1use std::{
2 error::Error,
3 fmt::{self, Debug},
4 path::Path,
5 sync::{Arc, Mutex},
6 time::Duration,
7};
8
9use crate::{
10 ToolMetrics,
11 assertions::{AssertionsReport, RanAssertion, RanAssertionResult},
12};
13use agent::{ContextLoadResult, Thread, ThreadEvent};
14use anyhow::{Result, anyhow};
15use assistant_settings::AgentProfileId;
16use async_trait::async_trait;
17use buffer_diff::DiffHunkStatus;
18use collections::HashMap;
19use futures::{FutureExt as _, StreamExt, channel::mpsc, select_biased};
20use gpui::{App, AppContext, AsyncApp, Entity};
21use language_model::{LanguageModel, Role, StopReason};
22
23pub const THREAD_EVENT_TIMEOUT: Duration = Duration::from_secs(60 * 2);
24
25#[async_trait(?Send)]
26pub trait Example {
27 fn meta(&self) -> ExampleMetadata;
28 async fn conversation(&self, cx: &mut ExampleContext) -> Result<()>;
29 fn diff_assertions(&self) -> Vec<JudgeAssertion> {
30 Vec::new()
31 }
32 fn thread_assertions(&self) -> Vec<JudgeAssertion> {
33 Vec::new()
34 }
35}
36
37#[derive(Clone, Debug)]
38pub struct JudgeAssertion {
39 pub id: String,
40 pub description: String,
41}
42
43#[derive(Clone, Debug)]
44pub struct ExampleMetadata {
45 pub name: String,
46 pub url: String,
47 pub revision: String,
48 pub language_server: Option<LanguageServer>,
49 pub max_assertions: Option<usize>,
50 pub profile_id: AgentProfileId,
51 pub existing_thread_json: Option<String>,
52}
53
54#[derive(Clone, Debug)]
55pub struct LanguageServer {
56 pub file_extension: String,
57 pub allow_preexisting_diagnostics: bool,
58}
59
60impl ExampleMetadata {
61 pub fn repo_name(&self) -> String {
62 self.url
63 .split('/')
64 .next_back()
65 .unwrap_or(&"")
66 .trim_end_matches(".git")
67 .into()
68 }
69}
70
71pub struct FailedAssertion(pub String);
72
73impl fmt::Debug for FailedAssertion {
74 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
75 write!(f, "Assertion failure: {}", self.0)
76 }
77}
78
79impl fmt::Display for FailedAssertion {
80 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
81 write!(f, "{}", self.0)
82 }
83}
84
85impl Error for FailedAssertion {}
86
87pub struct ExampleContext {
88 meta: ExampleMetadata,
89 log_prefix: String,
90 agent_thread: Entity<agent::Thread>,
91 app: AsyncApp,
92 model: Arc<dyn LanguageModel>,
93 pub assertions: AssertionsReport,
94 pub tool_metrics: Arc<Mutex<ToolMetrics>>,
95}
96
97impl ExampleContext {
98 pub fn new(
99 meta: ExampleMetadata,
100 log_prefix: String,
101 agent_thread: Entity<agent::Thread>,
102 model: Arc<dyn LanguageModel>,
103 app: AsyncApp,
104 ) -> Self {
105 let assertions = AssertionsReport::new(meta.max_assertions);
106
107 Self {
108 meta,
109 log_prefix,
110 agent_thread,
111 assertions,
112 model,
113 app,
114 tool_metrics: Arc::new(Mutex::new(ToolMetrics::default())),
115 }
116 }
117
118 pub fn push_user_message(&mut self, text: impl ToString) {
119 self.app
120 .update_entity(&self.agent_thread, |thread, cx| {
121 thread.insert_user_message(
122 text.to_string(),
123 ContextLoadResult::default(),
124 None,
125 Vec::new(),
126 cx,
127 );
128 })
129 .unwrap();
