| 1 | //! Offline evaluation harness for exercising representative tool loops. |
| 2 | //! |
| 3 | //! This module is intentionally self-contained so it can be wired into a CLI |
| 4 | //! command later without calling the network or any LLM endpoints. |
| 5 | |
| 6 | use anyhow::{Context, Result, anyhow}; |
| 7 | use ignore::WalkBuilder; |
| 8 | use regex::Regex; |
| 9 | use serde::{Deserialize, Serialize}; |
| 10 | use std::collections::BTreeMap; |
| 11 | use std::fs; |
| 12 | use std::io::Write; |
| 13 | use std::path::{Path, PathBuf}; |
| 14 | use std::time::{Duration, Instant}; |
| 15 | use tempfile::TempDir; |
| 16 | |
| 17 | #[cfg(test)] |
| 18 | #[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 19 | enum EvalShellPlatform { |
| 20 | Windows, |
| 21 | Unix, |
| 22 | } |
| 23 | |
| 24 | #[cfg(test)] |
| 25 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 26 | struct EvalShellInvocation { |
| 27 | program: &'static str, |
| 28 | args: Vec<String>, |
| 29 | raw_payload_on_windows: bool, |
| 30 | } |
| 31 | |
| 32 | #[cfg(test)] |
| 33 | fn eval_shell_invocation_for_platform( |
| 34 | command: &str, |
| 35 | platform: EvalShellPlatform, |
| 36 | ) -> EvalShellInvocation { |
| 37 | match platform { |
| 38 | EvalShellPlatform::Windows => EvalShellInvocation { |
| 39 | program: "cmd", |
| 40 | args: vec!["/C".to_string(), command.to_string()], |
| 41 | raw_payload_on_windows: true, |
| 42 | }, |
| 43 | EvalShellPlatform::Unix => EvalShellInvocation { |
| 44 | program: "sh", |
| 45 | args: vec!["-c".to_string(), command.to_string()], |
| 46 | raw_payload_on_windows: false, |
| 47 | }, |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | /// Representative tool steps covered by the evaluation harness. |
| 52 | #[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Serialize)] |
| 53 | pub enum ScenarioStepKind { |
| 54 | List, |
| 55 | Read, |
| 56 | Search, |
| 57 | Edit, |
| 58 | ApplyPatch, |
| 59 | Bash, |
| 60 | } |
| 61 | |
| 62 | impl ScenarioStepKind { |
| 63 | /// Canonical tool name associated with this step. |
| 64 | pub fn tool_name(self) -> &'static str { |
| 65 | match self { |
| 66 | ScenarioStepKind::List |
| 67 | | ScenarioStepKind::Read |
| 68 | | ScenarioStepKind::Search |
| 69 | | ScenarioStepKind::Edit => "File", |
| 70 | ScenarioStepKind::ApplyPatch => "apply_patch", |
| 71 | ScenarioStepKind::Bash => "Bash", |
| 72 | } |
| 73 | } |
| 74 | |
| 75 | /// Canonical action for action-based tools. |
| 76 | pub fn action(self) -> Option<&'static str> { |
| 77 | match self { |
| 78 | ScenarioStepKind::List => Some("list"), |
| 79 | ScenarioStepKind::Read => Some("read"), |
| 80 | ScenarioStepKind::Search => Some("search_content"), |
| 81 | ScenarioStepKind::Edit => Some("edit"), |
| 82 | ScenarioStepKind::ApplyPatch => None, |
| 83 | ScenarioStepKind::Bash => Some("run"), |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | /// Parse a step kind from CLI-friendly strings. |
| 88 | pub fn parse(value: &str) -> Option<Self> { |
| 89 | match value.trim().to_lowercase().as_str() { |
| 90 | "list" => Some(Self::List), |
| 91 | "read" => Some(Self::Read), |
| 92 | "search" | "grep" => Some(Self::Search), |
| 93 | "edit" => Some(Self::Edit), |
| 94 | "patch" | "apply_patch" => Some(Self::ApplyPatch), |
| 95 | "bash" | "shell" | "exec" => Some(Self::Bash), |
| 96 | _ => None, |
| 97 | } |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | /// Aggregate statistics for a single tool kind. |
| 102 | #[derive(Debug, Clone, Default, PartialEq, Eq, Serialize)] |
