| 1 | //! `MockLlmClient` — a queue-driven `LlmClient` implementation for tests. |
| 2 | //! |
| 3 | //! This client implements the [`LlmClient`](super::LlmClient) trait by replaying a |
| 4 | //! pre-loaded queue of canned responses (one per turn). It captures every |
| 5 | //! request the runtime sends so tests can assert on the outgoing payload — |
| 6 | //! e.g. confirming that prior `reasoning_content` is replayed in DeepSeek V4 |
| 7 | //! thinking-mode tool-calling turns (V4 §5.1.1; the bug that broke |
| 8 | //! v0.4.9-v0.5.1). |
| 9 | //! |
| 10 | //! # Mocking strategy |
| 11 | //! |
| 12 | //! Tests mock at the **trait boundary** (`LlmClient`), never at the `reqwest` |
| 13 | //! HTTP layer. The trait is the durable abstraction — internal HTTP plumbing |
| 14 | //! changes frequently and is not part of the public engine contract. |
| 15 | //! |
| 16 | //! # Example |
| 17 | //! |
| 18 | //! ```ignore |
| 19 | //! use crate::llm_client::mock::{MockLlmClient, canned}; |
| 20 | //! use crate::llm_client::LlmClient; |
| 21 | //! |
| 22 | //! // One canned turn that emits "hello world" as two text deltas, then |
| 23 | //! // finishes with stop_reason = "end_turn". |
| 24 | //! let turn = vec![ |
| 25 | //! canned::message_start("msg_1"), |
| 26 | //! canned::text_delta(0, "hello "), |
| 27 | //! canned::text_delta(0, "world"), |
| 28 | //! canned::message_stop(), |
| 29 | //! ]; |
| 30 | //! |
| 31 | //! let mock = MockLlmClient::new(vec![turn]); |
| 32 | //! let stream = mock.create_message_stream(/* ... */).await.unwrap(); |
| 33 | //! // ... drain the stream, assert deltas ... |
| 34 | //! assert_eq!(mock.call_count(), 1); |
| 35 | //! assert_eq!(mock.captured_requests().len(), 1); |
| 36 | //! ``` |
| 37 | |
| 38 | // This module ships methods + builder helpers that integration tests rely on |
| 39 | // individually. Not every helper is exercised by unit tests — that's expected |
| 40 | // (the goal is a usable mock surface for downstream tests), so we silence |
| 41 | // per-item dead-code warnings at the module level. |
| 42 | #![allow(dead_code)] |
| 43 | |
| 44 | use std::collections::VecDeque; |
| 45 | use std::pin::Pin; |
| 46 | use std::sync::Mutex; |
| 47 | use std::sync::atomic::{AtomicUsize, Ordering}; |
| 48 | |
| 49 | use anyhow::{Result, anyhow}; |
| 50 | use async_stream::try_stream; |
| 51 | use futures_util::Stream; |
| 52 | |
| 53 | use crate::models::{ |
| 54 | ContentBlock, MessageDelta, MessageRequest, MessageResponse, StreamEvent, Usage, |
| 55 | }; |
| 56 | |
| 57 | use super::{LlmClient, StreamEventBox}; |
| 58 | |
| 59 | /// A pre-recorded "turn" the mock will replay on the next streaming call. |
| 60 | /// |
| 61 | /// `MessageStop` does *not* need to be the final element — the mock will |
| 62 | /// auto-emit one if missing, mirroring the real client's behaviour. Likewise |
| 63 | /// the mock does not require `MessageStart` to be present. |
| 64 | pub type CannedTurn = Vec<StreamEvent>; |
| 65 | |
| 66 | /// A queued mock response step. |
| 67 | pub enum FauxStep { |
| 68 | Canned(CannedTurn), |
| 69 | /// Build a canned turn from the live outgoing request. |
| 70 | /// |
