| 1 | //! Shared test-only helpers. |
| 2 | |
| 3 | use std::ffi::{OsStr, OsString}; |
| 4 | use std::path::{Path, PathBuf}; |
| 5 | use std::sync::{Mutex, OnceLock}; |
| 6 | use std::time::{SystemTime, UNIX_EPOCH}; |
| 7 | |
| 8 | pub(crate) use crate::shell_dispatcher::test_env_lock::{ |
| 9 | EnvScopeMembership, EnvScopeTicket, TestEnvLock, current_env_scope_generation, |
| 10 | current_thread_holds_test_env_lock, env_scope_ticket, join_env_scope, lock_test_env, |
| 11 | with_test_env_lock, |
| 12 | }; |
| 13 | |
| 14 | /// Process-wide state root for unit tests that do not intentionally provide an |
| 15 | /// explicit config/settings path. |
| 16 | /// |
| 17 | /// The production fallback is the user's real home. That is useful at runtime |
| 18 | /// and unsafe in a parallel test binary: an unguarded save can otherwise read |
| 19 | /// or overwrite the developer's config. Tests that exercise path precedence |
| 20 | /// still hold [`lock_test_env`] and provide explicit temporary environment |
| 21 | /// values; every other test is confined here. |
| 22 | pub(crate) fn isolated_test_state_root() -> &'static Path { |
| 23 | static ROOT: OnceLock<PathBuf> = OnceLock::new(); |
| 24 | ROOT.get_or_init(|| { |
| 25 | let nonce = SystemTime::now() |
| 26 | .duration_since(UNIX_EPOCH) |
| 27 | .unwrap_or_default() |
| 28 | .as_nanos(); |
| 29 | let root = std::env::temp_dir().join(format!( |
| 30 | "codewhale-tui-test-state-{}-{nonce}", |
| 31 | std::process::id() |
| 32 | )); |
| 33 | std::fs::create_dir_all(&root).unwrap_or_else(|error| { |
| 34 | panic!( |
| 35 | "failed to create isolated unit-test state root {}: {error}", |
| 36 | root.display() |
| 37 | ) |
| 38 | }); |
| 39 | root |
| 40 | }) |
| 41 | } |
| 42 | |
| 43 | /// Build a syntactically valid, non-secret JWT fixture without embedding a |
| 44 | /// high-entropy token-shaped literal in Git history. |
| 45 | pub(crate) fn future_test_jwt(label: &str) -> String { |
| 46 | use base64::Engine as _; |
| 47 | |
| 48 | let payload = base64::engine::general_purpose::URL_SAFE_NO_PAD.encode(br#"{"exp":9999999999}"#); |
| 49 | format!("test.{payload}.{label}") |
| 50 | } |
| 51 | |
| 52 | fn state_io_lock() -> &'static Mutex<()> { |
| 53 | static LOCK: OnceLock<Mutex<()>> = OnceLock::new(); |
| 54 | LOCK.get_or_init(|| Mutex::new(())) |
| 55 | } |
| 56 | |
| 57 | /// Serialize read/merge/write operations against the process-wide isolated |
| 58 | /// test state root. |
| 59 | /// |
| 60 | /// Path isolation protects the developer's files, but parallel tests still |
| 61 | /// share the same temporary files. Settings persistence is a multi-step |
| 62 | /// operation, so it needs this second barrier around the complete I/O |
| 63 | /// transaction rather than only around path resolution. |
| 64 | pub(crate) fn with_test_state_io_lock<T>(operation: impl FnOnce() -> T) -> T { |
| 65 | let _guard = match state_io_lock().lock() { |
| 66 | Ok(guard) => guard, |
| 67 | Err(poisoned) => poisoned.into_inner(), |
| 68 | }; |
| 69 | operation() |
| 70 | } |
| 71 | |
| 72 | /// Restore one environment variable when dropped. |
| 73 | /// |
| 74 | /// Callers that mutate process-global environment variables must hold |
| 75 | /// [`lock_test_env`] until after this guard is dropped. |
| 76 | pub(crate) struct EnvVarGuard { |
| 77 | key: &'static str, |
| 78 | previous: Option<OsString>, |
| 79 | } |
| 80 | |
| 81 | impl EnvVarGuard { |
| 82 | pub(crate) fn set(key: &'static str, value: impl AsRef<OsStr>) -> Self { |
| 83 | debug_assert!( |
| 84 | current_thread_holds_test_env_lock(), |
| 85 | "EnvVarGuard::set({key}) requires lock_test_env()" |
| 86 | ); |
| 87 | let previous = std::env::var_os(key); |