130 }
131
132 pub fn assert(&mut self, expected: bool, message: impl ToString) -> Result<()> {
133 let message = message.to_string();
134 self.log_assertion(
135 if expected {
136 Ok(())
137 } else {
138 Err(anyhow::Error::from(FailedAssertion(message.clone())))
139 },
140 message,
141 )
142 }
143
144 pub fn assert_some<T>(&mut self, option: Option<T>, message: impl ToString) -> Result<T> {
145 let message = message.to_string();
146 self.log_assertion(
147 match option {
148 Some(value) => Ok(value),
149 None => Err(anyhow::Error::from(FailedAssertion(message.clone()))),
150 },
151 message,
152 )
153 }
154
155 #[allow(dead_code)]
156 pub fn assert_eq<T: PartialEq + Debug>(
157 &mut self,
158 left: T,
159 right: T,
160 message: impl ToString,
161 ) -> Result<()> {
162 let message = message.to_string();
163 self.log_assertion(
164 if left == right {
165 Ok(())
166 } else {
167 println!(
168 "{}{}",
169 self.log_prefix,
170 pretty_assertions::Comparison::new(&left, &right)
171 );
172 Err(anyhow::Error::from(FailedAssertion(message.clone())))
173 },
174 message,
175 )
176 }
177
178 fn log_assertion<T>(&mut self, result: Result<T>, message: String) -> Result<T> {
179 if let Some(max) = self.meta.max_assertions {
180 anyhow::ensure!(
181 self.assertions.run_count() <= max,
182 "More assertions were run than the stated max_assertions of {max}"
183 );
184 }
185
186 self.assertions.ran.push(RanAssertion {
187 id: message.clone(),
188 result: Ok(RanAssertionResult {
189 analysis: None,
190 passed: result.is_ok(),
191 }),
192 });
193
194 if result.is_ok() {
195 println!("{}✅ {}", self.log_prefix, message);
196 } else {
197 println!("{}❌ {}", self.log_prefix, message);
198 }
199
200 result
201 }
202
203 pub async fn run_to_end(&mut self) -> Result<Response> {
204 self.run_turns(u32::MAX).await
205 }
206
207 pub async fn run_turn(&mut self) -> Result<Response> {
208 self.run_turns(1).await
209 }
210
211 pub async fn run_turns(&mut self, iterations: u32) -> Result<Response> {
212 let (mut tx, mut rx) = mpsc::channel(1);
213
214 let tool_metrics = self.tool_metrics.clone();
215 let log_prefix = self.log_prefix.clone();
216 let _subscription = self.app.subscribe(
217 &self.agent_thread,
218 move |thread, event: &ThreadEvent, cx| match event {
219 ThreadEvent::ShowError(thread_error) => {
220 tx.try_send(Err(anyhow!(thread_error.clone()))).ok();
221 }
222 ThreadEvent::Stopped(reason) => match reason {
223 Ok(StopReason::EndTurn) => {
224 tx.close_channel();
225 }
226 Ok(StopReason::ToolUse) => {
227 if thread.read(cx).remaining_turns() == 0 {
228 tx.close_channel();
229 }
230 }
231 Ok(StopReason::MaxTokens) => {
232 tx.try_send(Err(anyhow!("Exceeded maximum tokens"))).ok();
233 }
234 Err(err) => {
235 tx.try_send(Err(anyhow!(err.clone()))).ok();
236 }
237 },
238 ThreadEvent::NewRequest
239 | ThreadEvent::StreamedAssistantText(_, _)
240 | ThreadEvent::StreamedAssistantThinking(_, _)
241 | ThreadEvent::UsePendingTools { .. }
242 | ThreadEvent::CompletionCanceled => {}
243 ThreadEvent::ToolFinished {
244 tool_use_id,
245 pending_tool_use,
246 ..
