Skip to main content

borg/
live.rs

1//! Delivery modes: when the conscience judges the work (docs/delivery.md).
2//!
3//! - **hook**: early delivery and queued lessons are free word matching; one graded model
4//!   call runs at the turn end (main.rs `stop`), and only a grade 3 blocks the stop.
5//! - **live** (the default, chosen by the live-vs-hook eval): the same graded call also runs *during* the turn. After a state-changing tool call
6//!   (prefilter.rs `changes`), the PostToolUse hook spawns a detached `borg check` and returns at once;
7//!   the check judges the work so far and leaves its verdict in the `.queued` file, which the next
8//!   prompt or tool call delivers (main.rs `queued`). The stop still runs as the backstop, and skips its
9//!   own model call when a live check this turn already covered every change (`covered`).
10//!
11//! The mode is chosen per collective (`"mode"` in collectives.json, collective.rs) and `BORG_MODE`
12//! overrides it; unset or anything else is `DEFAULT`. Every harness runs it: Claude Code and omp take the
13//! next tool call's context as PostToolUse `additionalContext`, the opencode adapter adds it to the tool
14//! output (1.x) or steers it into the session (2.x). Without a Claude transcript the work so far is the
15//! turn's tool calls (main.rs `turn_file`).
16//!
17//! Three small files per session beside its state file, no daemon:
18//! - `<S>.live.n`: one line per state-changing tool call (the change counter);
19//! - `<S>.live.mark`: `<covered> <started_ms> <turn>`, the counter value the last completed check covered,
20//!   when the last check started (the debounce runs from there) and the turn it started in (main.rs
21//!   `turn_id`: a check of an earlier turn never covers this one);
22//! - `<S>.live.lock`: `<pid> <started_ms>` while a check (or a live stop) runs, linked into place whole
23//!   so two quick tool calls never run two checks; stale when its process is gone or it is older than
24//!   `stale_ms`.
25
26use std::io::Write;
27use std::path::{Path, PathBuf};
28use std::time::Duration;
29
30#[derive(Debug, Clone, Copy, PartialEq, Eq)]
31pub enum Mode {
32    Hook,
33    Live,
34}
35
36impl Mode {
37    pub fn name(self) -> &'static str {
38        match self {
39            Mode::Hook => "hook",
40            Mode::Live => "live",
41        }
42    }
43}
44
45/// The mode in force and where it came from (`BORG_MODE`, `config` or `default`), with a warning
46/// when a value was not a mode (`borg doctor` shows it; the mode is then `DEFAULT`).
47#[derive(Debug, Clone, PartialEq, Eq)]
48pub struct Resolved {
49    pub mode: Mode,
50    pub source: &'static str,
51    pub warning: Option<String>,
52}
53
54/// The mode when neither BORG_MODE nor the collective sets one: live, which passed the eval's fixed
55/// rule against hook (docs/delivery.md, "Why live is the default"). `"mode": "hook"` or `BORG_MODE=hook` switches back.
56pub const DEFAULT: Mode = Mode::Live;
57
58/// `env` (BORG_MODE) wins over `configured` (the collective's `mode`), else `DEFAULT`. An unknown value
59/// is `DEFAULT` plus a warning: a typo must never break the hooks.
60pub fn resolve(env: Option<&str>, configured: Option<&str>) -> Resolved {
61    let pick = |v: &str, source: &'static str| match v.trim().to_ascii_lowercase().as_str() {
62        "hook" => Resolved { mode: Mode::Hook, source, warning: None },
63        "live" => Resolved { mode: Mode::Live, source, warning: None },
64        _ => Resolved { mode: DEFAULT, source, warning: Some(format!("{source}: mode {v:?} is neither \"hook\" nor \"live\"; using {}", DEFAULT.name())) },
65    };
66    match (env.filter(|v| !v.trim().is_empty()), configured.filter(|v| !v.trim().is_empty())) {
67        (Some(v), _) => pick(v, "BORG_MODE"),
68        (None, Some(v)) => pick(v, "config"),
69        (None, None) => Resolved { mode: DEFAULT, source: "default", warning: None },
70    }
71}
72
73/// The mode of collective `name` in `cfg` for this process (BORG_MODE first).
74pub fn of(cfg: Option<&crate::collective::Config>, name: &str) -> Resolved {
75    let configured = cfg.and_then(|c| c.collectives.iter().find(|e| e.name == name)).and_then(|e| e.mode.as_deref());
76    resolve(std::env::var("BORG_MODE").ok().as_deref(), configured)
77}
78
79/// The mode of this hook process: the hook exports the resolved mode as `BORG_MODE` (main.rs), so the
80/// stop and the spawned check read the same one.
