oxedyne/daimond/hand/src/verify.rs
102 KiB, 1 run
created by r2519314175:927, which is this file's identity for as long as the history lasts, whatever it is later renamed to
download · who wrote it · its history
| 1 | //! Running a named verifier from the tracked tree, and refusing to report a bare pass. |
| 2 | //! |
| 3 | //! Written on 2026-08-21. A daimon cannot produce browser evidence: `listen()` |
| 4 | //! is refused by [`crate::seccomp`], a browser needs the display server's unix |
| 5 | //! socket and [`crate::seccomp::Unix::Refuse`] takes that away, so every |
| 6 | //! `dev/verify_*.mjs` that drives a real page dies under the fence a command |
| 7 | //! gets. Half the proof of a release session is those scripts, so the machine |
| 8 | //! that could write the code could not check it. |
| 9 | //! |
| 10 | //! **The reframe this module is built on is about provenance, not sandboxing.** |
| 11 | //! The fence exists to contain a command a MODEL wrote. A verifier is the |
| 12 | //! repository's own code, and the model supplies no part of it: it supplies a |
| 13 | //! *name* that is looked up in the directory, and at most a *break* that is |
| 14 | //! looked up in the file's own declarations. So the verb is in the same trust |
| 15 | //! class as `cargo test`, and it runs the script outside the command fence |
| 16 | //! deliberately -- which is stated in the report, stated in the tool's |
| 17 | //! description, and recorded in the journal as `fence:none` so that nobody has |
| 18 | //! to take this paragraph's word for it. |
| 19 | //! |
| 20 | //! # And the reframe has to be enforced, not reported |
| 21 | //! |
| 22 | //! Until 2026-08-25 it was not. `REVIEW.md` §1.21: the trustworthiness of a |
| 23 | //! verifier was measured -- `git ls-files --error-unmatch` -- and its only use |
| 24 | //! was a phrase in the report's first line. A daimon writes `dev/verify_x.mjs` |
| 25 | //! with `file_write`, which every turn may do because the granted root is |
| 26 | //! writable, and the verb runs it with no fence on it at all. Proved with a |
| 27 | //! script whose whole body wrote a file outside the granted root, which appeared. |
| 28 | //! |
| 29 | //! **The question is content, not membership.** Appending one comment to a |
| 30 | //! tracked verifier leaves `git ls-files --error-unmatch` exiting 0 while |
| 31 | //! `git diff --quiet HEAD` on the same file exits 1, so enforcing what was |
| 32 | //! already measured would have narrowed the hole without closing it. |
| 33 | //! [`provenance`] therefore asks whether the working file IS the commit, and |
| 34 | //! [`unfenced_refusal`] is the gate; [`conduct`] passes it before anything runs |
| 35 | //! and [`once`] before every spawn. |
| 36 | //! |
| 37 | //! **The refusal is a capability difference and not a dead end**, which matters |
| 38 | //! because improving a verifier is exactly the self-development this app is for. |
| 39 | //! A verifier the daimon has just written still runs -- through `run`, inside |
| 40 | //! the same fence every command gets, which 253 of this tree's 282 verifiers |
| 41 | //! need nothing more than. What it cannot do fenced is drive a browser or open |
| 42 | //! a port, and the refusal says so, so a fenced failure is not misread as the |
| 43 | //! script being broken. |
| 44 | //! |
| 45 | //! **What this does not close, and cannot from here.** `.git` is inside the |
| 46 | //! granted root and every fence leaves it writable -- `src/tools.rs` says so in |
| 47 | //! its own words where it explains why a push injects `core.hooksPath` -- so a |
| 48 | //! daimon can commit its own verifier and then be told, truthfully, that the |
| 49 | //! bytes are the commit's. That is one `run` call away and no check on this |
| 50 | //! side reaches it. What changes is that the act is now a commit in the |
| 51 | //! repository's history rather than a `file_write` nobody sees, and that the |
| 52 | //! report's first line stops being false. Closing it needs either `.git` out |
| 53 | //! of the fence or the owner's consent bound to the bytes; both are the owner's |
| 54 | //! to rule on, and the reasoning is in `REVIEW.md` §1.21. |
| 55 | //! |
| 56 | //! # The one thing this module exists to make impossible |
| 57 | //! |
| 58 | //! Running a verifier clean and reporting "27 checks passed" is exactly the |
| 59 | //! evidence that has been lying. On 2026-08-20 three separate instruments |
| 60 | //! passed while aimed at nothing: a `--break` whose output was byte-identical to |
| 61 | //! a clean run, a leg count blind to a 1,856-request runaway, and a check |
| 62 | //! asserting a destination a re-route also reaches. All three were caught |
| 63 | //! because a person demanded a break and the agents chose to report honestly. |
| 64 | //! |
| 65 | //! Honesty must not be load-bearing. So the verb runs the clean pass AND each |
| 66 | //! declared break, and [`Verdict`] is an enum whose every arm carries the count |
| 67 | //! of breaks that reddened nothing. There is no accessor for the passed count |
| 68 | //! on its own: a caller who wants it must match, and both arms hand back the |
| 69 | //! bad news in the same breath. |
| 70 | |
| 71 | use crate::journal::{ |
| 72 | Event, |
| 73 | Journal, |
| 74 | }; |
| 75 | use crate::wire::{ |
| 76 | Capture, |
| 77 | FenceSpec, |
| 78 | Req, |
| 79 | Resp, |
| 80 | Stream, |
| 81 | CHUNK_MAX, |
| 82 | }; |
| 83 | |
| 84 | use oxedyne_fe2o3_core::prelude::*; |
| 85 | |
| 86 | use std::{ |
| 87 | collections::BTreeMap, |
| 88 | path::{ |
| 89 | Path, |
| 90 | PathBuf, |
| 91 | }, |
| 92 | process::Stdio, |
| 93 | sync::{ |
| 94 | atomic::{ |
| 95 | AtomicBool, |
| 96 | Ordering, |
| 97 | }, |
| 98 | Arc, |
| 99 | Mutex, |
| 100 | }, |
| 101 | time::Duration, |
| 102 | }; |
| 103 | |
| 104 | use tokio::{ |
| 105 | io::AsyncReadExt, |
| 106 | process::Command, |
| 107 | sync::mpsc::Sender, |
| 108 | }; |
| 109 | |
| 110 | // ┌───────────────────────────────────────────────────────────────┐ |
| 111 | // │ Where a verifier lives, and what it may be called │ |
| 112 | // └───────────────────────────────────────────────────────────────┘ |
| 113 | |
| 114 | /// The one directory under the granted root a verifier may be read from. |
| 115 | pub const DEV_DIR: &str = "dev"; |
| 116 | |
| 117 | /// The one file-name shape a verifier may have. |
| 118 | pub const PREFIX: &str = "verify_"; |
| 119 | |
| 120 | /// The one extension a verifier may have. |
| 121 | pub const SUFFIX: &str = ".mjs"; |
| 122 | |
| 123 | /// The longest a verifier's short name, or a break's, may be. |
| 124 | pub const NAME_MAX: usize = 64; |
| 125 | |
| 126 | /// How much of one run's output is kept in memory. |
| 127 | /// |
| 128 | /// Only the check lines and the clean run's text survive past the comparison, |
| 129 | /// so this bounds a working buffer rather than the report. |
| 130 | pub const OUT_MAX: usize = 8 * 1024 * 1024; |
| 131 | |
| 132 | /// The default whole-sequence budget, when the caller names none. |
| 133 | pub const BUDGET_DEFAULT_MS: u64 = 20 * 60 * 1_000; |
| 134 | |
| 135 | /// The largest whole-sequence budget a caller may ask for. |
| 136 | pub const BUDGET_MAX_MS: u64 = 4 * 60 * 60 * 1_000; |
| 137 | |
| 138 | /// The least a single run is given before the budget is called spent. |
| 139 | /// |
| 140 | /// A break handed four seconds is a break that will time out and be reported as |
| 141 | /// unrun, which is worse than saying plainly that the budget ran out. |
| 142 | pub const RUN_FLOOR_MS: u64 = 15 * 1_000; |
| 143 | |
| 144 | /// Where a verifier leaves pictures, relative to the granted root. |
| 145 | pub const SHOTS_DIR: &str = "dev/shots"; |
| 146 | |
| 147 | /// The most shot paths one report will name. |
| 148 | pub const SHOTS_MAX: usize = 40; |
| 149 | |
| 150 | /// The largest report this module will send, in bytes. |
| 151 | /// |
| 152 | /// Under [`CHUNK_MAX`] so the whole report is one chunk and a reader never has |
| 153 | /// to reassemble it. |
| 154 | pub const REPORT_MAX: usize = CHUNK_MAX - 4_096; |
| 155 | |
| 156 | /// The marker every report ends with, which the app checks for. |
| 157 | /// |
| 158 | /// A second line of defence at the far end: `Tool::Verify` refuses to hand a |
| 159 | /// model a result with no trailer, because a result with no trailer is a result |
| 160 | /// whose three numbers were never computed. |
| 161 | pub const TRAILER: &str = "[verify:"; |
| 162 | |
| 163 | /// The exit status of a sequence that proved itself. |
| 164 | pub const EXIT_PROVED: i32 = 0; |
| 165 | |
| 166 | /// The exit status of a sequence whose clean run had failures. |
| 167 | pub const EXIT_FAILED: i32 = 1; |
| 168 | |
| 169 | /// The exit status of a clean run nothing has proved. |
| 170 | pub const EXIT_UNPROVEN: i32 = 2; |
| 171 | |
| 172 | // ┌───────────────────────────────────────────────────────────────┐ |
| 173 | // │ Nobody is at the keyboard for a verifier this verb runs │ |
| 174 | // └───────────────────────────────────────────────────────────────┘ |
| 175 | // |
| 176 | // A verifier is spawned with the hand's own environment, and the hand's own |
| 177 | // environment is whatever started it. Started by the browser as a native |
| 178 | // messaging host, that is the browser's -- `DISPLAY=:0` and `WAYLAND_DISPLAY` |
| 179 | // among it, because the browser is on the owner's screen. Half the verifiers in |
| 180 | // `dev/` drive a HEADED browser, and `dev/display.mjs` allowed `:0` on the stated |
| 181 | // grounds that watching a headed run on one's own seat is a thing people do. |
| 182 | // |
| 183 | // It is. Nobody is watching this one. A model asked for it, and the window |
| 184 | // would land in front of whoever happens to be at the machine, mid-sentence, |
| 185 | // exactly as it did on 2026-08-24 -- the incident that file's header is written |
| 186 | // about. That fault was a check reading the display STRING and not the |
| 187 | // environment; this is the same shape one layer up, a check reading the |
| 188 | // environment and not knowing who asked. |
| 189 | // |
| 190 | // The hand knows who asked, and it is the only party that does. So a verifier is |
| 191 | // handed no display at all and is told that none is coming: it must start one of |
| 192 | // its own, which `dev/verify_reflux.mjs` does and every other headed verifier is |
| 193 | // free to. Refusing loudly is the point. A headed verifier reached this way |
| 194 | // used to paint on the seat and now says it cannot, which is a capability this |
| 195 | // verb never honestly had. |
| 196 | |
| 197 | /// The display names a verifier is never handed, whatever the hand was started with. |
| 198 | pub const SEAT_VARS: &[&str] = &["DISPLAY", "WAYLAND_DISPLAY", "XDG_SESSION_TYPE"]; |
| 199 | |
| 200 | /// The name that tells `dev/display.mjs` nobody is at the keyboard for this run. |
| 201 | pub const UNATTENDED_VAR: &str = "DAIMOND_UNATTENDED"; |
| 202 | |
| 203 | /// The environment a verifier is spawned with, from the hand's own. |
| 204 | /// |
| 205 | /// Separate and pure so a test can put an environment to it without spawning |
| 206 | /// anything -- the mistake `dev/verify_harness.mjs` exists because of was a rule |
| 207 | /// that had no test because testing it meant opening a window. |
| 208 | /// |
| 209 | /// # Arguments |
| 210 | /// * `env` - The hand's own environment, name to value. |
| 211 | pub fn unattended_env(env: &BTreeMap<String, String>) -> BTreeMap<String, String> { |
| 212 | let mut out = env.clone(); |
| 213 | for v in SEAT_VARS { |
| 214 | out.remove(*v); |
| 215 | } |
| 216 | out.insert(fmt!("{}", UNATTENDED_VAR), fmt!("1")); |
| 217 | out |
| 218 | } |
| 219 | |
| 220 | /// The hand's own environment, as a map. |
| 221 | fn own_env() -> BTreeMap<String, String> { |
| 222 | std::env::vars().collect() |
| 223 | } |
| 224 | |
| 225 | /// Is this a name a verifier may be looked up by? |
| 226 | /// |
| 227 | /// Deliberately narrower than a file name. Lower case, digits and underscore |
| 228 | /// are the whole alphabet every verifier in the tree is spelled with, and they |
| 229 | /// leave nothing for a path, a flag or a shell to be made of: no dot, so `..` |
| 230 | /// cannot be written; no slash; no dash, so nothing can begin with one and be |
| 231 | /// read as an option. |
| 232 | /// |
| 233 | /// # Arguments |
| 234 | /// * `name` - The short name, as the model wrote it. |
| 235 | pub fn name_ok(name: &str) -> bool { |
| 236 | !name.is_empty() |
| 237 | && name.len() <= NAME_MAX |
| 238 | && name.chars().all(|c| c.is_ascii_lowercase() || c.is_ascii_digit() || c == '_') |
| 239 | } |
| 240 | |
| 241 | /// Why a file is not the commit's. |
| 242 | #[derive(Clone, Copy, Debug, Eq, PartialEq)] |
| 243 | pub enum Change { |
| 244 | Untracked, // git has never heard of the path |
| 245 | Edited, // git has, and the working copy is not what the commit holds |
| 246 | Linked, // the path is a symlink, so its bytes are wherever it points |
| 247 | } |
| 248 | |
| 249 | impl Change { |
| 250 | /// The clause the refusal and the report both put after the file's name. |
| 251 | pub fn phrase(&self) -> &'static str { |
| 252 | match self { |
| 253 | Self::Untracked => "git has never heard of it", |
| 254 | Self::Edited => "it is tracked, and the working copy is not what the commit \ |
| 255 | holds -- staging a change is not committing it", |
| 256 | Self::Linked => "it is a symbolic link, so its bytes are wherever it points \ |
| 257 | and the commit says nothing about them", |
| 258 | } |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | /// Whose bytes these are, as the repository can answer it. |
| 263 | /// |
| 264 | /// **The question is content and not membership**, and the difference is the |
| 265 | /// whole of `REVIEW.md` §1.21. `git ls-files --error-unmatch` answers about the |