| 71 | /// Tests can assert DeepSeek V4's thinking-mode tool-call invariant here: |
| 72 | /// on the assistant turn that produced the previous tool call, the next |
| 73 | /// outgoing request must still carry `reasoning_content` (represented in |
| 74 | /// this model as a [`ContentBlock::Thinking`] block). If it is missing, |
| 75 | /// DeepSeek V4 returns HTTP 400 on the follow-up turn. This guards the |
| 76 | /// [v0.4.9-v0.5.1 regression range](https://github.com/Hmbown/CodeWhale/compare/v0.4.9...v0.5.1) |
| 77 | /// where that content was dropped. |
| 78 | Factory(Box<dyn Fn(&MessageRequest) -> CannedTurn + Send + Sync>), |
| 79 | } |
| 80 | |
| 81 | /// A queue-driven mock LLM client. |
| 82 | /// |
| 83 | /// The mock holds a FIFO queue of canned response turns. Each call to |
| 84 | /// [`LlmClient::create_message_stream`] dequeues the next turn and replays its |
| 85 | /// events as a stream. If the queue is exhausted, the call returns an error |
| 86 | /// — tests should ensure they push exactly as many turns as the runtime will |
| 87 | /// consume. |
| 88 | /// |
| 89 | /// The mock also captures the [`MessageRequest`] passed to every call so tests |
| 90 | /// can assert on the outgoing payload (e.g. that prior `reasoning_content` is |
| 91 | /// preserved across turns). |
| 92 | pub struct MockLlmClient { |
| 93 | canned: Mutex<VecDeque<FauxStep>>, |
| 94 | captured_requests: Mutex<Vec<MessageRequest>>, |
| 95 | calls: AtomicUsize, |
| 96 | provider_name: &'static str, |
| 97 | model: String, |
| 98 | /// If set, [`LlmClient::create_message`] returns this verbatim. Otherwise |
| 99 | /// it falls back to streaming + collection. Useful for non-streaming |
| 100 | /// compaction-style calls. |
| 101 | canned_messages: Mutex<VecDeque<MessageResponse>>, |
| 102 | } |
| 103 | |
| 104 | impl MockLlmClient { |
| 105 | /// Construct a mock that will replay the given canned turns in order. |
| 106 | #[must_use] |
| 107 | pub fn new(canned: Vec<CannedTurn>) -> Self { |
| 108 | Self { |
| 109 | canned: Mutex::new(canned.into_iter().map(FauxStep::Canned).collect()), |
| 110 | captured_requests: Mutex::new(Vec::new()), |
| 111 | calls: AtomicUsize::new(0), |
| 112 | provider_name: "mock", |
| 113 | model: "mock-model".to_string(), |
| 114 | canned_messages: Mutex::new(VecDeque::new()), |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | /// Set the provider-name string returned by [`LlmClient::provider_name`]. |
| 119 | #[must_use] |
| 120 | pub fn with_provider(mut self, name: &'static str) -> Self { |
| 121 | self.provider_name = name; |
| 122 | self |
| 123 | } |
| 124 | |
| 125 | /// Set the model identifier returned by [`LlmClient::model`]. |
| 126 | #[must_use] |
| 127 | pub fn with_model(mut self, model: impl Into<String>) -> Self { |
| 128 | self.model = model.into(); |
| 129 | self |
| 130 | } |
| 131 | |
| 132 | /// Push a canned turn onto the back of the queue. |
| 133 | pub fn push_turn(&self, turn: CannedTurn) { |
| 134 | self.canned |
| 135 | .lock() |
| 136 | .expect("MockLlmClient.canned mutex poisoned") |
| 137 | .push_back(FauxStep::Canned(turn)); |
| 138 | } |
| 139 | |
| 140 | /// Push a factory step onto the back of the queue. |