81pub fn current() -> Mode {
82    resolve(std::env::var("BORG_MODE").ok().as_deref(), None).mode
83}
84
85/// Why live mode cannot run here, if it cannot: it needs the consequence model, which word matching
86/// alone (`BORG_MATCH=bm25`) and the eval modes `BORG_ONLY` and `BORG_CRITIC` replace.
87pub fn unsupported() -> Option<&'static str> {
88    if std::env::var("BORG_MATCH").as_deref() == Ok("bm25") {
89        Some("BORG_MATCH=bm25")
90    } else if std::env::var_os("BORG_ONLY").is_some() {
91        Some("BORG_ONLY")
92    } else if std::env::var_os("BORG_CRITIC").is_some() {
93        Some("BORG_CRITIC")
94    } else {
95        None
96    }
97}
98
99/// The least time between the starts of two checks of a session: `BORG_LIVE_DEBOUNCE_S`, default 45 s.
100pub fn debounce_ms() -> i64 {
101    std::env::var("BORG_LIVE_DEBOUNCE_S").ok().and_then(|v| v.trim().parse::<i64>().ok()).unwrap_or(45).max(0) * 1000
102}
103
104/// A lock older than this belongs to a check that cannot still be waiting for its model (or to a pid
105/// that was reused): twice the predict timeout plus 30 s.
106pub fn stale_ms() -> i64 {
107    2 * crate::predict::timeout().as_millis() as i64 + 30_000
108}
109
110fn beside(ctx: &Path, ext: &str) -> PathBuf {
111    ctx.with_extension(ext)
112}
113
114pub fn counter_file(ctx: &Path) -> PathBuf {
115    beside(ctx, "live.n")
116}
117
118pub fn mark_file(ctx: &Path) -> PathBuf {
119    beside(ctx, "live.mark")
120}
121
122pub fn lock_file(ctx: &Path) -> PathBuf {
123    beside(ctx, "live.lock")
124}
125
126/// State-changing tool calls of the session so far.
127pub fn count(ctx: &Path) -> u64 {
128    std::fs::read_to_string(counter_file(ctx)).map_or(0, |t| t.lines().count() as u64)
129}
130
131/// Counts one state-changing tool call.
132pub fn bump(ctx: &Path, tool: &str) {
133    let _ = std::fs::OpenOptions::new().create(true).append(true).open(counter_file(ctx)).and_then(|mut f| writeln!(f, "{tool}"));
134}
135
136/// What the last check covered, when it started and in which turn (None: a mark without one).
137#[derive(Debug, Clone, Copy, PartialEq, Eq)]
138pub struct Mark {
139    pub covered: u64,
140    pub started_ms: i64,
141    pub turn: Option<u64>,
142}
143
144pub fn read_mark(ctx: &Path) -> Option<Mark> {
145    let t = std::fs::read_to_string(mark_file(ctx)).ok()?;
146    let mut it = t.split_whitespace();
147    Some(Mark { covered: it.next()?.parse().ok()?, started_ms: it.next()?.parse().ok()?, turn: it.next().and_then(|v| v.parse().ok()) })
148}
149
150pub fn write_mark(ctx: &Path, m: Mark) {
151    let turn = m.turn.map_or(String::new(), |t| format!(" {t}"));
152    let _ = std::fs::write(mark_file(ctx), format!("{} {}{turn}\n", m.covered, m.started_ms));
153}
154
155#[derive(Debug, Clone, Copy, PartialEq, Eq)]
156pub enum Lock {
157    Free,
158    Held,
159    /// The file is there, but its check is gone (or far too old): it may be taken over.
160    Stale,
161}
162
163/// Whether process `pid` still exists (signal 0 checks without sending).
164pub fn alive(pid: i32) -> bool {
165    // SAFETY: kill with signal 0 only checks that the process exists and may be signalled.
166    pid > 0 && (unsafe { libc::kill(pid, 0) } == 0 || std::io::Error::last_os_error().raw_os_error() == Some(libc::EPERM))
167}
168
169/// The state of a lock whose file reads `text` at `now`. A lock is written whole (`acquire`), so one
170/// that does not parse is someone's half-written file: held, until its file is older than `stale_after`.