| 266 | /// INDEX: append one comment to a tracked verifier and it still exits 0, while |
| 267 | /// `git diff --quiet HEAD` on the same file exits 1. The claim the unfenced run |
| 268 | /// rests on is *these bytes came with the checkout*, and only the second question |
| 269 | /// answers it. |
| 270 | #[derive(Clone, Debug, Eq, PartialEq)] |
| 271 | pub enum Provenance { |
| 272 | /// The working file is byte for byte what the commit holds. |
| 273 | Committed, |
| 274 | /// It is not, and this is how. |
| 275 | Changed(Change), |
| 276 | /// Nothing could be asked, and this is why. |
| 277 | /// |
| 278 | /// Refused rather than assumed either way. Four verifiers were once |
| 279 | /// promoted into `dev/` and left untracked, and `dev/gate.sh` builds its |
| 280 | /// tree with `git worktree add` -- a clean checkout of a commit -- so the |
| 281 | /// suite ran none of them while reporting a pass. |
| 282 | Unknown(String), |
| 283 | } |
| 284 | |
| 285 | impl Provenance { |
| 286 | /// May a file of this provenance run outside the command fence? |
| 287 | pub fn committed(&self) -> bool { |
| 288 | matches!(self, Self::Committed) |
| 289 | } |
| 290 | |
| 291 | /// The phrase the report carries. |
| 292 | pub fn phrase(&self) -> String { |
| 293 | match self { |
| 294 | Self::Committed => fmt!("byte for byte the commit's"), |
| 295 | Self::Changed(c) => fmt!("NOT THE COMMIT'S -- {}", c.phrase()), |
| 296 | Self::Unknown(w) => fmt!("provenance unknown ({})", w), |
| 297 | } |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | /// One verifier, as the tree actually holds it. |
| 302 | #[derive(Clone, Debug)] |
| 303 | pub struct Script { |
| 304 | /// The short name it was looked up by. |
| 305 | pub name: String, |
| 306 | /// The file name, taken from the directory entry and never from the caller. |
| 307 | pub file: String, |
| 308 | /// The absolute path, built from the directory and that entry. |
| 309 | pub path: PathBuf, |
| 310 | /// Every break the file declares, in the order it declares them. |
| 311 | pub breaks: Vec<String>, |
| 312 | /// Whose bytes these are, as the repository can answer it. |
| 313 | pub prov: Provenance, |
| 314 | } |
| 315 | |
| 316 | /// Every break name a verifier's own source declares. |
| 317 | /// |
| 318 | /// The declaration is a comment, because that is how this tree has always |
| 319 | /// spelled it -- `dev/breakcheck.mjs` already reads the same text for the same |
| 320 | /// purpose: |
| 321 | /// |
| 322 | /// ```text |
| 323 | /// // node dev/verify_about.mjs --break nobutton # 1: no About button in the bar |
| 324 | /// ``` |
| 325 | /// |
| 326 | /// So the set is derived from the file rather than kept beside it, and a break |
| 327 | /// that has been deleted stops being offered the moment it goes. A `--break` |
| 328 | /// followed by anything but a lower-case letter -- `<name>`, `${BREAK}`, a |
| 329 | /// closing quote -- is the file talking about the flag rather than declaring a |
| 330 | /// value, and is passed over. |
| 331 | /// |
| 332 | /// # Arguments |
| 333 | /// * `src` - The verifier's source. |
| 334 | pub fn declared_breaks(src: &str) -> Vec<String> { |
| 335 | let flag = "--break"; |
| 336 | let b = src.as_bytes(); |
| 337 | let mut out: Vec<String> = Vec::new(); |
| 338 | let mut i = 0usize; |
| 339 | while let Some(hit) = src[i..].find(flag) { |
| 340 | let mut j = i + hit + flag.len(); |
| 341 | i = j; |
| 342 | // The separator, which is the whole of what tells a declaration from a mention. |
| 343 | match b.get(j) { |
| 344 | Some(b' ') | Some(b'\t') | Some(b'=') => j += 1, |
| 345 | _ => continue, |
| 346 | } |
| 347 | while let Some(b' ') | Some(b'\t') = b.get(j) { |
| 348 | j += 1; |
| 349 | } |
| 350 | let start = j; |
| 351 | while let Some(c) = b.get(j) { |
| 352 | if c.is_ascii_lowercase() || c.is_ascii_digit() || *c == b'_' { |
| 353 | j += 1; |
| 354 | } else { |
| 355 | break; |
| 356 | } |
| 357 | } |
| 358 | if j == start { |
| 359 | continue; |
| 360 | } |
| 361 | let word = &src[start..j]; |
| 362 | // A leading digit or underscore is not a break name in this tree, and taking one |
| 363 | // would offer the model a flag it cannot use. |
| 364 | match word.chars().next() { |
| 365 | Some(c) if c.is_ascii_lowercase() => (), |
| 366 | _ => continue, |
| 367 | } |
| 368 | if word.len() <= NAME_MAX && !out.iter().any(|w| w == word) { |
| 369 | out.push(fmt!("{}", word)); |
| 370 | } |
| 371 | } |
| 372 | out |
| 373 | } |
| 374 | |
| 375 | /// Every verifier the tree holds, by short name. |
| 376 | /// |
| 377 | /// # Arguments |
| 378 | /// * `root` - The granted root. |
| 379 | pub fn catalogue(root: &Path) -> Outcome<Vec<String>> { |
| 380 | let dir = root.join(DEV_DIR); |
| 381 | let rd = res!(std::fs::read_dir(&dir).map_err(|e| err!(e, |
| 382 | "The verifiers live in '{}', which could not be read.", dir.display(); |
| 383 | IO, Missing))); |
| 384 | let mut out = Vec::new(); |
| 385 | for entry in rd { |
| 386 | let entry = match entry { |
| 387 | Ok(e) => e, |
| 388 | Err(_) => continue, |
| 389 | }; |
| 390 | let name = entry.file_name(); |
| 391 | let name = match name.to_str() { |
| 392 | Some(s) => s, |
| 393 | None => continue, |
| 394 | }; |
| 395 | if name.starts_with(PREFIX) && name.ends_with(SUFFIX) { |
| 396 | let short = &name[PREFIX.len()..name.len() - SUFFIX.len()]; |
| 397 | if name_ok(short) { |
| 398 | out.push(fmt!("{}", short)); |
| 399 | } |
| 400 | } |
| 401 | } |
| 402 | out.sort(); |
| 403 | Ok(out) |
| 404 | } |
| 405 | |
| 406 | /// Whether this machine has any verifier to run at all. |
| 407 | /// |
| 408 | /// What the `verify:` capability in the handshake is computed from, so the page |
| 409 | /// can tell a model "not on this computer" rather than letting it discover the |
| 410 | /// same thing one refusal at a time. |
| 411 | /// |
| 412 | /// # Arguments |
| 413 | /// * `root` - The granted root. |
| 414 | pub fn available(root: &Path) -> bool { |
| 415 | match catalogue(root) { |
| 416 | Ok(v) => !v.is_empty(), |
| 417 | Err(_) => false, |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | /// The capability entry the handshake carries. |
| 422 | /// |
| 423 | /// # Arguments |
| 424 | /// * `root` - The granted root. |
| 425 | pub fn cap(root: &Path) -> String { |
| 426 | match available(root) { |
| 427 | true => fmt!("verify:dev"), |
| 428 | false => fmt!("verify:none"), |
| 429 | } |
| 430 | } |
| 431 | |
| 432 | /// The verifier a name refers to, or the sentence saying why there is none. |
| 433 | /// |
| 434 | /// **Every string that leaves here came from the file system or from the file.** |
| 435 | /// `file` is the directory entry's own name, `path` is that entry joined onto |
| 436 | /// the directory, and `breaks` is parsed out of the source. The caller's `name` |
| 437 | /// is used to *match* and is never used to *build*, which is the whole of the |
| 438 | /// argument that nothing the model wrote can reach a process. |
| 439 | /// |
| 440 | /// # Arguments |
| 441 | /// * `root` - The granted root. |
| 442 | /// * `name` - The short name, as the model wrote it. |
| 443 | pub fn resolve(root: &Path, name: &str) -> Result<Script, String> { |
| 444 | if !name_ok(name) { |
| 445 | return Err(fmt!( |
| 446 | "Refused: '{}' is not a verifier name. A name is lower-case letters, digits and \ |
| 447 | underscores -- 'graph' for dev/verify_graph.mjs. This verb takes a NAME and not a \ |
| 448 | path or a command line: there is nothing here that would run a path you wrote.", |
| 449 | trim_for_message(name))); |
| 450 | } |
| 451 | let dir = root.join(DEV_DIR); |
| 452 | let wanted = fmt!("{}{}{}", PREFIX, name, SUFFIX); |
| 453 | let rd = match std::fs::read_dir(&dir) { |
| 454 | Ok(r) => r, |
| 455 | Err(e) => return Err(fmt!( |
| 456 | "Refused: this computer's granted folder has no readable '{}' directory ({}), so \ |
| 457 | there are no verifiers on it to run.", dir.display(), e)), |
| 458 | }; |
| 459 | let mut found: Option<std::ffi::OsString> = None; |
| 460 | for entry in rd { |
| 461 | let entry = match entry { |
| 462 | Ok(e) => e, |
| 463 | Err(_) => continue, |
| 464 | }; |
| 465 | if entry.file_name() == std::ffi::OsStr::new(&wanted) { |
| 466 | found = Some(entry.file_name()); |
| 467 | break; |
| 468 | } |
| 469 | } |
| 470 | let file = match found { |
| 471 | Some(f) => f, |
| 472 | None => return Err(no_such(root, name, &wanted)), |
| 473 | }; |
| 474 | let path = dir.join(&file); |
| 475 | match std::fs::metadata(&path) { |
| 476 | Ok(m) if m.is_file() => (), |
| 477 | Ok(_) => return Err(fmt!( |
| 478 | "Refused: '{}' is not a file, so there is nothing to run.", path.display())), |
| 479 | Err(e) => return Err(fmt!( |
| 480 | "Refused: '{}' could not be read ({}), so there is nothing to run.", |
| 481 | path.display(), e)), |
| 482 | } |
| 483 | let src = match std::fs::read_to_string(&path) { |
| 484 | Ok(s) => s, |
| 485 | Err(e) => return Err(fmt!( |
| 486 | "Refused: '{}' could not be read as text ({}), so its breaks cannot be listed and \ |
| 487 | it will not be run blind.", path.display(), e)), |
| 488 | }; |
| 489 | let file_str = match file.to_str() { |
| 490 | Some(s) => fmt!("{}", s), |
| 491 | None => return Err(fmt!( |
| 492 | "Refused: the name of '{}' is not UTF-8, so it cannot be reported honestly.", |
| 493 | path.display())), |
| 494 | }; |
| 495 | Ok(Script { |
| 496 | name: fmt!("{}", name), |
| 497 | breaks: declared_breaks(&src), |
| 498 | prov: provenance(root, &file_str), |
| 499 | file: file_str, |
| 500 | path, |
| 501 | }) |
| 502 | } |
| 503 | |
| 504 | /// The sentence for a name that names nothing, with the near misses in it. |
| 505 | fn no_such(root: &Path, name: &str, wanted: &str) -> String { |
| 506 | let near = match catalogue(root) { |
| 507 | Ok(all) => { |
| 508 | let mut n: Vec<String> = all.into_iter() |
| 509 | .filter(|c| c.contains(name) || name.contains(c.as_str())) |
| 510 | .take(8) |
| 511 | .collect(); |
| 512 | n.sort(); |
| 513 | n |
| 514 | }, |
| 515 | Err(_) => Vec::new(), |
| 516 | }; |
| 517 | let tail = match near.is_empty() { |
| 518 | true => fmt!("Ask for a file listing of 'dev' to see what there is."), |
| 519 | false => fmt!("Names close to it: {}.", near.join(", ")), |
| 520 | }; |
| 521 | fmt!( |
| 522 | "Refused: there is no 'dev/{}' in the folder this hand was granted, so '{}' names no \ |
| 523 | verifier. {} Nothing was run.", wanted, trim_for_message(name), tail) |
| 524 | } |
| 525 | |
| 526 | /// A caller's string, cut short and stripped of control characters, for a message. |
| 527 | /// |
| 528 | /// A refusal quotes what the model asked for so it can see its own mistake, and |
| 529 | /// a refusal is written to the journal -- so a hundred kilobytes of newlines |
| 530 | /// must not become a hundred kilobytes of journal. |
| 531 | fn trim_for_message(s: &str) -> String { |
| 532 | let mut out = String::new(); |
| 533 | for c in s.chars() { |
| 534 | if out.chars().count() >= 40 { |
| 535 | out.push('…'); |
| 536 | break; |
| 537 | } |
| 538 | match c.is_control() { |
| 539 | true => out.push('·'), |
| 540 | false => out.push(c), |
| 541 | } |
| 542 | } |
| 543 | out |
| 544 | } |
| 545 | |
| 546 | /// Whose bytes `dev/<file>` holds, asked of git rather than assumed. |
| 547 | /// |
| 548 | /// Three questions and the order matters. Is there a repository here at all -- |
| 549 | /// because a granted folder that is not one can vouch for nothing, and the |
| 550 | /// permissive reading of that was half of `REVIEW.md` §1.21's reproduction. Is |
| 551 | /// the path a symlink -- because a committed link's target is an ordinary file |
| 552 | /// inside the granted root that any command may rewrite, which is §1.1's shape |
| 553 | /// wearing §1.21's clothes. And last, does the working file differ from the |
| 554 | /// commit: `git diff --quiet HEAD` compares the WORKING TREE with `HEAD`, so it |
| 555 | /// catches a staged change and an unstaged one alike, and it applies the |
| 556 | /// repository's own end-of-line and filter settings, which a hash of the bytes |
| 557 | /// taken here would not. |
| 558 | /// |
| 559 | /// Every argument vector is fixed but for one directory entry's own name, and |
| 560 | /// there is no shell, so `--` before it is belt to the braces. |
| 561 | /// |
| 562 | /// # Arguments |
| 563 | /// * `root` - The granted root, which is where git is asked. |
| 564 | /// * `file` - The verifier's file name, from the directory entry. |
| 565 | pub fn provenance(root: &Path, file: &str) -> Provenance { |
| 566 | let git = match on_path("git") { |
| 567 | Some(p) => p, |
| 568 | None => return Provenance::Unknown(fmt!( |
| 569 | "there is no git on this hand's PATH, so nothing here can say whose code this is")), |
| 570 | }; |
| 571 | let rel = fmt!("{}/{}", DEV_DIR, file); |
| 572 | let ask = |args: &[&str]| -> std::io::Result<std::process::ExitStatus> { |
| 573 | std::process::Command::new(&git) |
| 574 | .arg("-C").arg(root) |
| 575 | .args(args) |
| 576 | .stdin(Stdio::null()) |
| 577 | .stdout(Stdio::null()) |
| 578 | .stderr(Stdio::null()) |
| 579 | .status() |
| 580 | }; |
| 581 | match ask(&["rev-parse", "--git-dir"]) { |
| 582 | Ok(s) if s.success() => (), |
| 583 | Ok(_) => return Provenance::Unknown(fmt!( |
| 584 | "the folder this hand was granted is not a git repository, so there is no commit \ |
| 585 | to compare this file with")), |
| 586 | Err(e) => return Provenance::Unknown(fmt!("git could not be run: {}", e)), |
| 587 | } |
| 588 | // Asked without following the link, which is the whole point: `metadata` above |
| 589 | // followed it and answered about the target. |
| 590 | match std::fs::symlink_metadata(root.join(&rel)) { |
| 591 | Ok(m) if m.file_type().is_symlink() => return Provenance::Changed(Change::Linked), |
| 592 | Ok(_) => (), |
| 593 | Err(e) => return Provenance::Unknown(fmt!( |
| 594 | "'{}' could not be looked at ({})", rel, e)), |
| 595 | } |