247 } => {
248 thread.update(cx, |thread, _cx| {
249 if let Some(tool_use) = pending_tool_use {
250 let mut tool_metrics = tool_metrics.lock().unwrap();
251 if let Some(tool_result) = thread.tool_result(&tool_use_id) {
252 let message = if tool_result.is_error {
253 format!("✖︎ {}", tool_use.name)
254 } else {
255 format!("✔︎ {}", tool_use.name)
256 };
257 println!("{log_prefix}{message}");
258 tool_metrics
259 .insert(tool_result.tool_name.clone(), !tool_result.is_error);
260 } else {
261 let message =
262 format!("TOOL FINISHED WITHOUT RESULT: {}", tool_use.name);
263 println!("{log_prefix}{message}");
264 tool_metrics.insert(tool_use.name.clone(), true);
265 }
266 }
267 });
268 }
269 ThreadEvent::InvalidToolInput { .. } => {
270 println!("{log_prefix} invalid tool input");
271 }
272 ThreadEvent::MissingToolUse {
273 tool_use_id: _,
274 ui_text,
275 } => {
276 println!("{log_prefix} {ui_text}");
277 }
278 ThreadEvent::ToolConfirmationNeeded => {
279 panic!(
280 "{}Bug: Tool confirmation should not be required in eval",
281 log_prefix
282 );
283 }
284 ThreadEvent::StreamedCompletion
285 | ThreadEvent::MessageAdded(_)
286 | ThreadEvent::MessageEdited(_)
287 | ThreadEvent::MessageDeleted(_)
288 | ThreadEvent::SummaryChanged
289 | ThreadEvent::SummaryGenerated
290 | ThreadEvent::ReceivedTextChunk
291 | ThreadEvent::StreamedToolUse { .. }
292 | ThreadEvent::CheckpointChanged
293 | ThreadEvent::CancelEditing => {
294 tx.try_send(Ok(())).ok();
295 if std::env::var("ZED_EVAL_DEBUG").is_ok() {
296 println!("{}Event: {:#?}", log_prefix, event);
297 }
298 }
299 },
300 );
301
302 let model = self.model.clone();
303
304 let message_count_before = self.app.update_entity(&self.agent_thread, |thread, cx| {
305 thread.set_remaining_turns(iterations);
306 thread.send_to_model(model, None, cx);
307 thread.messages().len()
308 })?;
309
310 loop {
311 select_biased! {
312 result = rx.next() => {
313 if let Some(result) = result {
314 result?;
315 } else {
316 break;
317 }
318 }
319 _ = self.app.background_executor().timer(THREAD_EVENT_TIMEOUT).fuse() => {
320 anyhow::bail!("Agentic loop stalled - waited {THREAD_EVENT_TIMEOUT:?} without any events");
321 }
322 }
323 }
324
325 let messages = self.app.read_entity(&self.agent_thread, |thread, cx| {
326 let mut messages = Vec::new();
327 for message in thread.messages().skip(message_count_before) {
328 messages.push(Message {
329 _role: message.role,
330 text: message.to_string(),
331 tool_use: thread
332 .tool_uses_for_message(message.id, cx)
333 .into_iter()
334 .map(|tool_use| ToolUse {
335 name: tool_use.name.to_string(),
336 value: tool_use.input,
337 })
338 .collect(),
339 });
340 }
341 messages
342 })?;
343
344 let response = Response::new(messages);
345
346 Ok(response)
347 }
348
349 pub fn edits(&self) -> HashMap<Arc<Path>, FileEdits> {
350 self.agent_thread
351 .read_with(&self.app, |thread, cx| {
352 let action_log = thread.action_log().read(cx);
353 HashMap::from_iter(action_log.changed_buffers(cx).into_iter().map(
354 |(buffer, diff)| {
355 let snapshot = buffer.read(cx).snapshot();
356
357 let file = snapshot.file().unwrap();
358 let diff = diff.read(cx);
359 let base_text = diff.base_text().text();
360
361 let hunks = diff
362 .hunks(&snapshot, cx)
363 .map(|hunk| FileEditHunk {
364 base_text: base_text[hunk.diff_base_byte_range.clone()].to_string(),
365 text: snapshot
366 .text_for_range(hunk.range.clone())
367 .collect::<String>(),
368 status: hunk.status(),
369 })
370 .collect();
371
372 (file.path().clone(), FileEdits { hunks })
373 },
374 ))
375 })
376 .unwrap()
377 }
378
379 pub fn agent_thread(&self) -> Entity<Thread> {
380 self.agent_thread.clone()
381 }
382}
383
384impl AppContext for ExampleContext {