| 141 | /// |
| 142 | /// The closure receives the live outgoing [`MessageRequest`] before the |
| 143 | /// response stream is built, so assertions panic directly from the client |
| 144 | /// call rather than later while polling the returned stream. |
| 145 | pub fn push_factory<F>(&self, factory: F) |
| 146 | where |
| 147 | F: Fn(&MessageRequest) -> CannedTurn + Send + Sync + 'static, |
| 148 | { |
| 149 | self.canned |
| 150 | .lock() |
| 151 | .expect("MockLlmClient.canned mutex poisoned") |
| 152 | .push_back(FauxStep::Factory(Box::new(factory))); |
| 153 | } |
| 154 | |
| 155 | /// Push a canned non-streaming `MessageResponse`. Consumed by |
| 156 | /// [`LlmClient::create_message`] (FIFO). |
| 157 | pub fn push_message_response(&self, response: MessageResponse) { |
| 158 | self.canned_messages |
| 159 | .lock() |
| 160 | .expect("MockLlmClient.canned_messages mutex poisoned") |
| 161 | .push_back(response); |
| 162 | } |
| 163 | |
| 164 | /// Number of completed calls to either `create_message` or |
| 165 | /// `create_message_stream`. |
| 166 | #[must_use] |
| 167 | pub fn call_count(&self) -> usize { |
| 168 | self.calls.load(Ordering::SeqCst) |
| 169 | } |
| 170 | |
| 171 | /// Number of canned turns still queued. |
| 172 | #[must_use] |
| 173 | pub fn remaining_turns(&self) -> usize { |
| 174 | self.canned |
| 175 | .lock() |
| 176 | .expect("MockLlmClient.canned mutex poisoned") |
| 177 | .len() |
| 178 | } |
| 179 | |
| 180 | /// Snapshot of every request the mock has been asked to handle, in order. |
| 181 | #[must_use] |
| 182 | pub fn captured_requests(&self) -> Vec<MessageRequest> { |
| 183 | self.captured_requests |
| 184 | .lock() |
| 185 | .expect("MockLlmClient.captured_requests mutex poisoned") |
| 186 | .clone() |
| 187 | } |
| 188 | |
| 189 | /// Convenience: return the most recently captured request, or `None` if |
| 190 | /// the mock has not been called yet. |
| 191 | #[must_use] |
| 192 | pub fn last_request(&self) -> Option<MessageRequest> { |
| 193 | self.captured_requests |
| 194 | .lock() |
| 195 | .expect("MockLlmClient.captured_requests mutex poisoned") |
| 196 | .last() |
| 197 | .cloned() |
| 198 | } |
| 199 | |
| 200 | fn record_request(&self, request: &MessageRequest) { |
| 201 | self.captured_requests |
| 202 | .lock() |
| 203 | .expect("MockLlmClient.captured_requests mutex poisoned") |
| 204 | .push(request.clone()); |
| 205 | self.calls.fetch_add(1, Ordering::SeqCst); |
| 206 | } |
| 207 | |
| 208 | fn pop_step(&self) -> Option<FauxStep> { |
| 209 | self.canned |
| 210 | .lock() |
| 211 | .expect("MockLlmClient.canned mutex poisoned") |
| 212 | .pop_front() |
| 213 | } |
| 214 | |
| 215 | fn turn_from_step(&self, step: FauxStep, request: &MessageRequest) -> CannedTurn { |
| 216 | match step { |
| 217 | FauxStep::Canned(turn) => turn, |
| 218 | FauxStep::Factory(factory) => factory(request), |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | fn pop_message(&self) -> Option<MessageResponse> { |
| 223 | self.canned_messages |
| 224 | .lock() |
| 225 | .expect("MockLlmClient.canned_messages mutex poisoned") |
| 226 | .pop_front() |
| 227 | } |
| 228 | } |
| 229 | |