| 596 | match ask(&["ls-files", "--error-unmatch", "--", &rel]) { |
| 597 | Ok(s) if s.success() => (), |
| 598 | Ok(_) => return Provenance::Changed(Change::Untracked), |
| 599 | Err(e) => return Provenance::Unknown(fmt!("git could not be run: {}", e)), |
| 600 | } |
| 601 | // `git diff` says 0 for no difference and 1 for a difference; anything else is git |
| 602 | // failing to answer, and an unborn HEAD in a repository with no commits is exactly |
| 603 | // that. A failure to answer is never read as agreement. |
| 604 | match ask(&["diff", "--quiet", "HEAD", "--", &rel]) { |
| 605 | Ok(s) if s.success() => Provenance::Committed, |
| 606 | Ok(s) if s.code() == Some(1) => Provenance::Changed(Change::Edited), |
| 607 | Ok(s) => Provenance::Unknown(fmt!( |
| 608 | "git could not compare '{}' with the commit (it exited {}); a repository with no \ |
| 609 | commit in it yet is the usual reason", rel, match s.code() { |
| 610 | Some(c) => fmt!("{}", c), |
| 611 | None => fmt!("on a signal"), |
| 612 | })), |
| 613 | Err(e) => Provenance::Unknown(fmt!("git could not be run: {}", e)), |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | /// The sentence a verifier that is not the commit's is refused with, or `None`. |
| 618 | /// |
| 619 | /// **This is the gate `REVIEW.md` §1.21 was open for**, and it is asked fresh |
| 620 | /// rather than read off the [`Script`] resolved earlier: a verifier sequence |
| 621 | /// runs for minutes, a background command started by an earlier turn can rewrite |
| 622 | /// a file while it does, and a check taken once at the start would be answering |
| 623 | /// about bytes that are no longer there. [`conduct`] asks before anything runs |
| 624 | /// and [`once`] asks again before each spawn, so the window between the answer |
| 625 | /// and the `execve` is as small as this side can make it. |
| 626 | /// |
| 627 | /// **The sentence is written to be acted on in one call.** A refusal a model |
| 628 | /// cannot converge on costs the run anyway -- so it hands over the command that |
| 629 | /// runs the same file INSIDE the fence, which is where a command a model wrote |
| 630 | /// belongs and is enough for the nine verifiers in ten here that only read the |
| 631 | /// tree; it says what that cannot do, so a fenced failure is not misread as the |
| 632 | /// verifier being broken; and it names the two things that reach the unfenced |
| 633 | /// run, one of which is a person. |
| 634 | /// |
| 635 | /// # Arguments |
| 636 | /// * `root` - The granted root. |
| 637 | /// * `file` - The verifier's file name, from the directory entry. |
| 638 | pub fn unfenced_refusal(root: &Path, file: &str) -> Option<String> { |
| 639 | let prov = provenance(root, file); |
| 640 | let why = match &prov { |
| 641 | Provenance::Committed => return None, |
| 642 | Provenance::Changed(c) => fmt!("{}", c.phrase()), |
| 643 | Provenance::Unknown(w) => fmt!("{}", w), |
| 644 | }; |
| 645 | Some(fmt!( |
| 646 | "Refused: dev/{} is not this repository's committed code -- {}. Nothing was run. A \ |
| 647 | verifier is the one thing this hand runs OUTSIDE the command fence, and the whole of \ |
| 648 | the reason is that its bytes came with the checkout instead of from a model; this \ |
| 649 | hand cannot tell your edit from anybody else's, so it will not run one unfenced. \ |
| 650 | RUN IT YOURSELF INSTEAD: 'run' with [\"node\",\"dev/{}\"] runs this same file inside \ |
| 651 | the fence every command gets, which is all a verifier that reads the tree needs. It \ |
| 652 | is not enough for one that drives a browser or opens a port -- the fence refuses \ |
| 653 | both, so a failure there is the fence and not your script. To get the unfenced run, \ |
| 654 | commit the file and ask again, or ask the person to run it themselves.", |
| 655 | file, why, file)) |
| 656 | } |
| 657 | |
| 658 | /// The first entry on `PATH` that is a runnable file with this name. |
| 659 | /// |
| 660 | /// The hand's own `PATH`, which is the browser's, since Chrome hands a native |
| 661 | /// messaging host its own environment. A name and never a path: this is asked |
| 662 | /// only about `node` and `git`, both of which are written down here. |
| 663 | /// |
| 664 | /// # Arguments |
| 665 | /// * `prog` - The program's bare name. |
| 666 | pub fn on_path(prog: &str) -> Option<PathBuf> { |
| 667 | let path = match std::env::var("PATH") { |
| 668 | Ok(p) => p, |
| 669 | Err(_) => return None, |
| 670 | }; |
| 671 | for dir in path.split(':') { |
| 672 | if dir.is_empty() { |
| 673 | continue; |
| 674 | } |
| 675 | let cand = Path::new(dir).join(prog); |
| 676 | if runnable(&cand) { |
| 677 | return Some(cand); |
| 678 | } |
| 679 | } |
| 680 | None |
| 681 | } |
| 682 | |
| 683 | /// Is this a file the kernel would execute? |
| 684 | #[cfg(unix)] |
| 685 | fn runnable(p: &Path) -> bool { |
| 686 | use std::os::unix::fs::PermissionsExt; |
| 687 | match std::fs::metadata(p) { |
| 688 | Ok(m) => m.is_file() && (m.permissions().mode() & 0o111) != 0, |
| 689 | Err(_) => false, |
| 690 | } |
| 691 | } |
| 692 | |
| 693 | /// Is this a file the kernel would execute? |
| 694 | #[cfg(not(unix))] |
| 695 | fn runnable(p: &Path) -> bool { |
| 696 | match std::fs::metadata(p) { |
| 697 | Ok(m) => m.is_file(), |
| 698 | Err(_) => false, |
| 699 | } |
| 700 | } |
| 701 | |
| 702 | /// Every file under `dev/shots` this sequence wrote, workspace-relative. |
| 703 | /// |
| 704 | /// Named rather than returned. A daimon reads a picture by asking `file_read` |
| 705 | /// for it with `"as":"image"`, or by dispatching a worker who can see -- so what |
| 706 | /// the report owes it is the PATH, and handing back the bytes would spend a |
| 707 | /// context window on an image the model may not be able to look at anyway. |
| 708 | /// |
| 709 | /// Compared by modification time against the moment the sequence started, so an |
| 710 | /// old screenshot from last week's run is not claimed as this run's evidence. |
| 711 | /// |
| 712 | /// # Arguments |
| 713 | /// * `root` - The granted root. |
| 714 | /// * `since` - When the sequence began. |
| 715 | pub fn shots_since(root: &Path, since: std::time::SystemTime) -> Vec<String> { |
| 716 | let mut out = Vec::new(); |
| 717 | let base = root.join(SHOTS_DIR); |
| 718 | let mut todo = vec![base.clone()]; |
| 719 | while let Some(dir) = todo.pop() { |
| 720 | if out.len() >= SHOTS_MAX { |
| 721 | break; |
| 722 | } |
| 723 | let rd = match std::fs::read_dir(&dir) { |
| 724 | Ok(r) => r, |
| 725 | Err(_) => continue, |
| 726 | }; |
| 727 | for entry in rd { |
| 728 | let entry = match entry { |
| 729 | Ok(e) => e, |
| 730 | Err(_) => continue, |
| 731 | }; |
| 732 | let path = entry.path(); |
| 733 | let meta = match entry.metadata() { |
| 734 | Ok(m) => m, |
| 735 | Err(_) => continue, |
| 736 | }; |
| 737 | if meta.is_dir() { |
| 738 | todo.push(path); |
| 739 | continue; |
| 740 | } |
| 741 | let fresh = match meta.modified() { |
| 742 | Ok(t) => t >= since, |
| 743 | Err(_) => false, |
| 744 | }; |
| 745 | if !fresh { |
| 746 | continue; |
| 747 | } |
| 748 | if let Ok(rel) = path.strip_prefix(root) { |
| 749 | out.push(fmt!("{}", rel.display())); |
| 750 | } |
| 751 | } |
| 752 | } |
| 753 | out.sort(); |
| 754 | out.truncate(SHOTS_MAX); |
| 755 | out |
| 756 | } |
| 757 | |
| 758 | // ┌───────────────────────────────────────────────────────────────┐ |
| 759 | // │ Reading what a verifier said │ |
| 760 | // └───────────────────────────────────────────────────────────────┘ |
| 761 | |
| 762 | /// Every check a run reported, by name, and whether it passed every time it |
| 763 | /// appeared. |
| 764 | /// |
| 765 | /// A `BTreeMap` because two runs are compared and the comparison must not depend |
| 766 | /// on the order the lines arrived in; a name that appears twice folds to the |
| 767 | /// worse of the two, since a check that failed once is not a check that passed. |
| 768 | pub type Checks = BTreeMap<String, bool>; |
| 769 | |
| 770 | /// The checks a run's output reports. |
| 771 | /// |
| 772 | /// The convention is one line per check, and it is the same in all 265 files of |
| 773 | /// `dev/`, because they all copy the same four-line helper: |
| 774 | /// |
| 775 | /// ```text |
| 776 | /// console.log((pass ? ' ok ' : ' FAIL ') + name + (detail ? ' — ' + detail : '')); |
| 777 | /// ``` |
| 778 | /// |
| 779 | /// The detail is cut off before the name is kept. It carries counts, paths and |
| 780 | /// timings that differ between two runs of the same passing check, and comparing |
| 781 | /// it would make every break look as though it had bitten. |
| 782 | /// |
| 783 | /// # Arguments |
| 784 | /// * `text` - Everything the run printed, on both streams. |
| 785 | pub fn parse_checks(text: &str) -> Checks { |
| 786 | let mut out: Checks = BTreeMap::new(); |
| 787 | for line in text.lines() { |
| 788 | let t = line.trim_start(); |
| 789 | let (pass, rest) = if let Some(r) = t.strip_prefix("ok ") { |
| 790 | (true, r) |
| 791 | } else if let Some(r) = t.strip_prefix("FAIL ") { |
| 792 | (false, r) |
| 793 | } else if let Some(r) = t.strip_prefix("PASS ") { |
| 794 | (true, r) |
| 795 | } else { |
| 796 | continue; |
| 797 | }; |
| 798 | let head = match rest.find(" — ") { |
| 799 | Some(i) => &rest[..i], |
| 800 | None => rest, |
| 801 | }; |
| 802 | let name = head.trim(); |
| 803 | if name.is_empty() { |
| 804 | continue; |
| 805 | } |
| 806 | let e = out.entry(fmt!("{}", name)).or_insert(true); |
| 807 | *e = *e && pass; |
| 808 | } |
| 809 | out |
| 810 | } |
| 811 | |
| 812 | /// The text of a run with the parts that legitimately differ taken out. |
| 813 | /// |
| 814 | /// Used for one detail only -- saying that a break's output was *identical* to |
| 815 | /// the clean run's, which is the most damning thing that can be said about an |
| 816 | /// instrument. A verifier usually announces its own break in a banner, so a raw |
| 817 | /// comparison would never find two runs identical and the worst case would never |
| 818 | /// be reported. Lines mentioning the flag are therefore dropped, along with |
| 819 | /// blank lines and trailing space. |
| 820 | /// |
| 821 | /// # Arguments |
| 822 | /// * `text` - Everything the run printed. |
| 823 | pub fn normalise(text: &str) -> String { |
| 824 | let mut out = String::with_capacity(text.len()); |
| 825 | for line in text.lines() { |
| 826 | if line.contains("--break") || line.contains("BREAK") || line.contains("break '") { |
| 827 | continue; |
| 828 | } |
| 829 | let t = line.trim_end(); |
| 830 | if t.is_empty() { |
| 831 | continue; |
| 832 | } |
| 833 | out.push_str(t); |
| 834 | out.push('\n'); |
| 835 | } |
| 836 | out |
| 837 | } |
| 838 | |
| 839 | /// What one run of a verifier did. |
| 840 | #[derive(Clone, Debug)] |
| 841 | pub struct Pass { |
| 842 | /// `clean`, or the break's name. |
| 843 | pub label: String, |
| 844 | /// The exit status, or -1 where there was none. |
| 845 | pub exit: i32, |
| 846 | /// Whether the budget killed it. |
| 847 | pub timed: bool, |
| 848 | /// The checks it reported. |
| 849 | pub checks: Checks, |
| 850 | /// Its output with the varying parts removed, for the identity comparison. |
| 851 | pub norm: String, |
| 852 | /// Up to a few failing lines, verbatim, for the report. |
| 853 | pub fails: Vec<String>, |
| 854 | /// How long it took, in milliseconds. |
| 855 | pub ms: u64, |
| 856 | } |
| 857 | |
| 858 | impl Pass { |
| 859 | /// How many checks passed. |
| 860 | pub fn passed(&self) -> usize { |
| 861 | self.checks.values().filter(|v| **v).count() |
| 862 | } |
| 863 | |
| 864 | /// How many checks failed. |
| 865 | pub fn failed(&self) -> usize { |
| 866 | self.checks.values().filter(|v| !**v).count() |
| 867 | } |
| 868 | } |
| 869 | |
| 870 | /// What a break did to the checks that passed clean. |
| 871 | #[derive(Clone, Debug, Eq, PartialEq)] |
| 872 | pub enum Bite { |
| 873 | /// At least one check that passed clean failed under it. |
| 874 | Red { |
| 875 | /// Which ones, up to a handful. |
| 876 | names: Vec<String>, |
| 877 | }, |
| 878 | /// No check that passed clean failed under it. A lying instrument. |
| 879 | None { |
| 880 | /// Whether its output was the clean run's, banner aside. |
| 881 | same: bool, |
| 882 | /// Checks that passed clean and did not appear at all. |
| 883 | vanished: usize, |
| 884 | }, |
| 885 | /// It never ran, and this is why. |
| 886 | Unrun { |
| 887 | /// The reason, in a phrase. |
| 888 | why: String, |
| 889 | }, |
| 890 | } |
| 891 | |
| 892 | /// What a break did, compared with the clean run. |
| 893 | /// |
| 894 | /// **Deliberately the weaker claim.** A break declares in a comment which check |
| 895 | /// it aims at, and a comment is not something to hold a verdict on -- so what is |
| 896 | /// asserted is that *some* check which passed clean now fails. A break that |
| 897 | /// reddens a different check than it meant to is still an instrument that has |
| 898 | /// been seen to move; a break that reddens nothing is not. |
| 899 | /// |
| 900 | /// # Arguments |
| 901 | /// * `clean` - The clean run. |
| 902 | /// * `brk` - The run under the break. |
| 903 | pub fn bite(clean: &Pass, brk: &Pass) -> Bite { |
| 904 | let mut newly: Vec<String> = Vec::new(); |
| 905 | let mut vanished: usize = 0; |
| 906 | for (name, ok) in clean.checks.iter() { |
| 907 | if !*ok { |
| 908 | continue; // it was already failing; it can say nothing about this break |
| 909 | } |
| 910 | match brk.checks.get(name) { |
| 911 | Some(true) => (), |
| 912 | Some(false) => newly.push(fmt!("{}", name)), |
| 913 | None => vanished += 1, |
| 914 | } |
| 915 | } |
| 916 | if !newly.is_empty() { |
| 917 | newly.truncate(6); |
| 918 | return Bite::Red { names: newly }; |
| 919 | } |
| 920 | Bite::None { |
| 921 | same: clean.norm == brk.norm, |
| 922 | vanished, |
| 923 | } |
| 924 | } |
| 925 | |
| 926 | // ┌───────────────────────────────────────────────────────────────┐ |