| 88 | // SAFETY: callers hold the process-wide test env mutex. |
| 89 | unsafe { std::env::set_var(key, value) }; |
| 90 | Self { key, previous } |
| 91 | } |
| 92 | |
| 93 | pub(crate) fn remove(key: &'static str) -> Self { |
| 94 | debug_assert!( |
| 95 | current_thread_holds_test_env_lock(), |
| 96 | "EnvVarGuard::remove({key}) requires lock_test_env()" |
| 97 | ); |
| 98 | let previous = std::env::var_os(key); |
| 99 | // SAFETY: callers hold the process-wide test env mutex. |
| 100 | unsafe { std::env::remove_var(key) }; |
| 101 | Self { key, previous } |
| 102 | } |
| 103 | |
| 104 | pub(crate) fn previous(&self) -> Option<OsString> { |
| 105 | self.previous.clone() |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | impl Drop for EnvVarGuard { |
| 110 | fn drop(&mut self) { |
| 111 | // SAFETY: callers hold the process-wide test env mutex until after this |
| 112 | // guard is dropped. |
| 113 | unsafe { |
| 114 | if let Some(value) = self.previous.take() { |
| 115 | std::env::set_var(self.key, value); |
| 116 | } else { |
| 117 | std::env::remove_var(self.key); |
| 118 | } |
| 119 | } |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | /// Find the byte position of the first divergence between two strings, |
| 124 | /// returning a windowed view (`±32 bytes` around the divergence) so failures |
| 125 | /// in cache-prefix-stability tests show *which* bytes drifted, not just that |
| 126 | /// they did. Returns `None` when the strings are byte-identical. |
| 127 | pub(crate) fn first_divergence(a: &str, b: &str) -> Option<(usize, String, String)> { |
| 128 | let a_bytes = a.as_bytes(); |
| 129 | let b_bytes = b.as_bytes(); |
| 130 | let max = a_bytes.len().min(b_bytes.len()); |
| 131 | for i in 0..max { |
| 132 | if a_bytes[i] != b_bytes[i] { |
| 133 | let lo = i.saturating_sub(32); |
| 134 | let a_hi = (i + 32).min(a_bytes.len()); |
| 135 | let b_hi = (i + 32).min(b_bytes.len()); |
| 136 | let a_ctx = String::from_utf8_lossy(&a_bytes[lo..a_hi]).into_owned(); |
| 137 | let b_ctx = String::from_utf8_lossy(&b_bytes[lo..b_hi]).into_owned(); |
| 138 | return Some((i, a_ctx, b_ctx)); |
| 139 | } |
| 140 | } |
| 141 | if a_bytes.len() != b_bytes.len() { |
| 142 | return Some(( |
| 143 | max, |
| 144 | format!("(len={})", a_bytes.len()), |
| 145 | format!("(len={})", b_bytes.len()), |
| 146 | )); |
| 147 | } |
| 148 | None |
| 149 | } |
| 150 | |
| 151 | /// Assert two strings are byte-identical, panicking with a windowed diff |
| 152 | /// around the first divergence when they aren't. Used by the prefix-cache |
| 153 | /// stability harness (#263, #280) to pin construction surfaces that land in |
| 154 | /// DeepSeek's KV cache prefix. |
| 155 | #[track_caller] |
| 156 | pub(crate) fn assert_byte_identical(label: &str, a: &str, b: &str) { |
| 157 | if let Some((pos, a_ctx, b_ctx)) = first_divergence(a, b) { |
| 158 | panic!( |
| 159 | "{label}: prompt construction is non-deterministic — first diff at byte {pos}\n\ |
| 160 | ── side A (±32B) ──\n{a_ctx:?}\n── side B (±32B) ──\n{b_ctx:?}", |
| 161 | ); |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | // ── Shared App/TuiOptions fixtures (#3923) ────────────────────────────── |
| 166 | // |
| 167 | // Before this module owned them, `create_test_app` was copy-pasted across 28 |
| 168 | // test modules, each spelling out the full `TuiOptions` literal — 87 literals |
| 169 | // in all. The copies had drifted: different modules pinned different locales, |
| 170 | // currencies, and onboarding flags without anyone having chosen that, which is |
| 171 | // the non-hermeticity behind the intermittent `config_command_allow_shell_*` |