| 230 | impl LlmClient for MockLlmClient { |
| 231 | fn provider_name(&self) -> &'static str { |
| 232 | self.provider_name |
| 233 | } |
| 234 | |
| 235 | fn model(&self) -> &str { |
| 236 | &self.model |
| 237 | } |
| 238 | |
| 239 | async fn create_message(&self, request: MessageRequest) -> Result<MessageResponse> { |
| 240 | self.record_request(&request); |
| 241 | |
| 242 | if let Some(canned) = self.pop_message() { |
| 243 | return Ok(canned); |
| 244 | } |
| 245 | |
| 246 | // Fallback: synthesize a MessageResponse from the next streaming turn. |
| 247 | let Some(step) = self.pop_step() else { |
| 248 | return Err(anyhow!( |
| 249 | "MockLlmClient: create_message called but no canned response queued (request #{})", |
| 250 | self.calls.load(Ordering::SeqCst) |
| 251 | )); |
| 252 | }; |
| 253 | |
| 254 | let turn = self.turn_from_step(step, &request); |
| 255 | Ok(synthesize_message_response(turn, &self.model)) |
| 256 | } |
| 257 | |
| 258 | async fn create_message_stream(&self, request: MessageRequest) -> Result<StreamEventBox> { |
| 259 | self.record_request(&request); |
| 260 | |
| 261 | let Some(step) = self.pop_step() else { |
| 262 | return Err(anyhow!( |
| 263 | "MockLlmClient: create_message_stream called but no canned turn queued (call #{})", |
| 264 | self.calls.load(Ordering::SeqCst) |
| 265 | )); |
| 266 | }; |
| 267 | |
| 268 | let turn = self.turn_from_step(step, &request); |
| 269 | Ok(stream_from_canned(turn)) |
| 270 | } |
| 271 | |
| 272 | async fn health_check(&self) -> Result<bool> { |
| 273 | Ok(true) |
| 274 | } |
| 275 | } |
| 276 | |
| 277 | /// Wrap a canned event vector as a stream that yields each event in order and |
| 278 | /// auto-appends `MessageStop` if the trailing event is not already one. |
| 279 | fn stream_from_canned(turn: CannedTurn) -> StreamEventBox { |
| 280 | let s = try_stream! { |
| 281 | let has_stop = matches!(turn.last(), Some(StreamEvent::MessageStop)); |
| 282 | for ev in turn { |
| 283 | yield ev; |
| 284 | } |
| 285 | if !has_stop { |
| 286 | yield StreamEvent::MessageStop; |
| 287 | } |
| 288 | }; |
| 289 | Box::pin(s) as Pin<Box<dyn Stream<Item = Result<StreamEvent>> + Send + 'static>> |
| 290 | } |
| 291 | |
| 292 | /// Best-effort: collapse a streaming turn into a non-streaming |
| 293 | /// `MessageResponse` by concatenating text deltas. Used only as a fallback |
| 294 | /// when callers `create_message` without a queued `MessageResponse`. |
| 295 | fn synthesize_message_response(turn: CannedTurn, model: &str) -> MessageResponse { |
| 296 | use crate::models::Delta; |
| 297 | |
| 298 | let mut text = String::new(); |
| 299 | let mut stop_reason: Option<String> = None; |
| 300 | |
| 301 | for ev in turn { |
| 302 | match ev { |
| 303 | StreamEvent::ContentBlockDelta { |
| 304 | delta: Delta::TextDelta { text: t }, |
| 305 | .. |
| 306 | } => text.push_str(&t), |
| 307 | StreamEvent::MessageDelta { |
| 308 | delta: MessageDelta { |
| 309 | stop_reason: sr, .. |
| 310 | }, |
| 311 | .. |
| 312 | } => stop_reason = sr, |
| 313 | _ => {} |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | MessageResponse { |
| 318 | id: "mock_msg".to_string(), |