| 927 | // │ The verdict, which cannot be given without its bad half │ |
| 928 | // └───────────────────────────────────────────────────────────────┘ |
| 929 | |
| 930 | /// What a whole sequence proved. |
| 931 | /// |
| 932 | /// **An enum with no field accessors on purpose.** The count of checks that |
| 933 | /// passed is reachable only by matching, and every arm that carries it also |
| 934 | /// carries what is wrong with it -- so there is no expression in this program |
| 935 | /// that yields "27 passed" without yielding, in the same breath, the number of |
| 936 | /// breaks that proved nothing or the fact that none was run. That is the whole |
| 937 | /// design of this type, and a struct with three public fields would not have it. |
| 938 | #[derive(Clone, Debug, Eq, PartialEq)] |
| 939 | pub enum Verdict { |
| 940 | /// The clean run and at least one break were run. |
| 941 | Proven { |
| 942 | /// Checks that passed clean. |
| 943 | passed: usize, |
| 944 | /// Checks that failed clean. |
| 945 | failed: usize, |
| 946 | /// Breaks that reddened a check which passed clean. |
| 947 | red: Vec<String>, |
| 948 | /// Breaks that reddened nothing. **The lying-instrument count.** |
| 949 | dead: Vec<String>, |
| 950 | /// Breaks the budget or the machine never let run. |
| 951 | unrun: Vec<String>, |
| 952 | }, |
| 953 | /// Only the clean run was made, so nothing here is known to be able to fail. |
| 954 | Unproven { |
| 955 | /// Checks that passed clean. |
| 956 | passed: usize, |
| 957 | /// Checks that failed clean. |
| 958 | failed: usize, |
| 959 | /// The breaks the file declares and this run did not use. |
| 960 | declared: Vec<String>, |
| 961 | }, |
| 962 | } |
| 963 | |
| 964 | impl Verdict { |
| 965 | /// The one-line trailer, which is the sentence a model repeats. |
| 966 | /// |
| 967 | /// Both arms are here so that neither can be written without the other being |
| 968 | /// looked at, and the `Unproven` arm names itself in the words the model |
| 969 | /// should carry back: *proves nothing*. |
| 970 | pub fn trailer(&self) -> String { |
| 971 | match self { |
| 972 | Self::Proven { passed, failed, red, dead, unrun } => { |
| 973 | let mut s = fmt!( |
| 974 | "{} {} checks passed, {} failed, {} breaks confirmed red, {} breaks proved \ |
| 975 | nothing", TRAILER, passed, failed, red.len(), dead.len()); |
| 976 | if !unrun.is_empty() { |
| 977 | s.push_str(&fmt!(", {} breaks never ran", unrun.len())); |
| 978 | } |
| 979 | s.push(']'); |
| 980 | s |
| 981 | }, |
| 982 | Self::Unproven { passed, failed, declared } => fmt!( |
| 983 | "{} {} checks passed, {} failed -- NOT PROVEN: no break was run, so no check \ |
| 984 | here has been shown to be able to fail. This is not evidence that anything \ |
| 985 | works. {} declared breaks were skipped; run again without 'clean_only' to prove \ |
| 986 | the instrument.]", |
| 987 | TRAILER, passed, failed, declared.len()), |
| 988 | } |
| 989 | } |
| 990 | |
| 991 | /// The exit status this verdict deserves. |
| 992 | pub fn exit(&self) -> i32 { |
| 993 | match self { |
| 994 | Self::Proven { failed, dead, unrun, .. } => { |
| 995 | if *failed > 0 { |
| 996 | EXIT_FAILED |
| 997 | } else if !dead.is_empty() || !unrun.is_empty() { |
| 998 | EXIT_UNPROVEN |
| 999 | } else { |
| 1000 | EXIT_PROVED |
| 1001 | } |
| 1002 | }, |
| 1003 | Self::Unproven { failed, .. } => match *failed > 0 { |
| 1004 | true => EXIT_FAILED, |
| 1005 | false => EXIT_UNPROVEN, |
| 1006 | }, |
| 1007 | } |
| 1008 | } |
| 1009 | |
| 1010 | /// The paragraph under the trailer that says what to do about it. |
| 1011 | pub fn advice(&self) -> String { |
| 1012 | match self { |
| 1013 | Self::Proven { failed, dead, unrun, .. } => { |
| 1014 | let mut s = String::new(); |
| 1015 | if *failed > 0 { |
| 1016 | s.push_str( |
| 1017 | "Checks failed in the CLEAN run: that is the code, not the instrument. \ |
| 1018 | Read the failing lines above before running anything again.\n"); |
| 1019 | } |
| 1020 | if !dead.is_empty() { |
| 1021 | s.push_str(&fmt!( |
| 1022 | "A break that proved nothing means the check it aims at cannot be made \ |
| 1023 | to fail, so that check is measuring nothing and its pass is worth \ |
| 1024 | nothing. Treat these as unmeasured and say so: {}.\n", |
| 1025 | dead.join(", "))); |
| 1026 | } |
| 1027 | if !unrun.is_empty() { |
| 1028 | s.push_str(&fmt!( |
| 1029 | "These breaks never ran, so this sequence is an incomplete proof: {}.\n", |
| 1030 | unrun.join(", "))); |
| 1031 | } |
| 1032 | if s.is_empty() { |
| 1033 | s.push_str( |
| 1034 | "Every declared break reddened a check that passed clean, so these \ |
| 1035 | checks have each been seen to fail. That is what makes the pass above \ |
| 1036 | evidence rather than an assertion.\n"); |
| 1037 | } |
| 1038 | s |
| 1039 | }, |
| 1040 | Self::Unproven { declared, .. } => { |
| 1041 | let mut s = fmt!( |
| 1042 | "THIS RUN PROVES NOTHING. A check that has never been observed failing is \ |
| 1043 | not a check, and no check here was observed failing. Do not report the \ |
| 1044 | passing count as evidence.\n"); |
| 1045 | if !declared.is_empty() { |
| 1046 | s.push_str(&fmt!( |
| 1047 | "The breaks this verifier declares, and which were not run: {}.\n", |
| 1048 | declared.join(", "))); |
| 1049 | } |
| 1050 | s |
| 1051 | }, |
| 1052 | } |
| 1053 | } |
| 1054 | } |
| 1055 | |
| 1056 | // ┌───────────────────────────────────────────────────────────────┐ |
| 1057 | // │ Which breaks to run │ |
| 1058 | // └───────────────────────────────────────────────────────────────┘ |
| 1059 | |
| 1060 | /// The breaks a sequence will attempt, or the sentence saying why it cannot. |
| 1061 | /// |
| 1062 | /// # Arguments |
| 1063 | /// * `script` - The resolved verifier. |
| 1064 | /// * `want` - What the caller asked for. |
| 1065 | pub fn chosen(script: &Script, want: &crate::wire::Breaks) -> Result<Vec<String>, String> { |
| 1066 | match want { |
| 1067 | crate::wire::Breaks::None => Ok(Vec::new()), |
| 1068 | crate::wire::Breaks::All => Ok(script.breaks.clone()), |
| 1069 | crate::wire::Breaks::One(asked) => { |
| 1070 | if !name_ok(asked) { |
| 1071 | return Err(fmt!( |
| 1072 | "Refused: '{}' is not a break name. A break is lower-case letters, digits \ |
| 1073 | and underscores, and it must be one this verifier declares.", |
| 1074 | trim_for_message(asked))); |
| 1075 | } |
| 1076 | // Matched against the file's own declarations, and the string that goes on to |
| 1077 | // the command line is the DECLARED copy, never the caller's. |
| 1078 | match script.breaks.iter().find(|b| b.as_str() == asked.as_str()) { |
| 1079 | Some(b) => Ok(vec![b.clone()]), |
| 1080 | None => Err(match script.breaks.is_empty() { |
| 1081 | true => fmt!( |
| 1082 | "Refused: dev/{} declares no breaks at all, so '{}' is not one of \ |
| 1083 | them and there is no way to prove any of its checks. Run it without a \ |
| 1084 | break and treat the result as unproven, or give the verifier a break \ |
| 1085 | mode first.", script.file, trim_for_message(asked)), |
| 1086 | false => fmt!( |
| 1087 | "Refused: dev/{} does not declare a break called '{}'. The ones it \ |
| 1088 | declares are: {}. A break has to be one the verifier itself knows how \ |
| 1089 | to apply; naming any other string would run the file unchanged and \ |
| 1090 | report a pass that means nothing.", |
| 1091 | script.file, trim_for_message(asked), script.breaks.join(", ")), |
| 1092 | }), |
| 1093 | } |
| 1094 | }, |
| 1095 | } |
| 1096 | } |
| 1097 | |
| 1098 | // ┌───────────────────────────────────────────────────────────────┐ |
| 1099 | // │ The record │ |
| 1100 | // └───────────────────────────────────────────────────────────────┘ |
| 1101 | |
| 1102 | /// The journal, and whether it is still being written. |
| 1103 | /// |
| 1104 | /// A handle rather than the whole dispatcher, so the sequence can be run from a |
| 1105 | /// task of its own without carrying `main.rs`'s furniture into this module. |
| 1106 | #[derive(Clone)] |
| 1107 | pub struct Ledger { |
| 1108 | jr: Arc<Mutex<Journal>>, |
| 1109 | sound: Arc<AtomicBool>, |
| 1110 | } |
| 1111 | |
| 1112 | impl Ledger { |
| 1113 | /// # Arguments |
| 1114 | /// * `jr` - The record. |
| 1115 | /// * `sound` - Whether it is still being written. |
| 1116 | pub fn new(jr: Arc<Mutex<Journal>>, sound: Arc<AtomicBool>) -> Self { |
| 1117 | Self { jr, sound } |
| 1118 | } |
| 1119 | |
| 1120 | /// Writes one event, and remembers a failure exactly as the dispatcher does. |
| 1121 | pub fn record(&self, ev: &Event) -> Outcome<()> { |
| 1122 | let done = { |
| 1123 | let mut g = lock_mutex!(self.jr); |
| 1124 | g.append(ev) |
| 1125 | }; |
| 1126 | match done { |
| 1127 | Ok(_) => Ok(()), |
| 1128 | Err(e) => { |
| 1129 | self.sound.store(false, Ordering::SeqCst); |
| 1130 | Err(e) |
| 1131 | }, |
| 1132 | } |
| 1133 | } |
| 1134 | |
| 1135 | /// Is the record still sound? |
| 1136 | pub fn sound(&self) -> bool { |
| 1137 | self.sound.load(Ordering::SeqCst) |
| 1138 | } |
| 1139 | } |
| 1140 | |
| 1141 | // ┌───────────────────────────────────────────────────────────────┐ |
| 1142 | // │ Running the sequence │ |
| 1143 | // └───────────────────────────────────────────────────────────────┘ |
| 1144 | |
| 1145 | /// Everything one sequence needs. |
| 1146 | pub struct Job { |
| 1147 | /// The caller's identifier, echoed on every response. |
| 1148 | pub id: String, |
| 1149 | /// The granted root, which is where the verifier runs. |
| 1150 | pub root: PathBuf, |
| 1151 | /// The verifier. |
| 1152 | pub script: Script, |
| 1153 | /// The breaks to attempt, already checked against the file's declarations. |
| 1154 | pub breaks: Vec<String>, |
| 1155 | /// Node, found on the hand's own `PATH`. |
| 1156 | pub node: PathBuf, |
| 1157 | /// The whole sequence's wall-clock budget. |
| 1158 | pub budget: Duration, |
| 1159 | /// The record. |
| 1160 | pub ledger: Ledger, |
| 1161 | } |
| 1162 | |
| 1163 | /// The argument vector one run takes. |
| 1164 | /// |
| 1165 | /// Three elements at most, and every one of them is either a constant of this |
| 1166 | /// program, a path built from a directory entry, or a break name parsed out of |
| 1167 | /// the verifier's own source. **There is no shell**: this vector is handed to |
| 1168 | /// `execve` through [`Command`], so a semicolon or a `$(…)` in any element would |
| 1169 | /// be an argument and not a command. A model's string is not in it at all. |
| 1170 | /// |
| 1171 | /// # Arguments |
| 1172 | /// * `job` - The sequence. |
| 1173 | /// * `brk` - The break, or `None` for the clean run. |
| 1174 | pub fn argv_for(job: &Job, brk: Option<&str>) -> Vec<String> { |
| 1175 | let mut v = vec![ |
| 1176 | fmt!("{}", job.node.display()), |
| 1177 | fmt!("{}", job.script.path.display()), |
| 1178 | ]; |
| 1179 | if let Some(b) = brk { |
| 1180 | v.push(fmt!("--break")); |
| 1181 | v.push(fmt!("{}", b)); |
| 1182 | } |
| 1183 | v |
| 1184 | } |
| 1185 | |
| 1186 | /// Reads a stream to its end, keeping at most `cap` bytes. |
| 1187 | /// |
| 1188 | /// The rest is drained rather than left, because a child whose pipe is full |
| 1189 | /// stops running and a run that never ends is a budget spent on nothing. |
| 1190 | async fn drain<R>(mut r: R, cap: usize) -> (Vec<u8>, u64) |
| 1191 | where |
| 1192 | R: tokio::io::AsyncRead + Unpin, |
| 1193 | { |
| 1194 | let mut kept = Vec::new(); |
| 1195 | let mut total = 0u64; |
| 1196 | let mut buf = [0u8; 32 * 1024]; |
| 1197 | loop { |
| 1198 | let n = match r.read(&mut buf).await { |
| 1199 | Ok(0) => break, |
| 1200 | Ok(n) => n, |
| 1201 | Err(_) => break, |
| 1202 | }; |
| 1203 | total = total.saturating_add(n as u64); |
| 1204 | if kept.len() < cap { |
| 1205 | let room = cap - kept.len(); |
| 1206 | let take = room.min(n); |
| 1207 | kept.extend_from_slice(&buf[..take]); |
| 1208 | } |
| 1209 | } |
| 1210 | (kept, total) |
| 1211 | } |
| 1212 | |
| 1213 | /// Runs the verifier once, clean or under one break. |
| 1214 | /// |
| 1215 | /// # Arguments |
| 1216 | /// * `job` - The sequence. |
| 1217 | /// * `brk` - The break, or `None`. |
| 1218 | /// * `left` - What is left of the budget. |
| 1219 | /// * `announce` - Where to send [`Resp::Started`], for the first run only. |
| 1220 | async fn once( |
| 1221 | job: &Job, |
| 1222 | brk: Option<&str>, |
| 1223 | left: Duration, |
| 1224 | announce: Option<&Sender<Resp>>, |
| 1225 | ) |
| 1226 | -> Result<Pass, String> |
| 1227 | { |
| 1228 | let label = match brk { |
| 1229 | Some(b) => fmt!("{}", b), |
| 1230 | None => fmt!("clean"), |
| 1231 | }; |
| 1232 | let argv = argv_for(job, brk); |
| 1233 | // The record is written before the process exists, exactly as a command's is, and with the |
| 1234 | // REAL argument vector -- so a reader of the journal sees the node invocation rather than a |
| 1235 | // verb they would have to trust this module's account of. |
| 1236 | if !job.ledger.sound() { |
| 1237 | return Err(fmt!("the hand's journal cannot be written")); |
| 1238 | } |
| 1239 | let ev = Event::from_req(&Req::Exec { |
| 1240 | id: fmt!("{}#{}", job.id, label), |
| 1241 | argv: argv.clone(), |
| 1242 | cwd: fmt!("{}", job.root.display()), |
| 1243 | env: Vec::new(), |
| 1244 | stdin: None, |
| 1245 | timeout_ms: left.as_millis() as u64, |
| 1246 | capture: Capture::Both, |
| 1247 | // What the process may touch, as intent. `mechs` below is what was actually in force, |
| 1248 | // and it says `fence:none` -- because this verb runs a TRACKED SCRIPT outside the |
| 1249 | // command fence on purpose, and a record that hid that would be worse than no record. |
| 1250 | fence: FenceSpec { |
| 1251 | rw: vec![fmt!("{}", job.root.display())], |
| 1252 | ro: Vec::new(), |
| 1253 | deny: Vec::new(), |
| 1254 | net: true, |
| 1255 | }, |
| 1256 | toolkits: Vec::new(), |
| 1257 | }, &[ |
| 1258 | fmt!("fence:none"), |
