feature/cargo-workspace
claude/github-issue-40-0au5ua
claude/linked-issue-acp-tui-u4vvpa
claude/minor-release-cut-vkugow
claude/sigit-acp-local-chat-cx6380
claude/sigit-cloud-agent-expansion-reox0a
claude/tool-permission-system
claude/zen-feynman-0u78dk
development
feature/acp-cancel-request
feature/acp-session-list-cursor
feature/acp-v1-conformance
feature/agent-tools-multiedit-glob-todos-remember
feature/cargo-workspace
feature/hooks
feature/load-local-model-explicitly
feature/sigit-agent-profile
feature/sigit-code-cloud
feature/xcode-mcp-server
fix/acp-replay-loaded-session-history
fix/acp-replay-reasoning-and-attachments
fix/dsml-doubled-bar-markers
main
release/v1.5.10
release/v1.5.6
release/v1.6.1
| 1 | //! What the editor shows when the endpoint refuses a turn. |
| 2 | //! |
| 3 | //! Every endpoint siGit talks to reports failures in the OpenAI error envelope, |
| 4 | //! and siGit Code Cloud writes those messages for the person reading them |
| 5 | //! ("Monthly siGit Code Cloud allowance reached…"). The agent used to hand the |
| 6 | //! raw status line and JSON body to the client, so Zed's error banner showed |
| 7 | //! the envelope rather than the sentence inside it. |
| 8 | //! |
| 9 | //! Two shapes to cover: a failing HTTP status, and an endpoint that only |
| 10 | //! discovers the problem after the status line is out and reports it as a |
| 11 | //! `data:` frame in the stream. |
| 12 | |
| 13 | use std::collections::VecDeque; |
| 14 | use std::io::{BufRead, BufReader, Read, Write}; |
| 15 | use std::net::TcpListener; |
| 16 | use std::process::{Child, ChildStdin, Command, Stdio}; |
| 17 | use std::sync::Mutex; |
| 18 | use std::sync::mpsc::{Receiver, channel}; |
| 19 | use std::time::{Duration, Instant}; |
| 20 | |
| 21 | use serde_json::{Value, json}; |
| 22 | |
| 23 | /// This is the first test file to launch the freshly built binary, and on the |
| 24 | /// Intel macOS runner that first launch alone has taken 25 seconds. |
| 25 | const TIMEOUT: Duration = Duration::from_secs(60); |
| 26 | |
| 27 | const ALLOWANCE_MESSAGE: &str = "Monthly siGit Code Cloud allowance reached. It resets at the start of your next billing period."; |
| 28 | |
| 29 | /// One scripted reply per request: an HTTP status plus a body. |
| 30 | struct Reply { |
| 31 | status: &'static str, |
| 32 | content_type: &'static str, |
| 33 | body: String, |
| 34 | } |
| 35 | |
| 36 | fn error_status_reply() -> Reply { |
| 37 | Reply { |
| 38 | status: "429 Too Many Requests", |
| 39 | content_type: "application/json", |
| 40 | body: json!({"error": {"message": ALLOWANCE_MESSAGE, "type": "server_error"}}).to_string(), |
| 41 | } |
| 42 | } |
| 43 | |
| 44 | /// A 200 that turns into an error partway through, the way an endpoint has to |
| 45 | /// report a failure it only sees once the response is already open. |
| 46 | fn error_in_stream_reply() -> Reply { |
| 47 | let frame = json!({"error": {"message": ALLOWANCE_MESSAGE, "type": "server_error"}}); |
| 48 | Reply { |
| 49 | status: "200 OK", |
| 50 | content_type: "text/event-stream", |
| 51 | body: format!("data: {frame}\n\n"), |
| 52 | } |
| 53 | } |
| 54 | |
| 55 | fn sse_tool_call(id: &str, name: &str, arguments: &str) -> Reply { |
| 56 | Reply { |
| 57 | status: "200 OK", |
| 58 | content_type: "text/event-stream", |
| 59 | body: format!( |
| 60 | "data: {}\n\ndata: [DONE]\n\n", |
| 61 | json!({ |
| 62 | "choices": [{"delta": {"tool_calls": [{ |
| 63 | "index": 0, |
| 64 | "id": id, |
| 65 | "function": {"name": name, "arguments": arguments}, |
| 66 | }]}}] |
| 67 | }) |
| 68 | ), |
| 69 | } |
| 70 | } |