| 172 | // failures. Adding a `TuiOptions` field meant editing up to 87 sites. |
| 173 | // |
| 174 | // Express intentional differences by mutating the returned value at the call |
| 175 | // site, so the difference is visible as a deliberate line of test code rather |
| 176 | // than hidden inside another near-identical literal. |
| 177 | |
| 178 | /// Default `TuiOptions` for tests, pinned to the deepseek-v4-pro fixture route. |
| 179 | pub(crate) fn test_tui_options(workspace: impl AsRef<Path>) -> crate::tui::app::TuiOptions { |
| 180 | let workspace = workspace.as_ref().to_path_buf(); |
| 181 | crate::tui::app::TuiOptions { |
| 182 | model: "deepseek-v4-pro".to_string(), |
| 183 | workspace, |
| 184 | config_path: None, |
| 185 | config_profile: None, |
| 186 | allow_shell: false, |
| 187 | use_alt_screen: true, |
| 188 | use_mouse_capture: false, |
| 189 | use_bracketed_paste: true, |
| 190 | max_subagents: 1, |
| 191 | skills_dir: PathBuf::from("."), |
| 192 | memory_path: PathBuf::from("memory.md"), |
| 193 | notes_path: PathBuf::from("notes.txt"), |
| 194 | mcp_config_path: PathBuf::from("mcp.json"), |
| 195 | use_memory: false, |
| 196 | // Majority-of-fixtures defaults, measured across the 89 literals this |
| 197 | // helper replaced. Modules that need the other value say so explicitly. |
| 198 | start_in_agent_mode: false, |
| 199 | skip_onboarding: true, |
| 200 | yolo: false, |
| 201 | resume_session_id: None, |
| 202 | initial_input: None, |
| 203 | startup_notice: None, |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | /// Build an `App` whose observable state does not depend on the developer's |
| 208 | /// machine. |
| 209 | /// |
| 210 | /// `App::new` consults real persisted settings (provider/model maps, |
| 211 | /// auto-model, route limits, locale, currency), so an un-pinned fixture |
| 212 | /// computes against whatever the developer last configured. Every pin below |
| 213 | /// exists because some test was observed to depend on it. |
| 214 | pub(crate) fn test_app_with_options(options: crate::tui::app::TuiOptions) -> crate::tui::app::App { |
| 215 | let config = crate::config::Config::default(); |
| 216 | let mut app = crate::tui::app::App::new(options, &config); |
| 217 | |
| 218 | // Deterministic presentation regardless of host locale. |
| 219 | app.cost_currency = crate::pricing::CostCurrency::Usd; |
| 220 | app.ui_locale = crate::localization::Locale::En; |
| 221 | // Transcript tests must not depend on a concurrently swapped settings |
| 222 | // home. Tests for hidden reasoning opt out explicitly. |
| 223 | app.show_thinking = true; |
| 224 | // Pin the route identity: without this, a machine with customized |
| 225 | // settings computes context-window assertions against a different model |
| 226 | // than the requested deepseek-v4-pro. |
| 227 | app.set_provider_identity(crate::config::ApiProvider::Deepseek, "deepseek"); |
| 228 | app.billing_presentation = crate::route_billing::BillingPresentation::Metered; |
| 229 | app.model = "deepseek-v4-pro".to_string(); |
| 230 | app.auto_model = false; |
| 231 | app.last_effective_model = None; |
| 232 | app.active_route_limits = None; |
| 233 | app.active_context_window_override = None; |
| 234 | // Fixtures replace `app.workspace` freely. Do not retain `App::new`'s real |
| 235 | // process cwd as a second discovery root: parallel tests and a large |
| 236 | // developer checkout can otherwise consume the bounded mention index |
| 237 | // before the fixture workspace is scanned. |
| 238 | app.composer.mention_cwd = None; |
| 239 | app |
| 240 | } |
| 241 | |
| 242 | #[cfg(test)] |
| 243 | mod tests { |
| 244 | use super::*; |
| 245 | use std::sync::mpsc; |
| 246 | use std::time::Duration; |