| 1259 | fmt!("verify:tracked-script"), |
| 1260 | fmt!("net:open"), |
| 1261 | ]); |
| 1262 | if let Some(ev) = ev { |
| 1263 | if let Err(e) = job.ledger.record(&ev) { |
| 1264 | return Err(fmt!("it could not be written to the journal ({})", e.msgs().join(" "))); |
| 1265 | } |
| 1266 | } |
| 1267 | |
| 1268 | // Asked again, as late as anything on this side can be. The sequence has been |
| 1269 | // running for minutes by the time a late break starts, and a command an earlier turn |
| 1270 | // left in the background can rewrite a file in that time. |
| 1271 | if let Some(no) = unfenced_refusal(&job.root, &job.script.file) { |
| 1272 | // The sentence opens with "Refused:" and every caller of `once` puts its own |
| 1273 | // word in front of what comes back from here, so the word is said once. |
| 1274 | let said = match no.strip_prefix("Refused: ") { |
| 1275 | Some(t) => t, |
| 1276 | None => no.as_str(), |
| 1277 | }; |
| 1278 | return Err(fmt!( |
| 1279 | "it stopped being the commit's while the sequence was running. {}", said)); |
| 1280 | } |
| 1281 | |
| 1282 | let started = std::time::Instant::now(); |
| 1283 | let mut cmd = Command::new(&job.node); |
| 1284 | for a in argv.iter().skip(1) { |
| 1285 | cmd.arg(a); |
| 1286 | } |
| 1287 | cmd.current_dir(&job.root); |
| 1288 | // Cleared and rebuilt rather than trimmed, because `Command` has no way to |
| 1289 | // ask what it inherited: the map is the hand's own environment with the seat |
| 1290 | // taken out and the mark put in, which is the same thing said once. |
| 1291 | cmd.env_clear(); |
| 1292 | cmd.envs(unattended_env(&own_env())); |
| 1293 | cmd.stdin(Stdio::null()); |
| 1294 | cmd.stdout(Stdio::piped()); |
| 1295 | cmd.stderr(Stdio::piped()); |
| 1296 | cmd.kill_on_drop(true); |
| 1297 | let mut child = match cmd.spawn() { |
| 1298 | Ok(c) => c, |
| 1299 | Err(e) => return Err(fmt!("node could not be started ({})", e)), |
| 1300 | }; |
| 1301 | // The page starts a 30-second clock at the request and stops it at `Started`, so the |
| 1302 | // first child's real process id goes out the moment there is one. |
| 1303 | if let Some(tx) = announce { |
| 1304 | let _ = tx.send(Resp::Started { |
| 1305 | id: fmt!("{}", job.id), |
| 1306 | pid: child.id().unwrap_or(0), |
| 1307 | }).await; |
| 1308 | } |
| 1309 | let out_pipe = child.stdout.take(); |
| 1310 | let err_pipe = child.stderr.take(); |
| 1311 | // BOTH STREAMS AT ONCE, and it has to be. A pipe holds 64 KB; a child that fills stderr |
| 1312 | // while this side is still waiting for stdout to reach its end stops running, and the |
| 1313 | // deadlock lasts until the budget kills it. Every verifier in dev/ writes to both. |
| 1314 | let reading = async { |
| 1315 | let o = async { |
| 1316 | match out_pipe { |
| 1317 | Some(p) => drain(p, OUT_MAX).await, |
| 1318 | None => (Vec::new(), 0), |
| 1319 | } |
| 1320 | }; |
| 1321 | let e = async { |
| 1322 | match err_pipe { |
| 1323 | Some(p) => drain(p, OUT_MAX).await, |
| 1324 | None => (Vec::new(), 0), |
| 1325 | } |
| 1326 | }; |
| 1327 | tokio::join!(o, e) |
| 1328 | }; |
| 1329 | let waiting = async { |
| 1330 | let status = child.wait().await; |
| 1331 | status |
| 1332 | }; |
| 1333 | let both = async { tokio::join!(waiting, reading) }; |
| 1334 | let (status, (out, err), timed) = match tokio::time::timeout(left, both).await { |
| 1335 | Ok((s, pair)) => (s, pair, false), |
| 1336 | Err(_) => { |
| 1337 | // The budget is up. The child is asked to go; anything it started of its own |
| 1338 | // (a browser, a server) is not this module's to find, and the report says so. |
| 1339 | (Err(std::io::Error::new(std::io::ErrorKind::TimedOut, "budget spent")), |
| 1340 | ((Vec::new(), 0u64), (Vec::new(), 0u64)), |
| 1341 | true) |
| 1342 | }, |
| 1343 | }; |
| 1344 | let exit = match &status { |
| 1345 | Ok(s) => s.code().unwrap_or(-1), |
| 1346 | Err(_) => -1, |
| 1347 | }; |
| 1348 | let text = fmt!("{}{}", |
| 1349 | String::from_utf8_lossy(&out.0), |
| 1350 | String::from_utf8_lossy(&err.0)); |
| 1351 | let checks = parse_checks(&text); |
| 1352 | let mut fails: Vec<String> = text.lines() |
| 1353 | .filter(|l| l.trim_start().starts_with("FAIL ")) |
| 1354 | .map(|l| fmt!("{}", l.trim_end())) |
| 1355 | .collect(); |
| 1356 | fails.truncate(12); |
| 1357 | let ms = started.elapsed().as_millis() as u64; |
| 1358 | // The ending is recorded as a command's is. |
| 1359 | let ended = Event::Ended { |
| 1360 | id: fmt!("{}#{}", job.id, label), |
| 1361 | exit, |
| 1362 | timed_out: timed, |
| 1363 | killed: false, |
| 1364 | out_bytes: out.1, |
| 1365 | err_bytes: err.1, |
| 1366 | }; |
| 1367 | if let Err(e) = job.ledger.record(&ended) { |
| 1368 | eprintln!("daimond-hand: a verify run's ending was not journalled: {}", e); |
| 1369 | } |
| 1370 | if timed { |
| 1371 | return Err(fmt!("it ran past the budget and was killed")); |
| 1372 | } |
| 1373 | Ok(Pass { |
| 1374 | label, |
| 1375 | exit, |
| 1376 | timed, |
| 1377 | norm: normalise(&text), |
| 1378 | checks, |
| 1379 | fails, |
| 1380 | ms, |
| 1381 | }) |
| 1382 | } |
| 1383 | |
| 1384 | /// Runs the clean pass and every chosen break, and answers with all three |
| 1385 | /// numbers. |
| 1386 | /// |
| 1387 | /// **There is no path out of this function that yields the passing count |
| 1388 | /// alone.** It returns a [`Verdict`], whose arms each carry the bad half, and |
| 1389 | /// the report below is composed from that one value. |
| 1390 | /// |
| 1391 | /// # Arguments |
| 1392 | /// * `job` - The sequence. |
| 1393 | /// * `tx` - Where progress and the report are sent. |
| 1394 | pub async fn conduct(job: Job, tx: Sender<Resp>) -> Outcome<()> { |
| 1395 | // BEFORE THE JOURNAL AND BEFORE THE FIRST PROCESS. `REVIEW.md` §1.21: everything |
| 1396 | // below this line runs a script with no fence on it at all, and the only thing that |
| 1397 | // makes that defensible is that the script is the commit's. Asked here rather than |
| 1398 | // in `main.rs` because this function is the crate's public way in, and a gate in the |
| 1399 | // dispatcher is a gate the reproduction walked straight past. |
| 1400 | if let Some(no) = unfenced_refusal(&job.root, &job.script.file) { |
| 1401 | let _ = tx.send(Resp::Refused { id: fmt!("{}", job.id), reason: no }).await; |
| 1402 | return Ok(()); |
| 1403 | } |
| 1404 | let mut left = job.budget; |
| 1405 | let mut seq_err = 0u64; |
| 1406 | let began = std::time::SystemTime::now(); |
| 1407 | |
| 1408 | let say_err = |line: String, seq: &mut u64| { |
| 1409 | let n = *seq; |
| 1410 | *seq = seq.saturating_add(1); |
| 1411 | Resp::Chunk { |
| 1412 | id: fmt!("{}", job.id), |
| 1413 | stream: Stream::Err, |
| 1414 | seq: n, |
| 1415 | data: line, |
| 1416 | } |
| 1417 | }; |
| 1418 | |
| 1419 | // The clean run first, always. Its checks are the baseline every break is compared with, |
| 1420 | // so a sequence that could not make it has nothing to compare anything against. |
| 1421 | let note = say_err(fmt!(" .. clean run of dev/{}\n", job.script.file), &mut seq_err); |
| 1422 | let _ = tx.send(note).await; |
| 1423 | let clean = match once(&job, None, left, Some(&tx)).await { |
| 1424 | Ok(p) => p, |
| 1425 | Err(w) => { |
| 1426 | let _ = tx.send(Resp::Refused { |
| 1427 | id: fmt!("{}", job.id), |
| 1428 | reason: fmt!( |
| 1429 | "Refused: the clean run of dev/{} did not happen -- {}. Nothing was \ |
| 1430 | measured, so there is no result to report.", job.script.file, w), |
| 1431 | }).await; |
| 1432 | return Ok(()); |
| 1433 | }, |
| 1434 | }; |
| 1435 | left = left.saturating_sub(Duration::from_millis(clean.ms)); |
| 1436 | |
| 1437 | let mut passes: Vec<(Pass, Option<Bite>)> = vec![(clean.clone(), None)]; |
| 1438 | let mut red: Vec<String> = Vec::new(); |
| 1439 | let mut dead: Vec<String> = Vec::new(); |
| 1440 | let mut unrun: Vec<String> = Vec::new(); |
| 1441 | |
| 1442 | for b in job.breaks.iter() { |
| 1443 | if left < Duration::from_millis(RUN_FLOOR_MS) { |
| 1444 | unrun.push(fmt!("{}", b)); |
| 1445 | continue; |
| 1446 | } |
| 1447 | let note = say_err(fmt!(" .. --break {}\n", b), &mut seq_err); |
| 1448 | let _ = tx.send(note).await; |
| 1449 | match once(&job, Some(b.as_str()), left, None).await { |
| 1450 | Ok(p) => { |
| 1451 | left = left.saturating_sub(Duration::from_millis(p.ms)); |
| 1452 | let what = bite(&clean, &p); |
| 1453 | match &what { |
| 1454 | Bite::Red { .. } => red.push(fmt!("{}", b)), |
| 1455 | Bite::None { .. } => dead.push(fmt!("{}", b)), |
| 1456 | Bite::Unrun { .. } => unrun.push(fmt!("{}", b)), |
| 1457 | } |
| 1458 | passes.push((p, Some(what))); |
| 1459 | }, |
| 1460 | Err(w) => { |
| 1461 | unrun.push(fmt!("{}", b)); |
| 1462 | passes.push((Pass { |
| 1463 | label: fmt!("{}", b), |
| 1464 | exit: -1, |
| 1465 | timed: false, |
| 1466 | checks: Checks::new(), |
| 1467 | norm: String::new(), |
| 1468 | fails: Vec::new(), |
| 1469 | ms: 0, |
| 1470 | }, Some(Bite::Unrun { why: w }))); |
| 1471 | }, |
| 1472 | } |
| 1473 | } |
| 1474 | |
| 1475 | // The one place a verdict is made, and it is made from what actually ran. |
| 1476 | let verdict = match job.breaks.is_empty() { |
| 1477 | true => Verdict::Unproven { |
| 1478 | passed: clean.passed(), |
| 1479 | failed: clean.failed(), |
| 1480 | declared: job.script.breaks.clone(), |
| 1481 | }, |
| 1482 | false => Verdict::Proven { |
| 1483 | passed: clean.passed(), |
| 1484 | failed: clean.failed(), |
| 1485 | red, |
| 1486 | dead, |
| 1487 | unrun, |
| 1488 | }, |
| 1489 | }; |
| 1490 | let shots = shots_since(&job.root, began); |
| 1491 | let text = report(&job.script, &passes, &verdict, &shots); |
| 1492 | let bytes = text.len() as u64; |
| 1493 | let _ = tx.send(Resp::Chunk { |
| 1494 | id: fmt!("{}", job.id), |
| 1495 | stream: Stream::Out, |
| 1496 | seq: 0, |
| 1497 | data: text, |
| 1498 | }).await; |
| 1499 | let _ = tx.send(Resp::Ended { |
| 1500 | id: fmt!("{}", job.id), |
| 1501 | exit: verdict.exit(), |
| 1502 | timed_out: false, |
| 1503 | killed: false, |
| 1504 | out_bytes: bytes, |
| 1505 | err_bytes: seq_err, |
| 1506 | }).await; |
| 1507 | Ok(()) |
| 1508 | } |
| 1509 | |
| 1510 | /// The whole report, which is the only thing the model reads. |
| 1511 | /// |
| 1512 | /// # Arguments |
| 1513 | /// * `script` - The verifier. |
| 1514 | /// * `passes` - The clean run first, then each break with what it did. |
| 1515 | /// * `verdict` - What the sequence proved. |
| 1516 | /// * `shots` - Pictures this sequence wrote, workspace-relative. |
| 1517 | pub fn report( |
| 1518 | script: &Script, |
| 1519 | passes: &[(Pass, Option<Bite>)], |
| 1520 | verdict: &Verdict, |
| 1521 | shots: &[String], |
| 1522 | ) |
| 1523 | -> String |
| 1524 | { |
| 1525 | let mut s = String::new(); |
| 1526 | s.push_str(&fmt!("dev/{} — {} run{}, {}\n", |
| 1527 | script.file, |
| 1528 | passes.len(), |
| 1529 | match passes.len() { 1 => "", _ => "s" }, |
| 1530 | script.prov.phrase())); |
| 1531 | s.push_str( |
| 1532 | "Run OUTSIDE the command fence, because these bytes are the commit's rather than a \ |
| 1533 | command anybody's model wrote -- that was checked against the commit before each run, \ |
| 1534 | and a verifier that differs from it is refused rather than reported. It reaches this \ |
| 1535 | machine as any script you run yourself would.\n\n"); |
| 1536 | for (p, what) in passes.iter() { |
| 1537 | match what { |
| 1538 | None => { |
| 1539 | s.push_str(&fmt!("CLEAN {} passed, {} failed, exit {}, {} ms\n", |
| 1540 | p.passed(), p.failed(), p.exit, p.ms)); |
| 1541 | for f in p.fails.iter() { |
| 1542 | s.push_str(&fmt!(" {}\n", f)); |
| 1543 | } |
| 1544 | }, |
| 1545 | Some(Bite::Red { names }) => { |
| 1546 | s.push_str(&fmt!( |
| 1547 | "BREAK {:<10} {} passed, {} failed, exit {}, {} ms RED: {} went red\n", |
| 1548 | p.label, p.passed(), p.failed(), p.exit, p.ms, names.join("; "))); |
| 1549 | }, |
| 1550 | Some(Bite::None { same, vanished }) => { |
| 1551 | let how = match (*same, *vanished) { |
| 1552 | (true, _) => fmt!("its output was the clean run's, its own banner aside"), |
| 1553 | (_, 0) => fmt!("the output moved, and no check that passed clean failed"), |
| 1554 | (_, n) => fmt!( |
| 1555 | "no check went red; {} checks that passed clean did not appear at all, \ |
| 1556 | so it may have aborted rather than measured", n), |
| 1557 | }; |
| 1558 | s.push_str(&fmt!( |
| 1559 | "BREAK {:<10} {} passed, {} failed, exit {}, {} ms PROVED NOTHING: {}\n", |
| 1560 | p.label, p.passed(), p.failed(), p.exit, p.ms, how)); |
| 1561 | }, |
| 1562 | Some(Bite::Unrun { why }) => { |
| 1563 | s.push_str(&fmt!( |
| 1564 | "BREAK {:<10} NEVER RAN: {}\n", p.label, why)); |
| 1565 | }, |
| 1566 | } |
| 1567 | } |
| 1568 | if !shots.is_empty() { |
| 1569 | s.push_str(&fmt!("\n{} pictures were written; read one with file_read and \"as\":\"image\", \ |
| 1570 | or hand the path to a worker who can see:\n", shots.len())); |
| 1571 | for p in shots.iter() { |
| 1572 | s.push_str(&fmt!(" {}\n", p)); |
| 1573 | } |
| 1574 | } |
| 1575 | s.push('\n'); |
| 1576 | s.push_str(&verdict.advice()); |
| 1577 | s.push('\n'); |
| 1578 | s.push_str(&verdict.trailer()); |
| 1579 | s.push('\n'); |
| 1580 | if s.len() > REPORT_MAX { |
| 1581 | // The trailer is the one line that must survive, since it carries all three numbers |
| 1582 | // and the far end refuses a result without it. |
| 1583 | let tail = fmt!("\n[the report was cut here]\n{}\n", verdict.trailer()); |
| 1584 | let room = REPORT_MAX.saturating_sub(tail.len()); |
| 1585 | let mut cut = room; |
| 1586 | while cut > 0 && !s.is_char_boundary(cut) { |
| 1587 | cut -= 1; |
| 1588 | } |
| 1589 | s.truncate(cut); |
| 1590 | s.push_str(&tail); |
| 1591 | } |
| 1592 | s |
| 1593 | } |
| 1594 | |
| 1595 | #[cfg(test)] |
| 1596 | mod tests { |
| 1597 | use super::*; |
| 1598 | |
| 1599 | use crate::wire::Breaks; |
| 1600 | |
| 1601 | use std::fs; |
| 1602 | |
| 1603 | /// Where the fixtures go. |
| 1604 | /// |