| 71 | |
| 72 | fn upstream_500_reply() -> Reply { |
| 73 | Reply { |
| 74 | status: "500 Internal Server Error", |
| 75 | content_type: "application/json", |
| 76 | body: json!({ |
| 77 | "error": {"message": "Onde Cloud upstream error (500)", "type": "server_error"} |
| 78 | }) |
| 79 | .to_string(), |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | fn start_fake_endpoint(replies: Vec<Reply>) -> u16 { |
| 84 | let listener = TcpListener::bind("127.0.0.1:0").expect("bind fake endpoint"); |
| 85 | let port = listener.local_addr().unwrap().port(); |
| 86 | let queue = Mutex::new(VecDeque::from(replies)); |
| 87 | |
| 88 | std::thread::spawn(move || { |
| 89 | for stream in listener.incoming() { |
| 90 | let Ok(mut stream) = stream else { continue }; |
| 91 | let mut reader = BufReader::new(match stream.try_clone() { |
| 92 | Ok(clone) => clone, |
| 93 | Err(_) => continue, |
| 94 | }); |
| 95 | let mut content_length = 0usize; |
| 96 | loop { |
| 97 | let mut line = String::new(); |
| 98 | if reader.read_line(&mut line).unwrap_or(0) == 0 { |
| 99 | break; |
| 100 | } |
| 101 | let line = line.trim(); |
| 102 | if line.is_empty() { |
| 103 | break; |
| 104 | } |
| 105 | if let Some(length) = line.to_ascii_lowercase().strip_prefix("content-length:") { |
| 106 | content_length = length.trim().parse().unwrap_or(0); |
| 107 | } |
| 108 | } |
| 109 | let mut body = vec![0u8; content_length]; |
| 110 | if reader.read_exact(&mut body).is_err() { |
| 111 | continue; |
| 112 | } |
| 113 | let reply = queue.lock().unwrap().pop_front().unwrap_or_else(|| Reply { |
| 114 | status: "200 OK", |
| 115 | content_type: "text/event-stream", |
| 116 | body: "data: [DONE]\n\n".to_string(), |
| 117 | }); |
| 118 | let response = format!( |
| 119 | "HTTP/1.1 {}\r\ncontent-type: {}\r\ncontent-length: {}\r\n\ |
| 120 | connection: close\r\n\r\n{}", |
| 121 | reply.status, |
| 122 | reply.content_type, |
| 123 | reply.body.len(), |
| 124 | reply.body |
| 125 | ); |
| 126 | let _ = stream.write_all(response.as_bytes()); |
| 127 | let _ = stream.flush(); |
| 128 | } |
| 129 | }); |
| 130 | |
| 131 | port |
| 132 | } |
| 133 | |
| 134 | struct AgentUnderTest { |
| 135 | child: Child, |
| 136 | stdin: ChildStdin, |
| 137 | incoming: Receiver<Value>, |
| 138 | next_id: u64, |
| 139 | } |
| 140 | |
| 141 | fn spawn_agent(port: u16, config_dir: &std::path::Path) -> AgentUnderTest { |
| 142 | let mut child = Command::new(env!("CARGO_BIN_EXE_sigit")) |
| 143 | .env("OPENAI_BASE_URL", format!("http://127.0.0.1:{port}")) |
| 144 | .env("OPENAI_API_KEY", "test-key") |
| 145 | .env("SIGIT_MODEL", "scripted-model") |
| 146 | .env("SIGIT_CONFIG_DIR", config_dir) |
| 147 | .env("SIGIT_MCP", "off") |
| 148 | .env("SIGIT_PERMISSIONS", "allow") |
| 149 | .env_remove("SIGIT_LOCAL_INFERENCE") |
| 150 | .stdin(Stdio::piped()) |
| 151 | .stdout(Stdio::piped()) |
| 152 | .stderr(Stdio::null()) |
| 153 | .spawn() |
| 154 | .expect("spawn sigit in ACP mode"); |
| 155 | |
| 156 | let stdout = child.stdout.take().unwrap(); |
| 157 | let (message_tx, incoming) = channel(); |
| 158 | std::thread::spawn(move || { |
| 159 | for line in BufReader::new(stdout).lines() { |
| 160 | let Ok(line) = line else { break }; |
| 161 | if let Ok(message) = serde_json::from_str::<Value>(&line) |
| 162 | && message_tx.send(message).is_err() |
| 163 | { |
| 164 | break; |
| 165 | } |
| 166 | } |
| 167 | }); |
| 168 | |
| 169 | let stdin = child.stdin.take().unwrap(); |
| 170 | AgentUnderTest { |
| 171 | child, |
| 172 | stdin, |
| 173 | incoming, |
| 174 | next_id: 0, |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | impl AgentUnderTest { |