| 319 | r#type: "message".to_string(), |
| 320 | role: "assistant".to_string(), |
| 321 | content: vec![ContentBlock::Text { |
| 322 | text, |
| 323 | cache_control: None, |
| 324 | }], |
| 325 | model: model.to_string(), |
| 326 | stop_reason: stop_reason.or_else(|| Some("end_turn".to_string())), |
| 327 | stop_sequence: None, |
| 328 | container: None, |
| 329 | usage: Usage::default(), |
| 330 | } |
| 331 | } |
| 332 | |
| 333 | /// Builders for common canned-event patterns. Re-exported so tests can build |
| 334 | /// realistic streams without wiring `StreamEvent` shapes by hand. |
| 335 | pub mod canned { |
| 336 | use serde_json::Value; |
| 337 | |
| 338 | use crate::models::{ |
| 339 | ContentBlockStart, Delta, MessageDelta, MessageResponse, StreamEvent, Usage, |
| 340 | }; |
| 341 | |
| 342 | /// `MessageStart` event with a synthetic message envelope. |
| 343 | #[must_use] |
| 344 | pub fn message_start(id: &str) -> StreamEvent { |
| 345 | StreamEvent::MessageStart { |
| 346 | message: MessageResponse { |
| 347 | id: id.to_string(), |
| 348 | r#type: "message".to_string(), |
| 349 | role: "assistant".to_string(), |
| 350 | content: vec![], |
| 351 | model: "mock-model".to_string(), |
| 352 | stop_reason: None, |
| 353 | stop_sequence: None, |
| 354 | container: None, |
| 355 | usage: Usage::default(), |
| 356 | }, |
| 357 | } |
| 358 | } |
| 359 | |
| 360 | /// Open a text content block at `index`. |
| 361 | #[must_use] |
| 362 | pub fn text_block_start(index: u32) -> StreamEvent { |
| 363 | StreamEvent::ContentBlockStart { |
| 364 | index, |
| 365 | content_block: ContentBlockStart::Text { |
| 366 | text: String::new(), |
| 367 | }, |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | /// Append `text` to the content block at `index`. |
| 372 | #[must_use] |
| 373 | pub fn text_delta(index: u32, text: &str) -> StreamEvent { |
| 374 | StreamEvent::ContentBlockDelta { |
| 375 | index, |
| 376 | delta: Delta::TextDelta { |
| 377 | text: text.to_string(), |
| 378 | }, |
| 379 | } |
| 380 | } |
| 381 | |
| 382 | /// Append a thinking-content delta at `index`. |
| 383 | #[must_use] |
| 384 | pub fn thinking_delta(index: u32, thinking: &str) -> StreamEvent { |
| 385 | StreamEvent::ContentBlockDelta { |
| 386 | index, |
| 387 | delta: Delta::ThinkingDelta { |
| 388 | thinking: thinking.to_string(), |
| 389 | }, |
| 390 | } |
| 391 | } |
| 392 | |
| 393 | /// Open a tool_use content block at `index`. |
| 394 | #[must_use] |
| 395 | pub fn tool_use_block_start(index: u32, id: &str, name: &str) -> StreamEvent { |
| 396 | StreamEvent::ContentBlockStart { |
| 397 | index, |
| 398 | content_block: ContentBlockStart::ToolUse { |
| 399 | id: id.to_string(), |
| 400 | name: name.to_string(), |
| 401 | input: Value::Null, |
| 402 | caller: None, |
| 403 | }, |
| 404 | } |
| 405 | } |
| 406 | |
| 407 | /// Stream partial JSON for a tool's input arguments. |
| 408 | #[must_use] |
| 409 | pub fn tool_input_delta(index: u32, partial_json: &str) -> StreamEvent { |
| 410 | StreamEvent::ContentBlockDelta { |
| 411 | index, |
| 412 | delta: Delta::InputJsonDelta { |
| 413 | partial_json: partial_json.to_string(), |
| 414 | }, |
| 415 | } |
| 416 | } |