| 1605 | /// Under the home cache and never `/tmp`: that is a tmpfs here, its pages are |
| 1606 | /// charged to whoever wrote them, and filling it has taken this machine down |
| 1607 | /// before. |
| 1608 | /// |
| 1609 | /// # Arguments |
| 1610 | /// * `name` - Unique to the calling test, so tests do not share state. |
| 1611 | fn tree(name: &str) -> Outcome<PathBuf> { |
| 1612 | let home = match std::env::var("HOME") { |
| 1613 | Ok(h) => h, |
| 1614 | Err(e) => return Err(err!( |
| 1615 | "The verify tests need HOME to know where to put fixtures: {}", e; |
| 1616 | Test, Configuration)), |
| 1617 | }; |
| 1618 | let root = PathBuf::from(home).join(".cache/daimond-hand-verify-tests").join(name); |
| 1619 | if root.exists() { |
| 1620 | res!(fs::remove_dir_all(&root).map_err(|e| err!(e, "clearing {:?}", root; Test, IO))); |
| 1621 | } |
| 1622 | res!(fs::create_dir_all(root.join(DEV_DIR)) |
| 1623 | .map_err(|e| err!(e, "making {:?}", root; Test, IO))); |
| 1624 | Ok(root) |
| 1625 | } |
| 1626 | |
| 1627 | /// Lays a verifier down in a fixture tree. |
| 1628 | fn put(root: &Path, name: &str, src: &str) -> Outcome<()> { |
| 1629 | let p = root.join(DEV_DIR).join(fmt!("{}{}{}", PREFIX, name, SUFFIX)); |
| 1630 | res!(fs::write(&p, src).map_err(|e| err!(e, "writing {:?}", p; Test, IO))); |
| 1631 | Ok(()) |
| 1632 | } |
| 1633 | |
| 1634 | // ── The alphabet ──────────────────────────────────────────────── |
| 1635 | |
| 1636 | #[test] |
| 1637 | fn a_name_is_a_name_and_never_a_path() { |
| 1638 | assert!(name_ok("graph")); |
| 1639 | assert!(name_ok("a11y_aria")); |
| 1640 | assert!(name_ok("relay_e2e")); |
| 1641 | // Everything a path, a flag or a shell would be made of. |
| 1642 | for bad in [ |
| 1643 | "", |
| 1644 | "dev/verify_graph.mjs", |
| 1645 | "../etc/passwd", |
| 1646 | "graph.mjs", |
| 1647 | "-rf", |
| 1648 | "graph;rm", |
| 1649 | "graph ", |
| 1650 | "Graph", |
| 1651 | "graph$(id)", |
| 1652 | "graph\nrm", |
| 1653 | ] { |
| 1654 | assert!(!name_ok(bad), "{:?} was accepted as a verifier name", bad); |
| 1655 | } |
| 1656 | assert!(!name_ok(&"a".repeat(NAME_MAX + 1)), "a name past the cap was accepted"); |
| 1657 | } |
| 1658 | |
| 1659 | // ── Nobody is at the keyboard ─────────────────────────────────── |
| 1660 | |
| 1661 | #[test] |
| 1662 | fn a_verifier_is_handed_no_display_however_the_hand_got_one() { |
| 1663 | // The hand as the browser starts it: a native messaging host inherits the |
| 1664 | // browser's environment, and the browser is on the owner's screen. |
| 1665 | let mine: BTreeMap<String, String> = [ |
| 1666 | ("DISPLAY", ":0"), |
| 1667 | ("WAYLAND_DISPLAY", "wayland-0"), |
| 1668 | ("XDG_SESSION_TYPE", "wayland"), |
| 1669 | ("HOME", "/home/jason"), |
| 1670 | ("PATH", "/usr/bin"), |
| 1671 | ].iter().map(|(k, v)| (fmt!("{}", k), fmt!("{}", v))).collect(); |
| 1672 | let out = unattended_env(&mine); |
| 1673 | for v in SEAT_VARS { |
| 1674 | assert!(!out.contains_key(*v), |
| 1675 | "{} reached a verifier, so a headed one could paint on the seat", v); |
| 1676 | } |
| 1677 | assert_eq!(Some(&fmt!("1")), out.get(UNATTENDED_VAR), |
| 1678 | "the run was not marked unattended, so dev/display.mjs would allow :0"); |
| 1679 | // Everything else is untouched: a verifier with no HOME dies on the first |
| 1680 | // line of any script in dev/, which is B13's neighbour in the same file. |
| 1681 | assert_eq!(Some(&fmt!("/home/jason")), out.get("HOME")); |
| 1682 | assert_eq!(Some(&fmt!("/usr/bin")), out.get("PATH")); |
| 1683 | } |
| 1684 | |
| 1685 | #[test] |
| 1686 | fn a_hand_with_no_display_still_says_nobody_is_looking() { |
| 1687 | // The mark is not conditional on there having been something to strip. An |
| 1688 | // absent DISPLAY and an unattended run are different sentences in |
| 1689 | // `dev/display.mjs`, and only the second one names the seat. |
| 1690 | let mine: BTreeMap<String, String> = |
| 1691 | [(fmt!("HOME"), fmt!("/home/jason"))].into_iter().collect(); |
| 1692 | let out = unattended_env(&mine); |
| 1693 | assert_eq!(Some(&fmt!("1")), out.get(UNATTENDED_VAR)); |
| 1694 | assert_eq!(2, out.len(), "something else was added: {:?}", out); |
| 1695 | } |
| 1696 | |
| 1697 | #[test] |
| 1698 | fn a_verifier_cannot_ask_to_be_watched() { |
| 1699 | // The mark WINS over an inherited one. Nothing sets this name today, and a |
| 1700 | // day when something does is a day when the last word has to be the hand's. |
| 1701 | let mine: BTreeMap<String, String> = [ |
| 1702 | (fmt!("{}", UNATTENDED_VAR), fmt!("")), |
| 1703 | (fmt!("DISPLAY"), fmt!(":0")), |
| 1704 | ].into_iter().collect(); |
| 1705 | let out = unattended_env(&mine); |
| 1706 | assert_eq!(Some(&fmt!("1")), out.get(UNATTENDED_VAR), |
| 1707 | "an inherited empty mark survived, and an empty mark is read as absent"); |
| 1708 | assert!(!out.contains_key("DISPLAY")); |
| 1709 | } |
| 1710 | |
| 1711 | // ── What a verifier declares ──────────────────────────────────── |
| 1712 | |
| 1713 | #[test] |
| 1714 | fn the_breaks_come_out_of_the_files_own_source() { |
| 1715 | let src = "\ |
| 1716 | // EACH CHECK IS PROVED AGAINST BROKEN CODE FIRST. `--break <name>` serves a |
| 1717 | // node dev/verify_about.mjs --break nobutton # 1: no About button |
| 1718 | // node dev/verify_about.mjs --break oldbadge # 1: the badge is still there |
| 1719 | // node dev/verify_about.mjs --break=stretch # 3: the splash is squashed |
| 1720 | const BREAK = process.argv.indexOf('--break'); |
| 1721 | if (BREAK) console.log(`running with --break ${BREAK}`); |
| 1722 | "; |
| 1723 | assert_eq!(vec![fmt!("nobutton"), fmt!("oldbadge"), fmt!("stretch")], declared_breaks(src), |
| 1724 | "the placeholder, the quoted flag and the interpolation must not be taken as names"); |
| 1725 | } |
| 1726 | |
| 1727 | #[test] |
| 1728 | fn a_file_that_declares_nothing_declares_nothing() { |
| 1729 | assert!(declared_breaks("console.log(' ok nothing');").is_empty()); |
| 1730 | // The flag mentioned but never given a value. |
| 1731 | assert!(declared_breaks("// node dev/verify_x.mjs --break <name>\n").is_empty()); |
| 1732 | } |
| 1733 | |
| 1734 | // ── Looking one up ────────────────────────────────────────────── |
| 1735 | |
| 1736 | #[test] |
| 1737 | fn a_name_that_is_not_there_is_refused_with_the_near_misses() -> Outcome<()> { |
| 1738 | let root = res!(tree("absent")); |
| 1739 | res!(put(&root, "graph", "// node dev/verify_graph.mjs --break nolinks\n")); |
| 1740 | res!(put(&root, "graphedit", "//\n")); |
| 1741 | let e = match resolve(&root, "graphs") { |
| 1742 | Err(e) => e, |
| 1743 | Ok(s) => return Err(err!("'graphs' resolved to {:?}", s.file; Test)), |
| 1744 | }; |
| 1745 | assert!(e.starts_with("Refused:"), "{}", e); |
| 1746 | assert!(e.contains("graph"), "the near misses are not named: {}", e); |
| 1747 | Ok(()) |
| 1748 | } |
| 1749 | |
| 1750 | #[test] |
| 1751 | fn a_path_is_refused_before_the_directory_is_read() -> Outcome<()> { |
| 1752 | let root = res!(tree("path")); |
| 1753 | for bad in ["../../etc/passwd", "dev/verify_graph.mjs", "graph.mjs"] { |
| 1754 | let e = match resolve(&root, bad) { |
| 1755 | Err(e) => e, |
| 1756 | Ok(_) => return Err(err!("{:?} resolved to something", bad; Test)), |
| 1757 | }; |
| 1758 | assert!(e.contains("not a verifier name"), "{}", e); |
| 1759 | } |
| 1760 | Ok(()) |
| 1761 | } |
| 1762 | |
| 1763 | #[test] |
| 1764 | fn what_reaches_the_command_line_came_from_the_directory() -> Outcome<()> { |
| 1765 | let root = res!(tree("argv")); |
| 1766 | res!(put(&root, "graph", "// node dev/verify_graph.mjs --break nolinks\n")); |
| 1767 | let script = match resolve(&root, "graph") { |
| 1768 | Ok(s) => s, |
| 1769 | Err(e) => return Err(err!("{}", e; Test)), |
| 1770 | }; |
| 1771 | let job = Job { |
| 1772 | id: fmt!("v1"), |
| 1773 | root: root.clone(), |
| 1774 | script: script.clone(), |
| 1775 | breaks: vec![fmt!("nolinks")], |
| 1776 | node: PathBuf::from("/usr/bin/node"), |
| 1777 | budget: Duration::from_millis(1000), |
| 1778 | ledger: Ledger::new( |
| 1779 | Arc::new(Mutex::new(res!(crate::journal::Journal::open( |
| 1780 | crate::journal::Cfg::at(root.join("journal")))))), |
| 1781 | Arc::new(AtomicBool::new(true))), |
| 1782 | }; |
| 1783 | let argv = argv_for(&job, Some("nolinks")); |
| 1784 | assert_eq!(4, argv.len(), "{:?}", argv); |
| 1785 | assert_eq!(fmt!("{}", root.join(DEV_DIR).join("verify_graph.mjs").display()), argv[1]); |
| 1786 | assert_eq!(fmt!("--break"), argv[2]); |
| 1787 | assert_eq!(fmt!("nolinks"), argv[3]); |
| 1788 | Ok(()) |
| 1789 | } |
| 1790 | |
| 1791 | #[test] |
| 1792 | fn a_break_the_file_does_not_declare_is_refused_and_the_real_ones_are_listed() |
| 1793 | -> Outcome<()> |
| 1794 | { |
| 1795 | let root = res!(tree("undeclared")); |
| 1796 | res!(put(&root, "graph", |
| 1797 | "// node dev/verify_graph.mjs --break nolinks\n\ |
| 1798 | // node dev/verify_graph.mjs --break stale\n")); |
| 1799 | let script = match resolve(&root, "graph") { |
| 1800 | Ok(s) => s, |
| 1801 | Err(e) => return Err(err!("{}", e; Test)), |
| 1802 | }; |
| 1803 | assert_eq!(vec![fmt!("nolinks"), fmt!("stale")], script.breaks); |
| 1804 | let e = match chosen(&script, &Breaks::One(fmt!("invented"))) { |
| 1805 | Err(e) => e, |
| 1806 | Ok(v) => return Err(err!("an undeclared break was accepted: {:?}", v; Test)), |
| 1807 | }; |
| 1808 | assert!(e.contains("nolinks") && e.contains("stale"), |
| 1809 | "the refusal does not list what it does declare: {}", e); |
| 1810 | // And the declared one is taken, as the file spells it. |
| 1811 | assert_eq!(Ok(vec![fmt!("stale")]), chosen(&script, &Breaks::One(fmt!("stale")))); |
| 1812 | assert_eq!(Ok(vec![fmt!("nolinks"), fmt!("stale")]), chosen(&script, &Breaks::All)); |
| 1813 | assert_eq!(Ok(Vec::new()), chosen(&script, &Breaks::None)); |
| 1814 | Ok(()) |
| 1815 | } |
| 1816 | |
| 1817 | #[test] |
| 1818 | fn a_verifier_with_no_breaks_cannot_be_asked_for_one() -> Outcome<()> { |
| 1819 | let root = res!(tree("nobreaks")); |
| 1820 | res!(put(&root, "bare", "console.log(' ok something');\n")); |
| 1821 | let script = match resolve(&root, "bare") { |
| 1822 | Ok(s) => s, |
| 1823 | Err(e) => return Err(err!("{}", e; Test)), |
| 1824 | }; |
| 1825 | let e = match chosen(&script, &Breaks::One(fmt!("anything"))) { |
| 1826 | Err(e) => e, |
| 1827 | Ok(_) => return Err(err!("a break was accepted from a file with none"; Test)), |
| 1828 | }; |
| 1829 | assert!(e.contains("declares no breaks"), "{}", e); |
| 1830 | Ok(()) |
| 1831 | } |
| 1832 | |
| 1833 | // ── Reading a run ─────────────────────────────────────────────── |
| 1834 | |
| 1835 | #[test] |
| 1836 | fn the_check_lines_are_read_and_the_detail_is_cut_off() { |
| 1837 | let out = "\ |
| 1838 | ok the badge counts — 3 of 3 |
| 1839 | FAIL the links are drawn — 0 found |
| 1840 | ok the badge counts — 4 of 4 |
| 1841 | "; |
| 1842 | let c = parse_checks(out); |
| 1843 | assert_eq!(Some(&true), c.get("the badge counts")); |
| 1844 | assert_eq!(Some(&false), c.get("the links are drawn")); |
| 1845 | assert_eq!(2, c.len(), "{:?}", c); |
| 1846 | } |
| 1847 | |
| 1848 | #[test] |
| 1849 | fn one_failure_beats_a_pass_of_the_same_name() { |
| 1850 | let c = parse_checks(" ok x\n FAIL x\n ok x\n"); |
| 1851 | assert_eq!(Some(&false), c.get("x"), "a name that failed once must not read as passing"); |
| 1852 | } |
| 1853 | |
| 1854 | // ── What a break did ──────────────────────────────────────────── |
| 1855 | |
| 1856 | fn pass_of(label: &str, out: &str) -> Pass { |
| 1857 | Pass { |
| 1858 | label: fmt!("{}", label), |
| 1859 | exit: 0, |
| 1860 | timed: false, |
| 1861 | checks: parse_checks(out), |
| 1862 | norm: normalise(out), |
| 1863 | fails: Vec::new(), |
| 1864 | ms: 1, |
| 1865 | } |
| 1866 | } |
| 1867 | |
| 1868 | #[test] |
| 1869 | fn a_break_that_reddens_a_check_is_red() { |
| 1870 | let clean = pass_of("clean", " ok a\n ok b\n"); |
| 1871 | let brk = pass_of("nolinks", " ok a\n FAIL b\n"); |
| 1872 | assert_eq!(Bite::Red { names: vec![fmt!("b")] }, bite(&clean, &brk)); |
| 1873 | } |
| 1874 | |
| 1875 | #[test] |
| 1876 | fn a_break_whose_output_is_the_clean_runs_proves_nothing_and_says_which() { |
| 1877 | let clean = pass_of("clean", " ok a\n ok b\n"); |
| 1878 | let brk = pass_of("dead", "*** RUNNING UNDER --break dead ***\n ok a\n ok b\n"); |
| 1879 | assert_eq!(Bite::None { same: true, vanished: 0 }, bite(&clean, &brk), |
| 1880 | "the break's own banner must not hide that nothing else changed"); |
| 1881 | } |
| 1882 | |
| 1883 | #[test] |
| 1884 | fn a_break_that_moves_the_output_and_reddens_nothing_still_proves_nothing() { |
| 1885 | let clean = pass_of("clean", " ok a\n ok b\nfound 3 links\n"); |
| 1886 | let brk = pass_of("dead", " ok a\n ok b\nfound 9 links\n"); |
| 1887 | assert_eq!(Bite::None { same: false, vanished: 0 }, bite(&clean, &brk), |
| 1888 | "a changed number is not a reddened check"); |
| 1889 | } |
| 1890 | |
| 1891 | #[test] |
| 1892 | fn a_check_already_failing_clean_says_nothing_about_a_break() { |
| 1893 | let clean = pass_of("clean", " ok a\n FAIL b\n"); |
| 1894 | let brk = pass_of("dead", " ok a\n FAIL b\n"); |
| 1895 | assert_eq!(Bite::None { same: true, vanished: 0 }, bite(&clean, &brk), |
| 1896 | "a break must not be credited with a failure that was already there"); |
| 1897 | } |
| 1898 | |
| 1899 | #[test] |
| 1900 | fn a_break_that_aborts_is_not_counted_red() { |
| 1901 | let clean = pass_of("clean", " ok a\n ok b\n ok c\n"); |
| 1902 | let brk = pass_of("crash", " ok a\n"); |
| 1903 | assert_eq!(Bite::None { same: false, vanished: 2 }, bite(&clean, &brk), |
| 1904 | "a run that stopped early has not shown a check able to fail"); |
| 1905 | } |
| 1906 | |
| 1907 | // ── The verdict, and the number that cannot be dropped ────────── |
| 1908 | |
| 1909 | #[test] |
| 1910 | fn every_proven_trailer_carries_all_three_numbers() { |
| 1911 | let v = Verdict::Proven { |
| 1912 | passed: 27, |
| 1913 | failed: 0, |