| 179 | fn request(&mut self, method: &str, params: Value) -> u64 { |
| 180 | self.next_id += 1; |
| 181 | let id = self.next_id; |
| 182 | let mut line = |
| 183 | json!({"jsonrpc": "2.0", "id": id, "method": method, "params": params}).to_string(); |
| 184 | line.push('\n'); |
| 185 | self.stdin.write_all(line.as_bytes()).expect("write stdin"); |
| 186 | self.stdin.flush().expect("flush stdin"); |
| 187 | id |
| 188 | } |
| 189 | |
| 190 | fn wait_for_message(&mut self, id: u64) -> Value { |
| 191 | let deadline = Instant::now() + TIMEOUT; |
| 192 | loop { |
| 193 | let remaining = deadline.saturating_duration_since(Instant::now()); |
| 194 | let Ok(message) = self.incoming.recv_timeout(remaining) else { |
| 195 | panic!("timed out waiting for the response to request {id}"); |
| 196 | }; |
| 197 | if message["id"] == id && message.get("method").is_none() { |
| 198 | return message; |
| 199 | } |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | fn wait_for_response(&mut self, id: u64) -> Value { |
| 204 | let message = self.wait_for_message(id); |
| 205 | assert!( |
| 206 | message.get("error").is_none(), |
| 207 | "request {id} failed: {message}" |
| 208 | ); |
| 209 | message |
| 210 | } |
| 211 | |
| 212 | fn wait_for_prompt_updates(&mut self, id: u64) -> (Value, Vec<Value>) { |
| 213 | let mut updates = Vec::new(); |
| 214 | let deadline = Instant::now() + TIMEOUT; |
| 215 | loop { |
| 216 | let remaining = deadline.saturating_duration_since(Instant::now()); |
| 217 | let Ok(message) = self.incoming.recv_timeout(remaining) else { |
| 218 | panic!("timed out waiting for the response to request {id}"); |
| 219 | }; |
| 220 | if message["method"] == "session/update" { |
| 221 | updates.push(message["params"]["update"].clone()); |
| 222 | } |
| 223 | if message["id"] == id && message.get("method").is_none() { |
| 224 | return (message, updates); |
| 225 | } |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | fn open_session(&mut self, cwd: &std::path::Path) -> String { |
| 230 | let id = self.request( |
| 231 | "initialize", |
| 232 | json!({"protocolVersion": 1, "clientCapabilities": {}}), |
| 233 | ); |
| 234 | self.wait_for_response(id); |
| 235 | |
| 236 | let id = self.request("session/new", json!({"cwd": cwd, "mcpServers": []})); |
| 237 | self.wait_for_response(id)["result"]["sessionId"] |
| 238 | .as_str() |
| 239 | .expect("session id") |
| 240 | .to_string() |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | impl Drop for AgentUnderTest { |
| 245 | fn drop(&mut self) { |
| 246 | let _ = self.child.kill(); |
| 247 | let _ = self.child.wait(); |
| 248 | } |
| 249 | } |
| 250 | |
| 251 | fn scratch(name: &str) -> std::path::PathBuf { |
| 252 | let dir = std::env::temp_dir().join(format!("sigit_acp_err_{name}_{}", std::process::id())); |
| 253 | std::fs::create_dir_all(dir.join("config")).unwrap(); |
| 254 | std::fs::create_dir_all(dir.join("work")).unwrap(); |
| 255 | dir |
| 256 | } |
| 257 | |
| 258 | /// The message the endpoint wrote is the message the client gets: no status |
| 259 | /// line in front of it, no JSON envelope around it. |
| 260 | fn assert_is_the_endpoints_message(error: &Value) { |
| 261 | let message = error["error"]["message"].as_str().unwrap_or_default(); |
| 262 | assert_eq!(message, ALLOWANCE_MESSAGE, "full error: {error}"); |
| 263 | } |
| 264 | |
| 265 | #[test] |
| 266 | fn a_refused_turn_reports_the_endpoints_own_message() { |
| 267 | let dir = scratch("status"); |
| 268 | let port = start_fake_endpoint(vec![error_status_reply()]); |