| 103 | pub struct ToolStats { |
| 104 | pub invocations: usize, |
| 105 | pub errors: usize, |
| 106 | pub total_duration: Duration, |
| 107 | } |
| 108 | |
| 109 | /// Top-level metrics produced by an evaluation run. |
| 110 | #[derive(Debug, Clone, PartialEq, Eq, Serialize)] |
| 111 | pub struct EvalMetrics { |
| 112 | pub success: bool, |
| 113 | pub tool_errors: usize, |
| 114 | pub steps: usize, |
| 115 | pub duration: Duration, |
| 116 | pub per_tool: BTreeMap<ScenarioStepKind, ToolStats>, |
| 117 | } |
| 118 | |
| 119 | /// One tool invocation recorded by the harness. |
| 120 | #[derive(Debug, Clone, PartialEq, Eq, Serialize)] |
| 121 | pub struct EvalStep { |
| 122 | pub kind: ScenarioStepKind, |
| 123 | pub tool_name: &'static str, |
| 124 | #[serde(skip_serializing_if = "Option::is_none")] |
| 125 | pub action: Option<&'static str>, |
| 126 | pub success: bool, |
| 127 | pub duration: Duration, |
| 128 | pub error: Option<String>, |
| 129 | pub output: Option<String>, |
| 130 | } |
| 131 | |
| 132 | /// Summary of the generated temporary workspace. |
| 133 | #[derive(Debug, Clone, PartialEq, Eq, Serialize)] |
| 134 | pub struct WorkspaceSummary { |
| 135 | pub root: PathBuf, |
| 136 | pub file_count: usize, |
| 137 | pub files: Vec<PathBuf>, |
| 138 | } |
| 139 | |
| 140 | /// Configuration for the offline evaluation harness. |
| 141 | #[derive(Debug, Clone)] |
| 142 | pub struct EvalHarnessConfig { |
| 143 | /// Human-readable scenario name for reporting. |
| 144 | pub scenario_name: String, |
| 145 | /// If set, the harness will intentionally fail this step to test metrics. |
| 146 | pub fail_step: Option<ScenarioStepKind>, |
| 147 | /// Shell command executed during the canonical `Bash.run` step. |
| 148 | pub shell_command: String, |
| 149 | /// Token that must appear in shell output for validation. |
| 150 | pub shell_expect_token: String, |
| 151 | /// Maximum characters stored for step output summaries. |
| 152 | pub max_output_chars: usize, |
| 153 | /// When set, every step is appended as a JSON Lines fixture to a file |
| 154 | /// inside this directory. The fixture file is named after the scenario |
| 155 | /// (e.g. `offline-tool-loop.jsonl`). Each line follows the schema: |
| 156 | /// `{ "request": <step descriptor>, "response_events": [<events>] }`. |
| 157 | /// The mock LLM client (`crate::llm_client::mock`) can replay these |
| 158 | /// fixtures for deterministic offline tests. See |
| 159 | /// `crates/tui/tests/README.md` for the full record/replay flow. |
| 160 | pub record_dir: Option<PathBuf>, |
| 161 | } |
| 162 | |
| 163 | impl Default for EvalHarnessConfig { |
| 164 | fn default() -> Self { |
| 165 | let shell_command = if cfg!(windows) { |
| 166 | "echo eval-harness".to_string() |
| 167 | } else { |
| 168 | "printf eval-harness".to_string() |
| 169 | }; |
| 170 | Self { |
| 171 | scenario_name: "offline-tool-loop".to_string(), |
| 172 | fail_step: None, |
| 173 | shell_command, |
| 174 | shell_expect_token: "eval-harness".to_string(), |
| 175 | max_output_chars: 240, |
| 176 | record_dir: None, |
| 177 | } |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | /// Offline harness that exercises representative tool loops in a temp workspace. |
| 182 | #[derive(Debug, Clone)] |
| 183 | pub struct EvalHarness { |
| 184 | config: EvalHarnessConfig, |
| 185 | } |
| 186 | |
| 187 | impl EvalHarness { |
| 188 | /// Create a new harness with the provided configuration. |