| 1914 | red: vec![fmt!("nolinks"), fmt!("stale")], |
| 1915 | dead: vec![fmt!("nosc")], |
| 1916 | unrun: Vec::new(), |
| 1917 | }; |
| 1918 | let t = v.trailer(); |
| 1919 | assert!(t.starts_with(TRAILER), "{}", t); |
| 1920 | assert!(t.contains("27 checks passed"), "{}", t); |
| 1921 | assert!(t.contains("2 breaks confirmed red"), "{}", t); |
| 1922 | assert!(t.contains("1 breaks proved nothing"), "{}", t); |
| 1923 | assert_eq!(EXIT_UNPROVEN, v.exit(), "a dead break is not a clean bill of health"); |
| 1924 | assert!(v.advice().contains("nosc"), "the dead break is not named: {}", v.advice()); |
| 1925 | } |
| 1926 | |
| 1927 | #[test] |
| 1928 | fn a_clean_only_run_says_it_proves_nothing_in_the_words_a_model_repeats() { |
| 1929 | let v = Verdict::Unproven { |
| 1930 | passed: 27, |
| 1931 | failed: 0, |
| 1932 | declared: vec![fmt!("nolinks"), fmt!("stale")], |
| 1933 | }; |
| 1934 | let t = v.trailer(); |
| 1935 | assert!(t.contains("NOT PROVEN"), "{}", t); |
| 1936 | assert!(t.contains("not evidence") || t.contains("NOT PROVEN"), "{}", t); |
| 1937 | assert_eq!(EXIT_UNPROVEN, v.exit()); |
| 1938 | assert!(v.advice().contains("PROVES NOTHING"), "{}", v.advice()); |
| 1939 | assert!(v.advice().contains("nolinks"), "the skipped breaks are not named"); |
| 1940 | } |
| 1941 | |
| 1942 | #[test] |
| 1943 | fn a_sequence_where_every_break_bit_is_the_only_clean_bill() { |
| 1944 | let v = Verdict::Proven { |
| 1945 | passed: 5, |
| 1946 | failed: 0, |
| 1947 | red: vec![fmt!("a"), fmt!("b")], |
| 1948 | dead: Vec::new(), |
| 1949 | unrun: Vec::new(), |
| 1950 | }; |
| 1951 | assert_eq!(EXIT_PROVED, v.exit()); |
| 1952 | assert!(v.trailer().contains("0 breaks proved nothing"), "{}", v.trailer()); |
| 1953 | } |
| 1954 | |
| 1955 | #[test] |
| 1956 | fn a_break_the_budget_never_reached_is_said_and_not_hidden() { |
| 1957 | let v = Verdict::Proven { |
| 1958 | passed: 5, |
| 1959 | failed: 0, |
| 1960 | red: vec![fmt!("a")], |
| 1961 | dead: Vec::new(), |
| 1962 | unrun: vec![fmt!("b")], |
| 1963 | }; |
| 1964 | assert!(v.trailer().contains("1 breaks never ran"), "{}", v.trailer()); |
| 1965 | assert_eq!(EXIT_UNPROVEN, v.exit()); |
| 1966 | } |
| 1967 | |
| 1968 | #[test] |
| 1969 | fn a_failing_clean_run_is_the_code_and_the_report_says_so() { |
| 1970 | let v = Verdict::Proven { |
| 1971 | passed: 20, |
| 1972 | failed: 3, |
| 1973 | red: vec![fmt!("a")], |
| 1974 | dead: Vec::new(), |
| 1975 | unrun: Vec::new(), |
| 1976 | }; |
| 1977 | assert_eq!(EXIT_FAILED, v.exit()); |
| 1978 | assert!(v.advice().contains("CLEAN run"), "{}", v.advice()); |
| 1979 | } |
| 1980 | |
| 1981 | // ── The report ────────────────────────────────────────────────── |
| 1982 | |
| 1983 | #[test] |
| 1984 | fn the_report_always_ends_with_the_trailer() -> Outcome<()> { |
| 1985 | let root = res!(tree("report")); |
| 1986 | res!(put(&root, "graph", "// node dev/verify_graph.mjs --break nolinks\n")); |
| 1987 | let script = match resolve(&root, "graph") { |
| 1988 | Ok(s) => s, |
| 1989 | Err(e) => return Err(err!("{}", e; Test)), |
| 1990 | }; |
| 1991 | let clean = pass_of("clean", " ok a\n ok b\n"); |
| 1992 | let brk = pass_of("nolinks", " ok a\n FAIL b\n"); |
| 1993 | let what = bite(&clean, &brk); |
| 1994 | let v = Verdict::Proven { |
| 1995 | passed: clean.passed(), |
| 1996 | failed: clean.failed(), |
| 1997 | red: vec![fmt!("nolinks")], |
| 1998 | dead: Vec::new(), |
| 1999 | unrun: Vec::new(), |
| 2000 | }; |
| 2001 | let text = report(&script, &[(clean, None), (brk, Some(what))], &v, &[]); |
| 2002 | let last = match text.lines().filter(|l| !l.is_empty()).last() { |
| 2003 | Some(l) => l, |
| 2004 | None => return Err(err!("the report is empty"; Test)), |
| 2005 | }; |
| 2006 | assert!(last.starts_with(TRAILER), "the last line is {:?}", last); |
| 2007 | assert!(text.contains("RED: b went red"), "{}", text); |
| 2008 | assert!(text.contains("OUTSIDE the command fence"), |
| 2009 | "the report does not say where it ran"); |
| 2010 | Ok(()) |
| 2011 | } |
| 2012 | |
| 2013 | #[test] |
| 2014 | fn a_report_too_long_to_send_keeps_its_trailer() -> Outcome<()> { |
| 2015 | let root = res!(tree("long")); |
| 2016 | res!(put(&root, "big", "//\n")); |
| 2017 | let script = match resolve(&root, "big") { |
| 2018 | Ok(s) => s, |
| 2019 | Err(e) => return Err(err!("{}", e; Test)), |
| 2020 | }; |
| 2021 | let mut clean = pass_of("clean", " ok a\n"); |
| 2022 | clean.fails = (0..40_000).map(|i| fmt!("FAIL a very long check name number {}", i)) |
| 2023 | .collect(); |
| 2024 | let v = Verdict::Unproven { passed: 1, failed: 0, declared: Vec::new() }; |
| 2025 | let text = report(&script, &[(clean, None)], &v, &[]); |
| 2026 | assert!(text.len() <= REPORT_MAX, "{} bytes", text.len()); |
| 2027 | assert!(text.trim_end().ends_with(']'), "the trailer was cut off"); |
| 2028 | assert!(text.contains(TRAILER), "the trailer is gone"); |
| 2029 | assert!(text.contains("NOT PROVEN"), "the unproven label was cut off"); |
| 2030 | Ok(()) |
| 2031 | } |
| 2032 | |
| 2033 | #[test] |
| 2034 | fn the_capability_says_whether_this_folder_has_verifiers_at_all() -> Outcome<()> { |
| 2035 | let bare = res!(tree("cap_bare")); |
| 2036 | assert_eq!(fmt!("verify:none"), cap(&bare)); |
| 2037 | res!(put(&bare, "one", "//\n")); |
| 2038 | assert_eq!(fmt!("verify:dev"), cap(&bare)); |
| 2039 | Ok(()) |
| 2040 | } |
| 2041 | |
| 2042 | // ── Whose code is this? ───────────────────────────────────────── |
| 2043 | // |
| 2044 | // `REVIEW.md` §1.21. A verifier runs OUTSIDE the command fence, and the |
| 2045 | // whole of the argument for that is that its bytes came with the checkout. |
| 2046 | // These tests aim at the argument rather than at the code: the fixture |
| 2047 | // verifier's entire body writes a file OUTSIDE the granted root, so the |
| 2048 | // marker's existence is the measurement. A provenance that may run gets |
| 2049 | // its marker; every other provenance must not. |
| 2050 | |
| 2051 | /// Runs one git command in a fixture tree. |
| 2052 | /// |
| 2053 | /// # Arguments |
| 2054 | /// * `root` - The fixture. |
| 2055 | /// * `args` - The argument vector after `git -C <root>`. |
| 2056 | fn git(root: &Path, args: &[&str]) -> Outcome<()> { |
| 2057 | let git = res!(on_path("git").ok_or_else(|| err!( |
| 2058 | "These tests ask git what the repository holds, and there is no git on PATH."; |
| 2059 | Test, Missing))); |
| 2060 | let out = res!(std::process::Command::new(git) |
| 2061 | .arg("-C").arg(root) |
| 2062 | .args(args) |
| 2063 | .output() |
| 2064 | .map_err(|e| err!(e, "running git {:?}", args; Test, IO))); |
| 2065 | if !out.status.success() { |
| 2066 | return Err(err!("git {:?} failed in {:?}: {}", |
| 2067 | args, root, String::from_utf8_lossy(&out.stderr); Test, IO)); |
| 2068 | } |
| 2069 | Ok(()) |
| 2070 | } |
| 2071 | |
| 2072 | /// A verifier whose whole body writes one file, at an absolute path. |
| 2073 | /// |
| 2074 | /// It prints a check line as well, so that a run which is allowed reports |
| 2075 | /// something a `Pass` can be built from and the fixture exercises the real |
| 2076 | /// path rather than an empty one. |
| 2077 | /// |
| 2078 | /// # Arguments |
| 2079 | /// * `marker` - Where it writes, which the tests place outside the root. |
| 2080 | fn escaper(marker: &Path) -> String { |
| 2081 | fmt!( |
| 2082 | "import fs from 'node:fs';\n\ |
| 2083 | fs.writeFileSync({:?}, 'the verifier ran unfenced\\n');\n\ |
| 2084 | console.log(' ok it ran');\n", |
| 2085 | fmt!("{}", marker.display())) |
| 2086 | } |
| 2087 | |
| 2088 | /// What one attempt at a fixture verifier did. |
| 2089 | struct Attempt { |
| 2090 | refused: Option<String>, // the sentence, where the hand declined |
| 2091 | escaped: bool, // whether the marker outside the root appeared |
| 2092 | } |
| 2093 | |
| 2094 | /// Resolves and conducts a fixture verifier, and says what came of it. |
| 2095 | /// |
| 2096 | /// Deliberately the crate's own public API and not the dispatcher's: a gate |
| 2097 | /// that lives in `main.rs` is a gate this function would walk past, and the |
| 2098 | /// reproduction in `REVIEW.md` §1.21 walked past exactly that. |
| 2099 | /// |
| 2100 | /// # Arguments |
| 2101 | /// * `root` - The fixture. |
| 2102 | /// * `name` - The verifier's short name. |
| 2103 | /// * `marker` - The path the verifier writes, outside `root`. |
| 2104 | async fn attempt(root: &Path, name: &str, marker: &Path) -> Outcome<Attempt> { |
| 2105 | if marker.exists() { |
| 2106 | res!(fs::remove_file(marker).map_err(|e| err!(e, "clearing {:?}", marker; Test, IO))); |
| 2107 | } |
| 2108 | let node = res!(on_path("node").ok_or_else(|| err!( |
| 2109 | "Every verifier is a Node script and there is no node on PATH, so this test would \ |
| 2110 | prove nothing about whether one ran."; |
| 2111 | Test, Missing))); |
| 2112 | let script = match resolve(root, name) { |
| 2113 | Ok(s) => s, |
| 2114 | Err(w) => return Ok(Attempt { refused: Some(w), escaped: marker.exists() }), |
| 2115 | }; |
| 2116 | let stem = match root.file_name().and_then(|n| n.to_str()) { |
| 2117 | Some(n) => fmt!("{}", n), |
| 2118 | None => fmt!("{}", name), |
| 2119 | }; |
| 2120 | let jdir = match root.parent() { |
| 2121 | Some(p) => p.join(fmt!("{}-journal", stem)), |
| 2122 | None => root.join("journal"), |
| 2123 | }; |
| 2124 | if jdir.exists() { |
| 2125 | res!(fs::remove_dir_all(&jdir).map_err(|e| err!(e, "clearing {:?}", jdir; Test, IO))); |
| 2126 | } |
| 2127 | let jr = res!(Journal::open(crate::journal::Cfg::at(&jdir))); |
| 2128 | let job = Job { |
| 2129 | id: fmt!("probe"), |
| 2130 | root: root.to_path_buf(), |
| 2131 | script, |
| 2132 | breaks: Vec::new(), |
| 2133 | node, |
| 2134 | budget: Duration::from_millis(60_000), |
| 2135 | ledger: Ledger::new(Arc::new(Mutex::new(jr)), Arc::new(AtomicBool::new(true))), |
| 2136 | }; |
| 2137 | let (tx, mut rx) = tokio::sync::mpsc::channel(64); |
| 2138 | // Boxed, so the whole sequence's future lives on the heap. Inlined, `conduct` |
| 2139 | // holds `once` holds `drain`'s buffers, and a debug build of that overflows a |
| 2140 | // test thread's stack before the first process starts. |
| 2141 | res!(Box::pin(conduct(job, tx)).await); |
| 2142 | let mut refused = None; |
| 2143 | while let Some(r) = rx.recv().await { |
| 2144 | if let Resp::Refused { reason, .. } = r { |
| 2145 | refused = Some(reason); |
| 2146 | } |
| 2147 | } |
| 2148 | Ok(Attempt { refused, escaped: marker.exists() }) |
| 2149 | } |
| 2150 | |
| 2151 | #[tokio::test] |
| 2152 | async fn a_verifier_the_checkout_brought_runs_outside_the_fence() -> Outcome<()> { |
| 2153 | let root = res!(tree("prov_committed")); |
| 2154 | let marker = root.join("..").join("prov_committed.escaped"); |
| 2155 | let marker = PathBuf::from(fmt!("{}", marker.display())); |
| 2156 | res!(git(&root, &["init", "-q"])); |
| 2157 | res!(put(&root, "probe", &escaper(&marker))); |
| 2158 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2159 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2160 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2161 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2162 | assert!(a.refused.is_none(), "the committed case must run: {:?}", a.refused); |
| 2163 | // The half that keeps this pair honest. Without it a guard that refused |
| 2164 | // everything would leave the three tests below green while making the verb |
| 2165 | // useless, which is the failure `REVIEW.md` was written about. |
| 2166 | assert!(a.escaped, |
| 2167 | "a verifier that IS the commit runs outside the fence, and this one wrote nothing \ |
| 2168 | outside the root -- so the three tests below prove nothing about a fence"); |
| 2169 | Ok(()) |
| 2170 | } |
| 2171 | |
| 2172 | /// The verb, over this very tree, driven by hand. |
| 2173 | /// |
| 2174 | /// `#[ignore]`d, so it is never part of a suite: it runs a real verifier from |
| 2175 | /// the repository this crate sits in, which starts browsers and takes minutes. |
| 2176 | /// It is here because the fixtures above cannot find what it finds. Run over |
| 2177 | /// `verify_reflux` on 2026-08-25 it turned up two defects in a file that had |
| 2178 | /// passed standalone twelve times -- a break declared by a sentence that named |
| 2179 | /// ANOTHER file's break, and an X server killed hard enough to keep its lock -- |
| 2180 | /// and neither is visible from a fixture, because a fixture has no header prose |
| 2181 | /// and starts no X server. |
| 2182 | /// |
| 2183 | /// LIVE=reflux cargo test --manifest-path hand/Cargo.toml --lib \ |
| 2184 | /// verify::tests::live_walk -- --ignored --nocapture |
| 2185 | #[tokio::test] |
| 2186 | #[ignore] |
| 2187 | async fn live_walk_over_this_very_tree() -> Outcome<()> { |
| 2188 | let root = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join(".."); |
| 2189 | let name = std::env::var("LIVE").unwrap_or(fmt!("harness")); |
| 2190 | let script = match resolve(&root, &name) { |
| 2191 | Ok(s) => s, |
| 2192 | Err(w) => { println!("REFUSED: {}", w); return Ok(()); }, |
| 2193 | }; |
| 2194 | println!("script {:?}\nprov {}\nbreaks {:?}", |
| 2195 | script.path, script.prov.phrase(), script.breaks); |
| 2196 | let node = res!(on_path("node").ok_or_else(|| err!( |
| 2197 | "Every verifier is a Node script and there is no node on PATH."; Test, Missing))); |
| 2198 | let jdir = res!(tree("live_walk")).join("journal"); |
| 2199 | let jr = res!(Journal::open(crate::journal::Cfg::at(&jdir))); |
| 2200 | let job = Job { |
| 2201 | id: fmt!("live"), root: root.clone(), breaks: script.breaks.clone(), script, |
| 2202 | node, budget: Duration::from_millis(1_800_000), |
| 2203 | ledger: Ledger::new(Arc::new(Mutex::new(jr)), Arc::new(AtomicBool::new(true))), |
| 2204 | }; |
| 2205 | let (tx, mut rx) = tokio::sync::mpsc::channel(64); |
| 2206 | res!(Box::pin(conduct(job, tx)).await); |
| 2207 | while let Some(r) = rx.recv().await { |
| 2208 | match r { |