| 269 | let mut agent = spawn_agent(port, &dir.join("config")); |
| 270 | let session_id = agent.open_session(&dir.join("work")); |
| 271 | |
| 272 | let id = agent.request( |
| 273 | "session/prompt", |
| 274 | json!({"sessionId": session_id, "prompt": [{"type": "text", "text": "hello"}]}), |
| 275 | ); |
| 276 | let response = agent.wait_for_message(id); |
| 277 | |
| 278 | assert!( |
| 279 | response.get("error").is_some(), |
| 280 | "expected a failure: {response}" |
| 281 | ); |
| 282 | assert_is_the_endpoints_message(&response); |
| 283 | |
| 284 | std::fs::remove_dir_all(&dir).ok(); |
| 285 | } |
| 286 | |
| 287 | /// An error that arrives as a stream frame used to parse as a chunk with no |
| 288 | /// choices and get skipped, so the turn ended as if the model had said nothing. |
| 289 | #[test] |
| 290 | fn an_error_frame_mid_stream_is_not_swallowed() { |
| 291 | let dir = scratch("stream"); |
| 292 | let port = start_fake_endpoint(vec![error_in_stream_reply()]); |
| 293 | let mut agent = spawn_agent(port, &dir.join("config")); |
| 294 | let session_id = agent.open_session(&dir.join("work")); |
| 295 | |
| 296 | let id = agent.request( |
| 297 | "session/prompt", |
| 298 | json!({"sessionId": session_id, "prompt": [{"type": "text", "text": "hello"}]}), |
| 299 | ); |
| 300 | let response = agent.wait_for_message(id); |
| 301 | |
| 302 | assert!( |
| 303 | response.get("error").is_some(), |
| 304 | "expected a failure: {response}" |
| 305 | ); |
| 306 | assert_is_the_endpoints_message(&response); |
| 307 | |
| 308 | std::fs::remove_dir_all(&dir).ok(); |
| 309 | } |
| 310 | |
| 311 | /// A failing summarization round must not strand the session (issue #125): |
| 312 | /// compaction falls back to plain truncation, the pending tool round runs, |
| 313 | /// and the turn completes instead of ending with "start a new thread". |
| 314 | #[test] |
| 315 | fn failed_compaction_falls_back_to_truncation_and_the_turn_continues() { |
| 316 | let dir = scratch("compact"); |
| 317 | let port = start_fake_endpoint(vec![ |
| 318 | sse_tool_call( |
| 319 | "call_1", |
| 320 | "run_command", |
| 321 | r#"{"command":"echo truncation-fallback-works"}"#, |
| 322 | ), |
| 323 | upstream_500_reply(), |
| 324 | ]); |
| 325 | let mut agent = spawn_agent(port, &dir.join("config")); |
| 326 | let session_id = agent.open_session(&dir.join("work")); |
| 327 | |
| 328 | let oversized_prompt = "x".repeat(120_000); |
| 329 | let id = agent.request( |
| 330 | "session/prompt", |
| 331 | json!({"sessionId": session_id, "prompt": [{"type": "text", "text": oversized_prompt}]}), |
| 332 | ); |
| 333 | let (response, updates) = agent.wait_for_prompt_updates(id); |
| 334 | |
| 335 | assert!( |
| 336 | response.get("error").is_none(), |
| 337 | "truncation fallback keeps the turn alive: {response}" |
| 338 | ); |
| 339 | assert_eq!(response["result"]["stopReason"], "end_turn"); |
| 340 | |
| 341 | assert!( |
| 342 | updates |
| 343 | .iter() |
| 344 | .any(|update| update["sessionUpdate"] == "tool_call"), |
| 345 | "the pending tool call should run after the truncation fallback: {updates:?}" |
| 346 | ); |
| 347 | let rendered = updates |
| 348 | .iter() |
| 349 | .filter_map(|update| { |
| 350 | (update["sessionUpdate"] == "agent_message_chunk") |
| 351 | .then(|| update["content"]["text"].as_str()) |
| 352 | .flatten() |
| 353 | }) |
| 354 | .collect::<String>(); |
| 355 | assert!( |
| 356 | !rendered.contains("could not"), |
| 357 | "no compaction-failure banner should be shown when the fallback succeeded: {rendered:?}" |
| 358 | ); |
| 359 | |
| 360 | std::fs::remove_dir_all(&dir).ok(); |
| 361 | } |