| 189 | pub fn new(config: EvalHarnessConfig) -> Self { |
| 190 | Self { config } |
| 191 | } |
| 192 | |
| 193 | /// Execute the offline evaluation scenario and return detailed results. |
| 194 | pub fn run(&self) -> Result<EvalRun> { |
| 195 | let started_at = Instant::now(); |
| 196 | let workspace = tempfile::Builder::new() |
| 197 | .prefix("deepseek-eval-") |
| 198 | .tempdir() |
| 199 | .context("failed to create evaluation workspace")?; |
| 200 | |
| 201 | let seed = seed_workspace(workspace.path())?; |
| 202 | |
| 203 | let mut steps = Vec::new(); |
| 204 | let mut per_tool: BTreeMap<ScenarioStepKind, ToolStats> = BTreeMap::new(); |
| 205 | |
| 206 | let list_output = self.run_step(ScenarioStepKind::List, &mut steps, &mut per_tool, || { |
| 207 | let entries = list_workspace(workspace.path())?; |
| 208 | Ok(entries.join(", ")) |
| 209 | }); |
| 210 | |
| 211 | let _read_output = self.run_step(ScenarioStepKind::Read, &mut steps, &mut per_tool, || { |
| 212 | let path = if self.config.fail_step == Some(ScenarioStepKind::Read) { |
| 213 | workspace.path().join("missing.txt") |
| 214 | } else { |
| 215 | seed.notes_path.clone() |
| 216 | }; |
| 217 | read_workspace_file(&path) |
| 218 | }); |
| 219 | |
| 220 | let search_output = |
| 221 | self.run_step(ScenarioStepKind::Search, &mut steps, &mut per_tool, || { |
| 222 | let root = if self.config.fail_step == Some(ScenarioStepKind::Search) { |
| 223 | workspace.path().join("missing-dir") |
| 224 | } else { |
| 225 | workspace.path().to_path_buf() |
| 226 | }; |
| 227 | let result = search_files(&root, "offline")?; |
| 228 | Ok(format!("matches={}", result.matches.len())) |
| 229 | }); |
| 230 | |
| 231 | let edit_output = self.run_step(ScenarioStepKind::Edit, &mut steps, &mut per_tool, || { |
| 232 | let path = if self.config.fail_step == Some(ScenarioStepKind::Edit) { |
| 233 | workspace.path().join("missing.txt") |
| 234 | } else { |
| 235 | seed.notes_path.clone() |
| 236 | }; |
| 237 | append_workspace_file(&path, "edited = true")?; |
| 238 | Ok("appended line".to_string()) |
| 239 | }); |
| 240 | |
| 241 | let patch_output = self.run_step( |
| 242 | ScenarioStepKind::ApplyPatch, |
| 243 | &mut steps, |
| 244 | &mut per_tool, |
| 245 | || { |
| 246 | let patch = if self.config.fail_step == Some(ScenarioStepKind::ApplyPatch) { |
| 247 | "*** Begin Patch\n*** Update File: notes.txt\n@@\n-THIS LINE DOES NOT EXIST\n+broken\n*** End Patch\n" |
| 248 | .to_string() |
| 249 | } else { |
| 250 | "*** Begin Patch\n*** Update File: notes.txt\n@@\n status = \"draft\"\n-todo: offline metrics\n+todo: offline metrics (patched)\n*** End Patch\n" |
| 251 | .to_string() |
| 252 | }; |
| 253 | apply_patch(workspace.path(), &patch)?; |
| 254 | Ok("patch applied".to_string()) |
| 255 | }, |
| 256 | ); |
| 257 | |
| 258 | let shell_output = self.run_step(ScenarioStepKind::Bash, &mut steps, &mut per_tool, || { |
| 259 | let command = if self.config.fail_step == Some(ScenarioStepKind::Bash) { |
| 260 | "command_that_does_not_exist".to_string() |
| 261 | } else { |
| 262 | self.config.shell_command.clone() |
| 263 | }; |
| 264 | run_bash(workspace.path(), &command) |
| 265 | }); |
| 266 | |
| 267 | let duration = started_at.elapsed(); |
| 268 | |
| 269 | let workspace_summary = summarize_workspace(workspace.path(), list_output.as_deref())?; |
| 270 | |
| 271 | let validation_success = validate_outputs( |
| 272 | workspace.path(), |
| 273 | &self.config.shell_expect_token, |
| 274 | search_output.as_deref(), |
| 275 | edit_output.as_deref(), |