| 2209 | Resp::Chunk { data, .. } => println!("{}", data), |
| 2210 | Resp::Refused { reason, .. } => println!("REFUSED: {}", reason), |
| 2211 | Resp::Ended { exit, .. } => println!("exit {}", exit), |
| 2212 | _ => {}, |
| 2213 | } |
| 2214 | } |
| 2215 | Ok(()) |
| 2216 | } |
| 2217 | |
| 2218 | #[tokio::test] |
| 2219 | async fn the_spawn_itself_hands_on_no_display() -> Outcome<()> { |
| 2220 | // The pure function above can be right and never called, which is this |
| 2221 | // repository's own recorded failure shape -- `dev/BLOCKERS.md` keeps a list |
| 2222 | // of things built and unreachable. So this one goes through `conduct`, |
| 2223 | // `once` and a real `execve`, and asks the child what it was actually given. |
| 2224 | let root = res!(tree("prov_seat")); |
| 2225 | let marker = root.join("..").join("prov_seat.saw"); |
| 2226 | let marker = PathBuf::from(fmt!("{}", marker.display())); |
| 2227 | res!(git(&root, &["init", "-q"])); |
| 2228 | res!(put(&root, "probe", &fmt!( |
| 2229 | "import fs from 'node:fs';\n\ |
| 2230 | const e = process.env;\n\ |
| 2231 | fs.writeFileSync({:?}, `DISPLAY=${{e.DISPLAY ?? '<unset>'}} \ |
| 2232 | WAYLAND_DISPLAY=${{e.WAYLAND_DISPLAY ?? '<unset>'}} \ |
| 2233 | DAIMOND_UNATTENDED=${{e.DAIMOND_UNATTENDED ?? '<unset>'}} \ |
| 2234 | HOME=${{e.HOME ? 'set' : '<unset>'}}`);\n\ |
| 2235 | console.log(' ok it ran');\n", |
| 2236 | fmt!("{}", marker.display())))); |
| 2237 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2238 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2239 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2240 | let mine = std::env::var("DISPLAY").unwrap_or_default(); |
| 2241 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2242 | assert!(a.refused.is_none(), "the committed case must run: {:?}", a.refused); |
| 2243 | let saw = res!(fs::read_to_string(&marker) |
| 2244 | .map_err(|e| err!(e, "the probe wrote nothing at {:?}", marker; Test, IO))); |
| 2245 | assert!(saw.contains("DISPLAY=<unset>"), |
| 2246 | "a verifier was handed a display. This process holds DISPLAY={:?}, and what \ |
| 2247 | reached the child was: {}", mine, saw); |
| 2248 | assert!(saw.contains("WAYLAND_DISPLAY=<unset>"), |
| 2249 | "a verifier was handed the compositor, which Chromium prefers over DISPLAY \ |
| 2250 | and which is the owner's own screen: {}", saw); |
| 2251 | assert!(saw.contains("DAIMOND_UNATTENDED=1"), |
| 2252 | "the run was not marked unattended, so dev/display.mjs would allow :0: {}", saw); |
| 2253 | // The half that keeps the three above honest: an environment cleared and not |
| 2254 | // rebuilt would satisfy every one of them and break every verifier in dev/. |
| 2255 | assert!(saw.contains("HOME=set"), |
| 2256 | "the environment was cleared and not rebuilt, so every script under dev/ \ |
| 2257 | dies on its first line: {}", saw); |
| 2258 | Ok(()) |
| 2259 | } |
| 2260 | |
| 2261 | #[tokio::test] |
| 2262 | async fn a_verifier_git_has_never_heard_of_does_not_run() -> Outcome<()> { |
| 2263 | let root = res!(tree("prov_untracked")); |
| 2264 | let marker = PathBuf::from(fmt!("{}", root.join("..").join("prov_untracked.escaped").display())); |
| 2265 | res!(git(&root, &["init", "-q"])); |
| 2266 | res!(put(&root, "probe", &escaper(&marker))); |
| 2267 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2268 | assert!(!a.escaped, "an untracked verifier ran and wrote outside the granted root"); |
| 2269 | let no = res!(a.refused.ok_or_else(|| err!( |
| 2270 | "nothing was refused, so the run happened"; Test, Invalid))); |
| 2271 | assert!(no.contains("dev/verify_probe.mjs"), "the refusal must name the file: {}", no); |
| 2272 | Ok(()) |
| 2273 | } |
| 2274 | |
| 2275 | #[tokio::test] |
| 2276 | async fn a_tracked_verifier_the_model_has_edited_does_not_run() -> Outcome<()> { |
| 2277 | let root = res!(tree("prov_edited")); |
| 2278 | let marker = PathBuf::from(fmt!("{}", root.join("..").join("prov_edited.escaped").display())); |
| 2279 | res!(git(&root, &["init", "-q"])); |
| 2280 | res!(put(&root, "probe", "console.log(' ok it ran');\n")); |
| 2281 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2282 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2283 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2284 | // Now the model edits it. `git ls-files --error-unmatch` goes on saying yes. |
| 2285 | res!(put(&root, "probe", &escaper(&marker))); |
| 2286 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2287 | assert!(!a.escaped, "an edited verifier ran and wrote outside the granted root"); |
| 2288 | assert!(a.refused.is_some(), "nothing was refused, so the run happened"); |
| 2289 | // And the whole of §1.21's second half: staging it is not committing it. |
| 2290 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2291 | let b = res!(attempt(&root, "probe", &marker).await); |
| 2292 | assert!(!b.escaped, |
| 2293 | "'git add' bought the model an unfenced run, so the check is about the index and \ |
| 2294 | not about the bytes"); |
| 2295 | assert!(b.refused.is_some(), "a staged edit was accepted"); |
| 2296 | Ok(()) |
| 2297 | } |
| 2298 | |
| 2299 | #[tokio::test] |
| 2300 | async fn a_folder_with_no_git_in_it_cannot_vouch_for_anything() -> Outcome<()> { |
| 2301 | let root = res!(tree("prov_nogit")); |
| 2302 | let marker = PathBuf::from(fmt!("{}", root.join("..").join("prov_nogit.escaped").display())); |
| 2303 | res!(put(&root, "probe", &escaper(&marker))); |
| 2304 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2305 | assert!(!a.escaped, |
| 2306 | "a granted folder with no repository in it ran a verifier outside the fence, and \ |
| 2307 | nothing there could have said whose code it was"); |
| 2308 | assert!(a.refused.is_some(), "nothing was refused, so the run happened"); |
| 2309 | Ok(()) |
| 2310 | } |
| 2311 | |
| 2312 | #[tokio::test] |
| 2313 | async fn the_refusal_hands_the_daimon_the_fenced_route() -> Outcome<()> { |
| 2314 | let root = res!(tree("prov_sentence")); |
| 2315 | let marker = PathBuf::from(fmt!("{}", root.join("..").join("prov_sentence.escaped").display())); |
| 2316 | res!(git(&root, &["init", "-q"])); |
| 2317 | res!(put(&root, "probe", &escaper(&marker))); |
| 2318 | let a = res!(attempt(&root, "probe", &marker).await); |
| 2319 | let no = res!(a.refused.ok_or_else(|| err!("nothing was refused"; Test, Invalid))); |
| 2320 | // A refusal a model cannot converge on costs the run anyway. Each of these is a |
| 2321 | // thing the daimon must be able to DO next, and the sentence is asserted for |
| 2322 | // meaning rather than for length. |
| 2323 | // The argv form and not a command line, because that is what 'run' takes: a |
| 2324 | // refusal that hands a model a shell string teaches it the one spelling the |
| 2325 | // tool refuses. |
| 2326 | assert!(no.contains(r#"["node","dev/verify_probe.mjs"]"#), |
| 2327 | "the refusal must hand over the argv that runs it fenced: {}", no); |
| 2328 | assert!(no.contains("'run'"), |
| 2329 | "the refusal must name the tool that command goes to: {}", no); |
| 2330 | assert!(no.contains("commit"), |
| 2331 | "the refusal must say what makes the unfenced run available: {}", no); |
| 2332 | assert!(no.contains("browser"), |
| 2333 | "the refusal must say what the fenced route cannot do, or a daimon will read the \ |
| 2334 | fenced run's failure as the verifier being broken: {}", no); |
| 2335 | Ok(()) |
| 2336 | } |
| 2337 | |
| 2338 | #[test] |
| 2339 | fn a_verifier_that_is_a_symlink_is_not_the_commit_whatever_git_says() -> Outcome<()> { |
| 2340 | let root = res!(tree("prov_link")); |
| 2341 | res!(git(&root, &["init", "-q"])); |
| 2342 | // The target is an ordinary file inside the granted root, which every fenced |
| 2343 | // command may rewrite. Committing the LINK commits the name and not the bytes. |
| 2344 | let target = root.join("notes.mjs"); |
| 2345 | res!(fs::write(&target, "console.log(' ok it ran');\n") |
| 2346 | .map_err(|e| err!(e, "writing {:?}", target; Test, IO))); |
| 2347 | #[cfg(unix)] |
| 2348 | res!(std::os::unix::fs::symlink("../notes.mjs", |
| 2349 | root.join(DEV_DIR).join("verify_probe.mjs")) |
| 2350 | .map_err(|e| err!(e, "linking"; Test, IO))); |
| 2351 | res!(git(&root, &["add", "-A"])); |
| 2352 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2353 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2354 | assert_eq!(Provenance::Changed(Change::Linked), |
| 2355 | provenance(&root, "verify_probe.mjs"), |
| 2356 | "a committed symlink read as the commit's own bytes"); |
| 2357 | let no = res!(unfenced_refusal(&root, "verify_probe.mjs").ok_or_else(|| err!( |
| 2358 | "a committed symlink was allowed to run unfenced"; Test, Invalid))); |
| 2359 | assert!(no.contains("symbolic link"), "{}", no); |
| 2360 | Ok(()) |
| 2361 | } |
| 2362 | |
| 2363 | #[test] |
| 2364 | fn membership_is_not_content_and_this_is_where_the_two_part() -> Outcome<()> { |
| 2365 | let root = res!(tree("prov_parts")); |
| 2366 | res!(git(&root, &["init", "-q"])); |
| 2367 | res!(put(&root, "probe", "console.log(' ok one');\n")); |
| 2368 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2369 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2370 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2371 | assert_eq!(Provenance::Committed, provenance(&root, "verify_probe.mjs")); |
| 2372 | assert!(unfenced_refusal(&root, "verify_probe.mjs").is_none()); |
| 2373 | |
| 2374 | // One appended comment. This is the case `REVIEW.md` §1.21 turns on: the index |
| 2375 | // goes on saying yes, and the bytes are the model's. |
| 2376 | res!(put(&root, "probe", "console.log(' ok one');\n// and now mine\n")); |
| 2377 | let git_path = res!(on_path("git").ok_or_else(|| err!("no git"; Test, Missing))); |
| 2378 | let listed = res!(std::process::Command::new(&git_path) |
| 2379 | .arg("-C").arg(&root) |
| 2380 | .args(["ls-files", "--error-unmatch", "--", "dev/verify_probe.mjs"]) |
| 2381 | .stdout(Stdio::null()).stderr(Stdio::null()) |
| 2382 | .status().map_err(|e| err!(e, "ls-files"; Test, IO))); |
| 2383 | assert!(listed.success(), |
| 2384 | "the premise of this test is that git goes on naming an edited file"); |
| 2385 | assert_eq!(Provenance::Changed(Change::Edited), provenance(&root, "verify_probe.mjs"), |
| 2386 | "the check followed the index rather than the bytes"); |
| 2387 | Ok(()) |
| 2388 | } |
| 2389 | |
| 2390 | #[test] |
| 2391 | fn the_report_says_whose_bytes_it_ran() -> Outcome<()> { |
| 2392 | let script = Script { |
| 2393 | name: fmt!("probe"), |
| 2394 | file: fmt!("verify_probe.mjs"), |
| 2395 | path: PathBuf::from("/w/dev/verify_probe.mjs"), |
| 2396 | breaks: Vec::new(), |
| 2397 | prov: Provenance::Committed, |
| 2398 | }; |
| 2399 | let clean = pass_of("clean", " ok one\n"); |
| 2400 | let txt = report(&script, &[(clean, None)], |
| 2401 | &Verdict::Unproven { passed: 1, failed: 0, declared: Vec::new() }, &[]); |
| 2402 | let first = res!(txt.lines().next().ok_or_else(|| err!("empty report"; Test, Invalid))); |
| 2403 | assert!(first.contains("byte for byte the commit's"), |
| 2404 | "the report's first line must say whose bytes ran: {}", first); |
| 2405 | Ok(()) |
| 2406 | } |
| 2407 | |
| 2408 | #[tokio::test] |
| 2409 | async fn a_verifier_that_rewrites_itself_gets_no_second_run() -> Outcome<()> { |
| 2410 | let root = res!(tree("prov_swap")); |
| 2411 | // Committed, and what it does when it runs is append to its own source. So the |
| 2412 | // clean run is the commit's and the break run is not, which is the race the |
| 2413 | // re-ask before every spawn exists for: no test can hold a file still for the |
| 2414 | // minutes a real sequence takes, and a verifier that changes itself is the same |
| 2415 | // event arriving on schedule. |
| 2416 | let src = "// node dev/verify_probe.mjs --break swap # 1: the break\n\ |
| 2417 | import fs from 'node:fs';\n\ |
| 2418 | console.log(' ok it ran');\n\ |
| 2419 | fs.appendFileSync('dev/verify_probe.mjs', '// and now it is mine\\n');\n"; |
| 2420 | res!(git(&root, &["init", "-q"])); |
| 2421 | res!(put(&root, "probe", src)); |
| 2422 | res!(git(&root, &["add", "dev/verify_probe.mjs"])); |
| 2423 | res!(git(&root, &["-c", "user.name=T", "-c", "user.email=t@t", |
| 2424 | "commit", "-q", "--no-verify", "-m", "probe"])); |
| 2425 | let script = res!(resolve(&root, "probe").map_err(|w| err!("{}", w; Test, Invalid))); |
| 2426 | assert_eq!(vec![fmt!("swap")], script.breaks); |
| 2427 | assert_eq!(Provenance::Committed, script.prov); |
| 2428 | let node = res!(on_path("node").ok_or_else(|| err!("no node"; Test, Missing))); |
| 2429 | let jdir = root.join("..").join("prov_swap-journal"); |
| 2430 | let jdir = PathBuf::from(fmt!("{}", jdir.display())); |
| 2431 | if jdir.exists() { |
| 2432 | res!(fs::remove_dir_all(&jdir).map_err(|e| err!(e, "clearing"; Test, IO))); |
| 2433 | } |
| 2434 | let jr = res!(Journal::open(crate::journal::Cfg::at(&jdir))); |
| 2435 | let job = Job { |
| 2436 | id: fmt!("swap"), |
| 2437 | root: root.clone(), |
| 2438 | script, |
| 2439 | breaks: vec![fmt!("swap")], |
| 2440 | node, |
| 2441 | budget: Duration::from_millis(60_000), |
| 2442 | ledger: Ledger::new(Arc::new(Mutex::new(jr)), Arc::new(AtomicBool::new(true))), |
| 2443 | }; |
| 2444 | let (tx, mut rx) = tokio::sync::mpsc::channel(64); |
| 2445 | res!(Box::pin(conduct(job, tx)).await); |
| 2446 | let mut out = String::new(); |
| 2447 | while let Some(r) = rx.recv().await { |
| 2448 | if let Resp::Chunk { stream: Stream::Out, data, .. } = r { |
| 2449 | out.push_str(&data); |
| 2450 | } |
| 2451 | } |
| 2452 | assert!(out.contains("BREAK swap"), "the break is not in the report: {}", out); |
| 2453 | assert!(out.contains("NEVER RAN"), |
| 2454 | "the break ran against a file the clean run had rewritten: {}", out); |
| 2455 | assert!(out.contains("stopped being the commit's"), |
| 2456 | "the report does not say why the break never ran: {}", out); |
| 2457 | Ok(()) |
| 2458 | } |
| 2459 | } |