| 417 | |
| 418 | /// Close the content block at `index`. |
| 419 | #[must_use] |
| 420 | pub fn block_stop(index: u32) -> StreamEvent { |
| 421 | StreamEvent::ContentBlockStop { index } |
| 422 | } |
| 423 | |
| 424 | /// Emit a `message_delta` carrying `stop_reason` and optional `usage`. |
| 425 | #[must_use] |
| 426 | pub fn message_delta(stop_reason: &str, usage: Option<Usage>) -> StreamEvent { |
| 427 | StreamEvent::MessageDelta { |
| 428 | delta: MessageDelta { |
| 429 | stop_reason: Some(stop_reason.to_string()), |
| 430 | stop_sequence: None, |
| 431 | }, |
| 432 | usage, |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | /// Final `message_stop` sentinel. |
| 437 | #[must_use] |
| 438 | pub fn message_stop() -> StreamEvent { |
| 439 | StreamEvent::MessageStop |
| 440 | } |
| 441 | |
| 442 | /// Convenience: a complete "assistant emits this text" turn ending with |
| 443 | /// `stop_reason = "end_turn"`. |
| 444 | #[must_use] |
| 445 | pub fn simple_text_turn(text: &str) -> Vec<StreamEvent> { |
| 446 | vec![ |
| 447 | message_start("mock_msg_1"), |
| 448 | text_block_start(0), |
| 449 | text_delta(0, text), |
| 450 | block_stop(0), |
| 451 | message_delta("end_turn", None), |
| 452 | message_stop(), |
| 453 | ] |
| 454 | } |
| 455 | |
| 456 | /// Convenience: a turn that emits one assistant tool_call and stops. |
| 457 | #[must_use] |
| 458 | pub fn tool_call_turn(call_id: &str, tool_name: &str, args_json: &str) -> Vec<StreamEvent> { |
| 459 | vec![ |
| 460 | message_start("mock_msg_tool"), |
| 461 | tool_use_block_start(0, call_id, tool_name), |
| 462 | tool_input_delta(0, args_json), |
| 463 | block_stop(0), |
| 464 | message_delta("tool_use", None), |
| 465 | message_stop(), |
| 466 | ] |
| 467 | } |
| 468 | } |
| 469 | |
| 470 | // === Tests === |
| 471 | |
| 472 | #[cfg(test)] |
| 473 | mod tests { |
| 474 | use futures_util::StreamExt; |
| 475 | |
| 476 | use super::*; |
| 477 | use crate::llm_client::LlmClient; |
| 478 | use crate::models::{Delta, Message, MessageRequest, StreamEvent}; |
| 479 | |
| 480 | fn empty_request() -> MessageRequest { |
| 481 | MessageRequest { |
| 482 | model: "mock-model".to_string(), |
| 483 | messages: vec![Message { |
| 484 | role: "user".to_string(), |
| 485 | content: vec![], |
| 486 | }], |
| 487 | max_tokens: 1024, |
| 488 | system: None, |
| 489 | tools: None, |
| 490 | tool_choice: None, |
| 491 | metadata: None, |
| 492 | thinking: None, |
| 493 | reasoning_effort: None, |
| 494 | stream: Some(true), |
| 495 | temperature: None, |
| 496 | top_p: None, |
| 497 | } |
| 498 | } |
| 499 | |
| 500 | #[tokio::test] |
| 501 | async fn replays_canned_turn_via_stream() { |
| 502 | let mock = MockLlmClient::new(vec![canned::simple_text_turn("hello world")]); |
| 503 | |
| 504 | let mut stream = mock |
| 505 | .create_message_stream(empty_request()) |
| 506 | .await |
| 507 | .expect("stream should open"); |
| 508 | |
| 509 | let mut text = String::new(); |
| 510 | let mut saw_stop = false; |
| 511 | while let Some(ev) = stream.next().await { |
| 512 | match ev.expect("event") { |
| 513 | StreamEvent::ContentBlockDelta { |
| 514 | delta: Delta::TextDelta { text: t }, |
| 515 | .. |