| 276 | patch_output.as_deref(), |
| 277 | shell_output.as_deref(), |
| 278 | ); |
| 279 | |
| 280 | let tool_errors = steps.iter().filter(|s| !s.success).count(); |
| 281 | let success = tool_errors == 0 && validation_success; |
| 282 | |
| 283 | let metrics = EvalMetrics { |
| 284 | success, |
| 285 | tool_errors, |
| 286 | steps: steps.len(), |
| 287 | duration, |
| 288 | per_tool, |
| 289 | }; |
| 290 | |
| 291 | Ok(EvalRun { |
| 292 | scenario_name: self.config.scenario_name.clone(), |
| 293 | workspace, |
| 294 | workspace_summary, |
| 295 | metrics, |
| 296 | steps, |
| 297 | }) |
| 298 | } |
| 299 | |
| 300 | fn run_step<T, F>( |
| 301 | &self, |
| 302 | kind: ScenarioStepKind, |
| 303 | steps: &mut Vec<EvalStep>, |
| 304 | per_tool: &mut BTreeMap<ScenarioStepKind, ToolStats>, |
| 305 | f: F, |
| 306 | ) -> Option<T> |
| 307 | where |
| 308 | F: FnOnce() -> Result<T>, |
| 309 | T: ToString, |
| 310 | { |
| 311 | let started_at = Instant::now(); |
| 312 | let result = f(); |
| 313 | let duration = started_at.elapsed(); |
| 314 | |
| 315 | let stats = per_tool.entry(kind).or_default(); |
| 316 | stats.invocations += 1; |
| 317 | stats.total_duration += duration; |
| 318 | |
| 319 | match result { |
| 320 | Ok(value) => { |
| 321 | let output = truncate_output(&value.to_string(), self.config.max_output_chars); |
| 322 | steps.push(EvalStep { |
| 323 | kind, |
| 324 | tool_name: kind.tool_name(), |
| 325 | action: kind.action(), |
| 326 | success: true, |
| 327 | duration, |
| 328 | error: None, |
| 329 | output: Some(output.clone()), |
| 330 | }); |
| 331 | if let Some(dir) = self.config.record_dir.as_deref() { |
| 332 | let _ = record_fixture( |
| 333 | dir, |
| 334 | &self.config.scenario_name, |
| 335 | FixtureRecord::ok(kind, &output), |
| 336 | ); |
| 337 | } |
| 338 | Some(value) |
| 339 | } |
| 340 | Err(err) => { |
| 341 | stats.errors += 1; |
| 342 | let err_str = err.to_string(); |
| 343 | steps.push(EvalStep { |
| 344 | kind, |
| 345 | tool_name: kind.tool_name(), |
| 346 | action: kind.action(), |
| 347 | success: false, |
| 348 | duration, |
| 349 | error: Some(err_str.clone()), |
| 350 | output: None, |
| 351 | }); |
| 352 | if let Some(dir) = self.config.record_dir.as_deref() { |
| 353 | let _ = record_fixture( |
| 354 | dir, |
| 355 | &self.config.scenario_name, |
| 356 | FixtureRecord::err(kind, &err_str), |
| 357 | ); |
| 358 | } |
| 359 | None |
| 360 | } |
| 361 | } |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | // === Fixture record/replay format =========================================== |
| 366 | // |
| 367 | // The `--record` flag writes one JSON object per line to a `.jsonl` file: |
| 368 | // |
| 369 | // { "request": { "tool": "File", "action": "list", "kind": "List" }, |
| 370 | // "response_events": [{ "type": "ok", "output": "…" }] } |
| 371 | // |
| 372 | // The mock LLM client replays these fixtures via |
| 373 | // `MockLlmClient::push_message_response` (or the streaming variant) by mapping |
| 374 | // each `response_events` array onto a canned `Vec<StreamEvent>`. |
| 375 | // |
| 376 | // This format is intentionally minimal — additional fields (timing, model, |
| 377 | // usage) can be added without breaking older fixtures because each line is a |
| 378 | // self-contained JSON object. |
| 379 | |
| 380 | /// Schema for one line of a `--record` JSONL fixture file. |
| 381 | #[derive(Debug, Clone, Serialize, Deserialize)] |
| 382 | pub struct FixtureRecord { |
| 383 | /// Step descriptor (`{ step, kind }`). |