| 247 | |
| 248 | #[test] |
| 249 | fn unguarded_state_writes_use_isolated_test_root() { |
| 250 | const PROBE_ENV: &str = "CODEWHALE_TEST_STATE_ISOLATION_PROBE"; |
| 251 | const RECEIPT_ENV: &str = "CODEWHALE_TEST_STATE_ISOLATION_RECEIPT"; |
| 252 | |
| 253 | if std::env::var_os(PROBE_ENV).is_some() { |
| 254 | let config_path = |
| 255 | crate::config_persistence::persist_root_bool_key(None, "allow_shell", true) |
| 256 | .expect("write isolated config"); |
| 257 | let direct_config_path = |
| 258 | crate::config::save_workspace_trust(Path::new("/tmp/codewhale-test-workspace")) |
| 259 | .expect("write through direct default config path"); |
| 260 | crate::settings::Settings::default() |
| 261 | .save() |
| 262 | .expect("write isolated settings"); |
| 263 | let settings_path = |
| 264 | crate::settings::Settings::path().expect("resolve isolated settings"); |
| 265 | let root = isolated_test_state_root(); |
| 266 | assert!(config_path.starts_with(root), "{}", config_path.display()); |
| 267 | assert!( |
| 268 | settings_path.starts_with(root), |
| 269 | "{}", |
| 270 | settings_path.display() |
| 271 | ); |
| 272 | assert!( |
| 273 | direct_config_path.starts_with(root), |
| 274 | "{}", |
| 275 | direct_config_path.display() |
| 276 | ); |
| 277 | let receipt = std::env::var_os(RECEIPT_ENV).expect("receipt path"); |
| 278 | std::fs::write( |
| 279 | receipt, |
| 280 | format!( |
| 281 | "{}\n{}\n{}\n{}\n", |
| 282 | root.display(), |
| 283 | config_path.display(), |
| 284 | settings_path.display(), |
| 285 | direct_config_path.display() |
| 286 | ), |
| 287 | ) |
| 288 | .expect("write isolation receipt"); |
| 289 | return; |
| 290 | } |
| 291 | |
| 292 | let sentinel = tempfile::tempdir().expect("sentinel home"); |
| 293 | let user_state = sentinel.path().join(".codewhale"); |
| 294 | std::fs::create_dir_all(&user_state).expect("create sentinel state"); |
| 295 | let config_path = user_state.join("config.toml"); |
| 296 | let settings_path = user_state.join("settings.toml"); |
| 297 | let config_sentinel = b"# developer config sentinel\n"; |
| 298 | let settings_sentinel = b"# developer settings sentinel\n"; |
| 299 | std::fs::write(&config_path, config_sentinel).expect("seed config"); |
| 300 | std::fs::write(&settings_path, settings_sentinel).expect("seed settings"); |
| 301 | let receipt_path = sentinel.path().join("receipt.txt"); |
| 302 | |
| 303 | let output = std::process::Command::new(std::env::current_exe().expect("test binary")) |
| 304 | .arg("--exact") |
| 305 | .arg("test_support::tests::unguarded_state_writes_use_isolated_test_root") |
| 306 | .arg("--test-threads=1") |
| 307 | .env(PROBE_ENV, "1") |
| 308 | .env(RECEIPT_ENV, &receipt_path) |
| 309 | .env("HOME", sentinel.path()) |
| 310 | .env("USERPROFILE", sentinel.path()) |
| 311 | .env_remove("CODEWHALE_HOME") |
| 312 | .env_remove("CODEWHALE_CONFIG_PATH") |
| 313 | .env_remove("DEEPSEEK_CONFIG_PATH") |
| 314 | .output() |
| 315 | .expect("run isolated-state probe"); |
| 316 | assert!( |
| 317 | output.status.success(), |
| 318 | "probe failed\nstdout:\n{}\nstderr:\n{}", |
| 319 | String::from_utf8_lossy(&output.stdout), |
| 320 | String::from_utf8_lossy(&output.stderr) |
| 321 | ); |
| 322 | |
| 323 | assert_eq!( |
| 324 | std::fs::read(&config_path).expect("read config sentinel"), |
| 325 | config_sentinel |
| 326 | ); |
| 327 | assert_eq!( |
| 328 | std::fs::read(&settings_path).expect("read settings sentinel"), |
| 329 | settings_sentinel |
| 330 | ); |
| 331 | |
| 332 | let receipt = std::fs::read_to_string(&receipt_path).expect("read isolation receipt"); |
| 333 | let mut paths = receipt.lines().map(PathBuf::from); |