| 516 | } => text.push_str(&t), |
| 517 | StreamEvent::MessageStop => { |
| 518 | saw_stop = true; |
| 519 | break; |
| 520 | } |
| 521 | _ => {} |
| 522 | } |
| 523 | } |
| 524 | |
| 525 | assert_eq!(text, "hello world"); |
| 526 | assert!(saw_stop); |
| 527 | assert_eq!(mock.call_count(), 1); |
| 528 | assert_eq!(mock.captured_requests().len(), 1); |
| 529 | assert_eq!(mock.remaining_turns(), 0); |
| 530 | } |
| 531 | |
| 532 | #[tokio::test] |
| 533 | async fn errors_when_queue_exhausted() { |
| 534 | let mock = MockLlmClient::new(Vec::new()); |
| 535 | let result = mock.create_message_stream(empty_request()).await; |
| 536 | match result { |
| 537 | Ok(_) => panic!("should error on empty queue"), |
| 538 | Err(err) => assert!(format!("{err}").contains("no canned")), |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | #[tokio::test] |
| 543 | async fn captures_request_payload_for_assertions() { |
| 544 | let mock = MockLlmClient::new(vec![canned::simple_text_turn("ok")]); |
| 545 | let mut req = empty_request(); |
| 546 | req.temperature = Some(0.42); |
| 547 | let _ = mock.create_message_stream(req).await.unwrap(); |
| 548 | |
| 549 | let captured = mock.last_request().expect("should have captured"); |
| 550 | assert_eq!(captured.temperature, Some(0.42)); |
| 551 | } |
| 552 | |
| 553 | #[tokio::test] |
| 554 | async fn stream_auto_appends_message_stop() { |
| 555 | // Queue a turn missing MessageStop — mock should append one. |
| 556 | let turn = vec![canned::text_block_start(0), canned::text_delta(0, "x")]; |
| 557 | let mock = MockLlmClient::new(vec![turn]); |
| 558 | |
| 559 | let mut stream = mock.create_message_stream(empty_request()).await.unwrap(); |
| 560 | let mut saw_stop = false; |
| 561 | while let Some(ev) = stream.next().await { |
| 562 | if matches!(ev.expect("event"), StreamEvent::MessageStop) { |
| 563 | saw_stop = true; |
| 564 | } |
| 565 | } |
| 566 | assert!(saw_stop, "auto MessageStop missing"); |
| 567 | } |
| 568 | |
| 569 | #[tokio::test] |
| 570 | async fn create_message_uses_canned_message_response_first() { |
| 571 | let mock = MockLlmClient::new(vec![canned::simple_text_turn("from stream")]); |
| 572 | mock.push_message_response(MessageResponse { |
| 573 | id: "preset".to_string(), |
| 574 | r#type: "message".to_string(), |
| 575 | role: "assistant".to_string(), |
| 576 | content: vec![ContentBlock::Text { |
| 577 | text: "from preset".to_string(), |
| 578 | cache_control: None, |
| 579 | }], |
| 580 | model: "mock-model".to_string(), |
| 581 | stop_reason: Some("end_turn".to_string()), |
| 582 | stop_sequence: None, |
| 583 | container: None, |
| 584 | usage: Usage::default(), |
| 585 | }); |
| 586 | |
| 587 | let resp = mock.create_message(empty_request()).await.unwrap(); |
| 588 | assert_eq!(resp.id, "preset"); |
| 589 | } |
| 590 | |
| 591 | #[tokio::test] |
| 592 | async fn create_message_synthesizes_from_streaming_turn_when_no_message_queued() { |
| 593 | let mock = MockLlmClient::new(vec![canned::simple_text_turn("synthesized")]); |
| 594 | let resp = mock.create_message(empty_request()).await.unwrap(); |
| 595 | let text = match &resp.content[0] { |