| 384 | pub request: serde_json::Value, |
| 385 | /// One or more synthetic response events. |
| 386 | pub response_events: Vec<serde_json::Value>, |
| 387 | } |
| 388 | |
| 389 | impl FixtureRecord { |
| 390 | fn ok(kind: ScenarioStepKind, output: &str) -> Self { |
| 391 | Self { |
| 392 | request: Self::request(kind), |
| 393 | response_events: vec![serde_json::json!({ |
| 394 | "type": "ok", |
| 395 | "output": output, |
| 396 | })], |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | fn err(kind: ScenarioStepKind, error: &str) -> Self { |
| 401 | Self { |
| 402 | request: Self::request(kind), |
| 403 | response_events: vec![serde_json::json!({ |
| 404 | "type": "error", |
| 405 | "error": error, |
| 406 | })], |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | fn request(kind: ScenarioStepKind) -> serde_json::Value { |
| 411 | let mut request = serde_json::json!({ |
| 412 | "tool": kind.tool_name(), |
| 413 | "kind": format!("{kind:?}"), |
| 414 | }); |
| 415 | if let Some(action) = kind.action() { |
| 416 | request["action"] = serde_json::Value::String(action.to_string()); |
| 417 | } |
| 418 | request |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | /// Append one fixture record to `<dir>/<scenario>.jsonl` (creating dir + file |
| 423 | /// if missing). Best-effort: I/O errors are returned but generally ignored by |
| 424 | /// the harness so a recording failure does not mask the run's primary result. |
| 425 | pub fn record_fixture(dir: &Path, scenario_name: &str, record: FixtureRecord) -> Result<PathBuf> { |
| 426 | fs::create_dir_all(dir) |
| 427 | .with_context(|| format!("failed to create fixture dir: {}", dir.display()))?; |
| 428 | let safe_scenario = scenario_name |
| 429 | .chars() |
| 430 | .map(|c| { |
| 431 | if c.is_ascii_alphanumeric() || c == '-' || c == '_' { |
| 432 | c |
| 433 | } else { |
| 434 | '_' |
| 435 | } |
| 436 | }) |
| 437 | .collect::<String>(); |
| 438 | let path = dir.join(format!("{safe_scenario}.jsonl")); |
| 439 | let line = serde_json::to_string(&record).context("failed to serialize fixture record")?; |
| 440 | |
| 441 | let mut file = fs::OpenOptions::new() |
| 442 | .create(true) |
| 443 | .append(true) |
| 444 | .open(&path) |
| 445 | .with_context(|| format!("failed to open fixture file: {}", path.display()))?; |
| 446 | writeln!(file, "{line}") |
| 447 | .with_context(|| format!("failed to write fixture line to {}", path.display()))?; |
| 448 | Ok(path) |
| 449 | } |
| 450 | |
| 451 | impl Default for EvalHarness { |
| 452 | fn default() -> Self { |
| 453 | Self::new(EvalHarnessConfig::default()) |
| 454 | } |
| 455 | } |
| 456 | |
| 457 | /// Result of running the evaluation harness. |
| 458 | #[derive(Debug)] |
| 459 | pub struct EvalRun { |
| 460 | pub scenario_name: String, |
| 461 | workspace: TempDir, |
| 462 | pub workspace_summary: WorkspaceSummary, |
| 463 | pub metrics: EvalMetrics, |
| 464 | pub steps: Vec<EvalStep>, |
| 465 | } |
| 466 | |
| 467 | impl EvalRun { |
| 468 | /// Get the root of the temporary workspace. |
| 469 | pub fn workspace_root(&self) -> &Path { |
| 470 | self.workspace.path() |
| 471 | } |
| 472 | |
| 473 | /// Convert the run into a serializable report for CLI output. |
| 474 | pub fn to_report(&self) -> EvalReport { |
| 475 | EvalReport { |
| 476 | scenario_name: self.scenario_name.clone(), |
| 477 | workspace_root: self.workspace_root().to_path_buf(), |
| 478 | workspace_summary: self.workspace_summary.clone(), |
| 479 | metrics: self.metrics.clone(), |