| 334 | let isolated_root = paths.next().expect("root receipt"); |
| 335 | let written_config = paths.next().expect("config receipt"); |
| 336 | let written_settings = paths.next().expect("settings receipt"); |
| 337 | let direct_config = paths.next().expect("direct config receipt"); |
| 338 | assert!(!isolated_root.starts_with(sentinel.path())); |
| 339 | assert!(written_config.starts_with(&isolated_root)); |
| 340 | assert!(written_settings.starts_with(&isolated_root)); |
| 341 | assert!(direct_config.starts_with(&isolated_root)); |
| 342 | assert!(written_config.exists()); |
| 343 | assert!(written_settings.exists()); |
| 344 | } |
| 345 | |
| 346 | #[test] |
| 347 | fn config_path_read_waits_for_foreign_env_redirect_to_restore() { |
| 348 | let (started_tx, started_rx) = mpsc::channel(); |
| 349 | let (tx, rx) = mpsc::channel(); |
| 350 | let redirected = std::env::temp_dir().join(format!( |
| 351 | "codewhale-config-path-read-barrier-{}", |
| 352 | std::process::id() |
| 353 | )); |
| 354 | |
| 355 | let reader = { |
| 356 | let lock = lock_test_env(); |
| 357 | let redirect = EnvVarGuard::set("DEEPSEEK_CONFIG_PATH", &redirected); |
| 358 | let reader = std::thread::spawn(move || { |
| 359 | started_tx.send(()).expect("signal config path read start"); |
| 360 | tx.send(crate::config_persistence::config_toml_path(None)) |
| 361 | .expect("send resolved config path"); |
| 362 | }); |
| 363 | |
| 364 | started_rx |
| 365 | .recv_timeout(Duration::from_secs(2)) |
| 366 | .expect("reader reached config path resolution"); |
| 367 | assert!( |
| 368 | rx.recv_timeout(Duration::from_millis(50)).is_err(), |
| 369 | "a foreign reader observed the temporary config redirect" |
| 370 | ); |
| 371 | drop(redirect); |
| 372 | drop(lock); |
| 373 | reader |
| 374 | }; |
| 375 | |
| 376 | let resolved = rx |
| 377 | .recv_timeout(Duration::from_secs(2)) |
| 378 | .expect("reader resumed after the redirect was restored") |
| 379 | .expect("resolve config path"); |
| 380 | reader.join().expect("reader thread"); |
| 381 | assert_ne!(resolved, redirected); |
| 382 | } |
| 383 | |
| 384 | #[test] |
| 385 | fn settings_save_waits_for_foreign_state_io_transaction() { |
| 386 | let (holder_ready_tx, holder_ready_rx) = mpsc::channel(); |
| 387 | let (release_tx, release_rx) = mpsc::channel(); |
| 388 | let holder = std::thread::spawn(move || { |
| 389 | with_test_state_io_lock(|| { |
| 390 | holder_ready_tx.send(()).expect("signal state lock held"); |
| 391 | release_rx.recv().expect("release state lock"); |
| 392 | }); |
| 393 | }); |
| 394 | holder_ready_rx |
| 395 | .recv_timeout(Duration::from_secs(2)) |
| 396 | .expect("holder acquired state I/O lock"); |
| 397 | |
| 398 | let (started_tx, started_rx) = mpsc::channel(); |
| 399 | let (saved_tx, saved_rx) = mpsc::channel(); |
| 400 | let writer = std::thread::spawn(move || { |
| 401 | started_tx.send(()).expect("signal settings save start"); |
| 402 | saved_tx |
| 403 | .send(crate::settings::Settings::default().save()) |
| 404 | .expect("send settings save result"); |
| 405 | }); |
| 406 | started_rx |
| 407 | .recv_timeout(Duration::from_secs(2)) |
| 408 | .expect("writer reached settings save"); |
| 409 | assert!( |
| 410 | saved_rx.recv_timeout(Duration::from_millis(50)).is_err(), |
| 411 | "settings save did not wait for an in-flight state transaction" |
| 412 | ); |
| 413 | |
| 414 | release_tx.send(()).expect("release holder"); |
| 415 | holder.join().expect("holder thread"); |
| 416 | saved_rx |
| 417 | .recv_timeout(Duration::from_secs(2)) |
| 418 | .expect("settings save resumed") |
| 419 | .expect("settings save succeeded"); |
| 420 | writer.join().expect("writer thread"); |
| 421 | } |
| 422 | } |
| 423 |