| 596 | ContentBlock::Text { text, .. } => text.clone(), |
| 597 | _ => panic!("expected text"), |
| 598 | }; |
| 599 | assert_eq!(text, "synthesized"); |
| 600 | assert_eq!(resp.stop_reason.as_deref(), Some("end_turn")); |
| 601 | } |
| 602 | |
| 603 | #[tokio::test] |
| 604 | async fn create_message_synthesizes_from_factory_turn() { |
| 605 | let mock = MockLlmClient::new(Vec::new()); |
| 606 | mock.push_factory(|request| { |
| 607 | assert_eq!(request.model, "mock-model"); |
| 608 | canned::simple_text_turn("from factory") |
| 609 | }); |
| 610 | |
| 611 | let resp = mock.create_message(empty_request()).await.unwrap(); |
| 612 | let text = match &resp.content[0] { |
| 613 | ContentBlock::Text { text, .. } => text.clone(), |
| 614 | _ => panic!("expected text"), |
| 615 | }; |
| 616 | assert_eq!(text, "from factory"); |
| 617 | } |
| 618 | |
| 619 | #[tokio::test] |
| 620 | async fn provider_and_model_are_overridable() { |
| 621 | let mock = MockLlmClient::new(vec![canned::simple_text_turn("x")]) |
| 622 | .with_provider("test-provider") |
| 623 | .with_model("test-model"); |
| 624 | assert_eq!(mock.provider_name(), "test-provider"); |
| 625 | assert_eq!(mock.model(), "test-model"); |
| 626 | } |
| 627 | |
| 628 | #[tokio::test] |
| 629 | async fn tool_call_turn_serializes_correctly() { |
| 630 | let mock = MockLlmClient::new(vec![canned::tool_call_turn( |
| 631 | "call_1", |
| 632 | "list_dir", |
| 633 | r#"{"path":"/tmp"}"#, |
| 634 | )]); |
| 635 | let mut stream = mock.create_message_stream(empty_request()).await.unwrap(); |
| 636 | |
| 637 | let mut saw_tool_use = false; |
| 638 | let mut json_seen = String::new(); |
| 639 | while let Some(ev) = stream.next().await { |
| 640 | match ev.unwrap() { |
| 641 | StreamEvent::ContentBlockStart { content_block, .. } => { |
| 642 | use crate::models::ContentBlockStart; |
| 643 | if let ContentBlockStart::ToolUse { name, .. } = content_block { |
| 644 | assert_eq!(name, "list_dir"); |
| 645 | saw_tool_use = true; |
| 646 | } |
| 647 | } |
| 648 | StreamEvent::ContentBlockDelta { |
| 649 | delta: Delta::InputJsonDelta { partial_json }, |
| 650 | .. |
| 651 | } => json_seen.push_str(&partial_json), |
| 652 | _ => {} |
| 653 | } |
| 654 | } |
| 655 | assert!(saw_tool_use, "expected tool_use start event"); |
| 656 | assert!(json_seen.contains("/tmp")); |
| 657 | } |
| 658 | |
| 659 | #[tokio::test] |
| 660 | async fn multiple_turns_consumed_in_order() { |
| 661 | let mock = MockLlmClient::new(vec![ |
| 662 | canned::simple_text_turn("turn-one"), |
| 663 | canned::simple_text_turn("turn-two"), |
| 664 | ]); |
| 665 | for expected in ["turn-one", "turn-two"] { |
| 666 | let mut stream = mock.create_message_stream(empty_request()).await.unwrap(); |
| 667 | let mut text = String::new(); |
| 668 | while let Some(ev) = stream.next().await { |
| 669 | if let StreamEvent::ContentBlockDelta { |
| 670 | delta: Delta::TextDelta { text: t }, |
| 671 | .. |
| 672 | } = ev.unwrap() |
| 673 | { |
| 674 | text.push_str(&t); |
| 675 | } |
| 676 | } |
| 677 | assert_eq!(text, expected); |
| 678 | } |
| 679 | assert_eq!(mock.call_count(), 2); |
| 680 | assert_eq!(mock.remaining_turns(), 0); |
| 681 | } |
| 682 | } |
| 683 |