| 480 | steps: self.steps.clone(), |
| 481 | } |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | /// Serializable report derived from an `EvalRun`. |
| 486 | #[derive(Debug, Clone, Serialize, PartialEq, Eq)] |
| 487 | pub struct EvalReport { |
| 488 | pub scenario_name: String, |
| 489 | pub workspace_root: PathBuf, |
| 490 | pub workspace_summary: WorkspaceSummary, |
| 491 | pub metrics: EvalMetrics, |
| 492 | pub steps: Vec<EvalStep>, |
| 493 | } |
| 494 | |
| 495 | #[derive(Debug, Clone)] |
| 496 | struct SeedWorkspace { |
| 497 | notes_path: PathBuf, |
| 498 | } |
| 499 | |
| 500 | fn seed_workspace(root: &Path) -> Result<SeedWorkspace> { |
| 501 | let src_dir = root.join("src"); |
| 502 | fs::create_dir_all(&src_dir) |
| 503 | .with_context(|| format!("failed to create seed directory: {}", src_dir.display()))?; |
| 504 | |
| 505 | let readme_path = root.join("README.md"); |
| 506 | fs::write( |
| 507 | &readme_path, |
| 508 | "# Eval Harness Workspace\n\nThis workspace is offline.\n", |
| 509 | ) |
| 510 | .with_context(|| format!("failed to write {}", readme_path.display()))?; |
| 511 | |
| 512 | let notes_path = root.join("notes.txt"); |
| 513 | fs::write( |
| 514 | ¬es_path, |
| 515 | "# Eval Harness\nstatus = \"draft\"\ntodo: offline metrics\n", |
| 516 | ) |
| 517 | .with_context(|| format!("failed to write {}", notes_path.display()))?; |
| 518 | |
| 519 | let lib_path = src_dir.join("lib.rs"); |
| 520 | fs::write( |
| 521 | &lib_path, |
| 522 | "pub fn add(a: i32, b: i32) -> i32 {\n a + b\n}\n", |
| 523 | ) |
| 524 | .with_context(|| format!("failed to write {}", lib_path.display()))?; |
| 525 | |
| 526 | Ok(SeedWorkspace { notes_path }) |
| 527 | } |
| 528 | |
| 529 | fn summarize_workspace(root: &Path, list_output: Option<&str>) -> Result<WorkspaceSummary> { |
| 530 | let mut files = Vec::new(); |
| 531 | |
| 532 | let walker = WalkBuilder::new(root) |
| 533 | .hidden(false) |
| 534 | .git_ignore(false) |
| 535 | .git_global(false) |
| 536 | .git_exclude(false) |
| 537 | .build(); |
| 538 | |
| 539 | for entry in walker { |
| 540 | let entry = entry.with_context(|| format!("failed to walk {}", root.display()))?; |
| 541 | if entry.file_type().is_some_and(|t| t.is_file()) { |
| 542 | files.push(entry.into_path()); |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | if files.is_empty() |
| 547 | && let Some(output) = list_output |
| 548 | && !output.trim().is_empty() |
| 549 | { |
| 550 | return Err(anyhow!( |
| 551 | "workspace appears empty after File.list: {}", |
| 552 | output.trim() |
| 553 | )); |
| 554 | } |
| 555 | |
| 556 | files.sort(); |
| 557 | |
| 558 | Ok(WorkspaceSummary { |
| 559 | root: root.to_path_buf(), |
| 560 | file_count: files.len(), |
| 561 | files, |
| 562 | }) |
| 563 | } |
| 564 | |
| 565 | fn validate_outputs( |
| 566 | root: &Path, |
| 567 | shell_expect_token: &str, |
| 568 | search_output: Option<&str>, |
| 569 | edit_output: Option<&str>, |
| 570 | patch_output: Option<&str>, |
| 571 | shell_output: Option<&str>, |
| 572 | ) -> bool { |
| 573 | let notes_path = root.join("notes.txt"); |
| 574 | let notes = match fs::read_to_string(¬es_path) { |
| 575 | Ok(content) => content, |
| 576 | Err(_) => return false, |
| 577 | }; |
| 578 | |
| 579 | let search_ok = search_output.is_some_and(|s| s.contains("matches=")); |
| 580 | let edit_ok = edit_output.is_some_and(|s| !s.is_empty()) && notes.contains("edited = true"); |
| 581 | let patch_ok = patch_output.is_some_and(|s| !s.is_empty()) |