385 type Result<T> = anyhow::Result<T>;
386
387 fn new<T: 'static>(
388 &mut self,
389 build_entity: impl FnOnce(&mut gpui::Context<T>) -> T,
390 ) -> Self::Result<Entity<T>> {
391 self.app.new(build_entity)
392 }
393
394 fn reserve_entity<T: 'static>(&mut self) -> Self::Result<gpui::Reservation<T>> {
395 self.app.reserve_entity()
396 }
397
398 fn insert_entity<T: 'static>(
399 &mut self,
400 reservation: gpui::Reservation<T>,
401 build_entity: impl FnOnce(&mut gpui::Context<T>) -> T,
402 ) -> Self::Result<Entity<T>> {
403 self.app.insert_entity(reservation, build_entity)
404 }
405
406 fn update_entity<T, R>(
407 &mut self,
408 handle: &Entity<T>,
409 update: impl FnOnce(&mut T, &mut gpui::Context<T>) -> R,
410 ) -> Self::Result<R>
411 where
412 T: 'static,
413 {
414 self.app.update_entity(handle, update)
415 }
416
417 fn read_entity<T, R>(
418 &self,
419 handle: &Entity<T>,
420 read: impl FnOnce(&T, &App) -> R,
421 ) -> Self::Result<R>
422 where
423 T: 'static,
424 {
425 self.app.read_entity(handle, read)
426 }
427
428 fn update_window<T, F>(&mut self, window: gpui::AnyWindowHandle, f: F) -> Result<T>
429 where
430 F: FnOnce(gpui::AnyView, &mut gpui::Window, &mut App) -> T,
431 {
432 self.app.update_window(window, f)
433 }
434
435 fn read_window<T, R>(
436 &self,
437 window: &gpui::WindowHandle<T>,
438 read: impl FnOnce(Entity<T>, &App) -> R,
439 ) -> Result<R>
440 where
441 T: 'static,
442 {
443 self.app.read_window(window, read)
444 }
445
446 fn background_spawn<R>(
447 &self,
448 future: impl std::future::Future<Output = R> + Send + 'static,
449 ) -> gpui::Task<R>
450 where
451 R: Send + 'static,
452 {
453 self.app.background_spawn(future)
454 }
455
456 fn read_global<G, R>(&self, callback: impl FnOnce(&G, &App) -> R) -> Self::Result<R>
457 where
458 G: gpui::Global,
459 {
460 self.app.read_global(callback)
461 }
462}
463
464#[derive(Debug)]
465pub struct Response {
466 messages: Vec<Message>,
467}
468
469impl Response {
470 pub fn new(messages: Vec<Message>) -> Self {
471 Self { messages }
472 }
473
474 pub fn expect_tool(
475 &self,
476 tool_name: &'static str,
477 cx: &mut ExampleContext,
478 ) -> Result<&ToolUse> {
479 let result = self.find_tool_call(tool_name);
480 cx.assert_some(result, format!("called `{}`", tool_name))
481 }
482
483 pub fn find_tool_call(&self, tool_name: &str) -> Option<&ToolUse> {
484 self.messages.iter().rev().find_map(|msg| {
485 msg.tool_use
486 .iter()
487 .find(|tool_use| tool_use.name == tool_name)
488 })
489 }
490
491 #[allow(dead_code)]
492 pub fn tool_uses(&self) -> impl Iterator<Item = &ToolUse> {
493 self.messages.iter().flat_map(|msg| &msg.tool_use)
494 }
495
496 pub fn texts(&self) -> impl Iterator<Item = String> {
497 self.messages.iter().map(|message| message.text.clone())
498 }
499}
500
501#[derive(Debug)]
502pub struct Message {
503 _role: Role,
504 text: String,
505 tool_use: Vec<ToolUse>,
506}
507
508#[derive(Debug)]
509pub struct ToolUse {
510 pub name: String,
511 value: serde_json::Value,
512}
513
514impl ToolUse {
515 pub fn parse_input<Input>(&self) -> Result<Input>
516 where
517 Input: for<'de> serde::Deserialize<'de>,
518 {
519 serde_json::from_value::<Input>(self.value.clone()).map_err(|err| anyhow!(err))
520 }
521}
522
523#[derive(Debug, Eq, PartialEq)]
524pub struct FileEdits {
525 pub hunks: Vec<FileEditHunk>,
526}
527
528#[derive(Debug, Eq, PartialEq)]
529pub struct FileEditHunk {
530 pub base_text: String,
531 pub text: String,
532 pub status: DiffHunkStatus,
533}
534
535impl FileEdits {
536 pub fn has_added_line(&self, line: &str) -> bool {
537 self.hunks.iter().any(|hunk| {
538 hunk.status == DiffHunkStatus::added_none()
539 && hunk.base_text.is_empty()
540 && hunk.text.contains(line)
541 })
542 }
543}