| 582 | && notes.contains("todo: offline metrics (patched)"); |
| 583 | let shell_ok = shell_output |
| 584 | .map(str::trim) |
| 585 | .is_some_and(|s| s.contains(shell_expect_token)); |
| 586 | |
| 587 | search_ok && edit_ok && patch_ok && shell_ok |
| 588 | } |
| 589 | |
| 590 | fn list_workspace(path: &Path) -> Result<Vec<String>> { |
| 591 | let mut entries = Vec::new(); |
| 592 | let dir = fs::read_dir(path) |
| 593 | .with_context(|| format!("failed to read directory: {}", path.display()))?; |
| 594 | |
| 595 | for entry in dir { |
| 596 | let entry = entry.with_context(|| format!("failed to list {}", path.display()))?; |
| 597 | entries.push(entry.file_name().to_string_lossy().to_string()); |
| 598 | } |
| 599 | |
| 600 | entries.sort(); |
| 601 | Ok(entries) |
| 602 | } |
| 603 | |
| 604 | fn read_workspace_file(path: &Path) -> Result<String> { |
| 605 | fs::read_to_string(path).with_context(|| format!("failed to read {}", path.display())) |
| 606 | } |
| 607 | |
| 608 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 609 | struct SearchMatch { |
| 610 | path: PathBuf, |
| 611 | line: usize, |
| 612 | content: String, |
| 613 | } |
| 614 | |
| 615 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 616 | struct SearchResult { |
| 617 | matches: Vec<SearchMatch>, |
| 618 | } |
| 619 | |
| 620 | fn search_files(root: &Path, pattern: &str) -> Result<SearchResult> { |
| 621 | if !root.exists() { |
| 622 | return Err(anyhow!("search root does not exist: {}", root.display())); |
| 623 | } |
| 624 | |
| 625 | let regex = Regex::new(pattern).context("failed to compile search regex")?; |
| 626 | let mut matches = Vec::new(); |
| 627 | |
| 628 | let walker = WalkBuilder::new(root) |
| 629 | .hidden(false) |
| 630 | .git_ignore(false) |
| 631 | .git_global(false) |
| 632 | .git_exclude(false) |
| 633 | .build(); |
| 634 | |
| 635 | for entry in walker { |
| 636 | let entry = entry.with_context(|| format!("failed to walk {}", root.display()))?; |
| 637 | if !entry.file_type().is_some_and(|t| t.is_file()) { |
| 638 | continue; |
| 639 | } |
| 640 | |
| 641 | let path = entry.path(); |
| 642 | let content = match fs::read_to_string(path) { |
| 643 | Ok(c) => c, |
| 644 | Err(_) => continue, |
| 645 | }; |
| 646 | |
| 647 | for (idx, line) in content.lines().enumerate() { |
| 648 | if regex.is_match(line) { |
| 649 | matches.push(SearchMatch { |
| 650 | path: path.to_path_buf(), |
| 651 | line: idx + 1, |
| 652 | content: line.to_string(), |
| 653 | }); |
| 654 | } |
| 655 | if matches.len() >= 64 { |
| 656 | break; |
| 657 | } |
| 658 | } |
| 659 | if matches.len() >= 64 { |
| 660 | break; |
| 661 | } |
| 662 | } |
| 663 | |
| 664 | Ok(SearchResult { matches }) |
| 665 | } |
| 666 | |
| 667 | fn append_workspace_file(path: &Path, line: &str) -> Result<()> { |
| 668 | let mut content = read_workspace_file(path)?; |
| 669 | if !content.ends_with('\n') { |
| 670 | content.push('\n'); |
| 671 | } |
| 672 | content.push_str(line); |
| 673 | content.push('\n'); |
| 674 | fs::write(path, content).with_context(|| format!("failed to write {}", path.display())) |
| 675 | } |
| 676 | |
| 677 | fn apply_patch(root: &Path, patch: &str) -> Result<()> { |
| 678 | let mut lines = patch.lines(); |
| 679 | |
| 680 | let begin = lines.next().unwrap_or_default(); |
| 681 | if begin != "*** Begin Patch" { |
| 682 | return Err(anyhow!("patch missing *** Begin Patch header")); |
| 683 | } |
| 684 | |
| 685 | let header = lines.next().unwrap_or_default(); |
| 686 | let file_rel = header |
| 687 | .strip_prefix("*** Update File: ") |
| 688 | .ok_or_else(|| anyhow!("only *** Update File patches are supported"))?; |
| 689 | if file_rel.contains("..") { |
| 690 | return Err(anyhow!("patch path must be workspace-relative")); |
| 691 | } |
| 692 | |
| 693 | let file_path = root.join(file_rel); |
| 694 | let original = read_workspace_file(&file_path)?; |
| 695 | let had_trailing_newline = original.ends_with('\n'); |
| 696 | let mut file_lines: Vec<String> = original.lines().map(|l| l.to_string()).collect(); |
| 697 | |
| 698 | let mut cursor = 0usize; |
| 699 | for raw_line in lines { |
| 700 | if raw_line == "*** End Patch" { |
| 701 | break; |
| 702 | } |
| 703 | if raw_line.starts_with("*** ") { |
| 704 | return Err(anyhow!("unexpected patch directive: {raw_line}")); |
| 705 | } |
| 706 | if raw_line.starts_with("@@") { |
| 707 | continue; |
| 708 | } |
| 709 | |
| 710 | let (kind, rest) = raw_line.split_at(1); |
| 711 | let content = rest.to_string(); |
| 712 | |
| 713 | match kind { |
| 714 | " " => { |
| 715 | let Some(found) = file_lines[cursor..] |
| 716 | .iter() |
| 717 | .position(|line| line == &content) |
| 718 | .map(|offset| cursor + offset) |
| 719 | else { |
| 720 | return Err(anyhow!( |
| 721 | "patch context not found in {}: {}", |
| 722 | file_path.display(), |
| 723 | content |
| 724 | )); |
| 725 | }; |
| 726 | cursor = found + 1; |
| 727 | } |
| 728 | "-" => { |
| 729 | if cursor >= file_lines.len() || file_lines[cursor] != content { |
| 730 | return Err(anyhow!( |
| 731 | "patch removal mismatch in {}: expected '{}'", |
| 732 | file_path.display(), |
| 733 | content |
| 734 | )); |
| 735 | } |
| 736 | file_lines.remove(cursor); |
| 737 | } |
| 738 | "+" => { |
| 739 | file_lines.insert(cursor, content); |
| 740 | cursor += 1; |
| 741 | } |
| 742 | _ => return Err(anyhow!("unsupported patch line: {raw_line}")), |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | let mut updated = file_lines.join("\n"); |
| 747 | if had_trailing_newline { |
| 748 | updated.push('\n'); |
| 749 | } |
| 750 | |
| 751 | fs::write(&file_path, updated) |
| 752 | .with_context(|| format!("failed to write patched file {}", file_path.display())) |
| 753 | } |
| 754 | |
| 755 | fn run_bash(root: &Path, command: &str) -> Result<String> { |
| 756 | crate::shell_dispatcher::global_dispatcher().run_foreground(command, root) |
| 757 | } |
| 758 | |
| 759 | fn truncate_output(value: &str, max_chars: usize) -> String { |
| 760 | if value.chars().count() <= max_chars { |
| 761 | return value.to_string(); |
| 762 | } |
| 763 | |
| 764 | let truncated: String = value.chars().take(max_chars).collect(); |
| 765 | format!("{truncated}...") |
| 766 | } |
| 767 | |
| 768 | #[cfg(test)] |
| 769 | mod tests { |
| 770 | use super::*; |
| 771 | |
| 772 | #[test] |
| 773 | fn eval_shell_invocation_preserves_quoted_payload_as_single_arg() { |
| 774 | let command = r#"git commit -m "feat: complete sub-pages""#; |
| 775 | |
| 776 | let windows = eval_shell_invocation_for_platform(command, EvalShellPlatform::Windows); |
| 777 | assert_eq!(windows.program, "cmd"); |
| 778 | assert_eq!(windows.args, vec!["/C".to_string(), command.to_string()]); |
| 779 | assert!(windows.raw_payload_on_windows); |
| 780 | |
| 781 | let unix = eval_shell_invocation_for_platform(command, EvalShellPlatform::Unix); |
| 782 | assert_eq!(unix.program, "sh"); |
| 783 | assert_eq!(unix.args, vec!["-c".to_string(), command.to_string()]); |
| 784 | assert!(!unix.raw_payload_on_windows); |
| 785 | } |
| 786 | } |
| 787 |