171fn state_of(text: &str, file_ms: i64, now: i64, stale_after: i64, alive: impl Fn(i32) -> bool) -> Lock {
172    let mut it = text.split_whitespace();
173    match (it.next().and_then(|p| p.parse::<i32>().ok()), it.next().and_then(|m| m.parse::<i64>().ok())) {
174        (Some(pid), Some(ms)) if alive(pid) && now - ms <= stale_after => Lock::Held,
175        (Some(_), Some(_)) => Lock::Stale,
176        _ if now - file_ms > stale_after => Lock::Stale,
177        _ => Lock::Held,
178    }
179}
180
181fn modified_ms(path: &Path) -> i64 {
182    std::fs::metadata(path).and_then(|m| m.modified()).ok().and_then(|t| t.duration_since(std::time::UNIX_EPOCH).ok()).map_or(0, |d| d.as_millis() as i64)
183}
184
185/// The lock's state at `now`; `alive` tells whether a pid still runs.
186pub fn lock_state(ctx: &Path, now: i64, stale_after: i64, alive: impl Fn(i32) -> bool) -> Lock {
187    let path = lock_file(ctx);
188    let Ok(t) = std::fs::read_to_string(&path) else { return Lock::Free };
189    state_of(&t, modified_ms(&path), now, stale_after, alive)
190}
191
192/// A held lock; dropping it removes the file, if it is still this one's.
193#[derive(Debug)]
194pub struct Guard {
195    path: PathBuf,
196    text: String,
197}
198
199impl Drop for Guard {
200    fn drop(&mut self) {
201        if std::fs::read_to_string(&self.path).is_ok_and(|t| t == self.text) {
202            let _ = std::fs::remove_file(&self.path);
203        }
204    }
205}
206
207/// Takes the session's lock for this process, or None when another check holds it. The lock is written
208/// to a file of its own first and hard-linked into place, which fails when a lock is there: no one ever
209/// reads a half-written lock. A stale lock is moved aside (a rename: only one process gets it) and taken
210/// once more; when what was moved aside is no longer the stale lock seen, someone took it over first,
211/// and it goes back.
212pub fn acquire(ctx: &Path, now: i64) -> Option<Guard> {
213    let path = lock_file(ctx);
214    let pid = std::process::id();
215    let text = format!("{pid} {now}\n");
216    let tmp = path.with_extension(format!("lock.{pid}"));
217    std::fs::write(&tmp, &text).ok()?;
218    let mut got = None;
219    for _ in 0..2 {
220        if std::fs::hard_link(&tmp, &path).is_ok() {
221            got = Some(Guard { path: path.clone(), text: text.clone() });
222            break;
223        }
224        let Ok(seen) = std::fs::read_to_string(&path) else { continue };
225        if state_of(&seen, modified_ms(&path), now, stale_ms(), alive) != Lock::Stale {
226            break;
227        }
228        let aside = path.with_extension(format!("lock.stale.{pid}"));
229        if std::fs::rename(&path, &aside).is_err() {
230            continue;
231        }
232        let moved = std::fs::read_to_string(&aside).unwrap_or_default();
233        if moved != seen {
234            let _ = std::fs::hard_link(&aside, &path);
235            let _ = std::fs::remove_file(&aside);
236            break;
237        }
238        let _ = std::fs::remove_file(&aside);
239    }
240    let _ = std::fs::remove_file(&tmp);
241    got
242}
243
244/// Runs `f` while holding an exclusive lock on `<file>.lock` (kept, never removed), so appends to `file`
245/// and taking it whole (a rename) never interleave: an append never lands in a file already taken.
246/// When the lock cannot be taken `f` runs unlocked: for a plain append no worse than without it.
247pub fn locked<T>(file: &Path, f: impl FnOnce() -> T) -> T {
248    let held = lock_of(file).ok();
249    let out = f();
250    if let Some(l) = held {
251        let _ = l.unlock();
252    }
253    out
254}
255
256/// `<file>.lock`, opened and locked exclusively.
257fn lock_of(file: &Path) -> std::io::Result<std::fs::File> {
258    let lock = std::fs::OpenOptions::new()
259        .create(true)
260        .truncate(false)
261        .write(true)
262        .open(file.with_extension(format!("{}.lock", file.extension().and_then(|e| e.to_str()).unwrap_or_default())))?;
263    lock.lock()?;
264    Ok(lock)
265}
266
267/// Whether a check should start now (Ok) or why not: not live mode, a check still running, no change
268/// since the last completed check, or the last check started less than `debounce` ago.
269pub fn decide(mode: Mode, lock: Lock, count: u64, mark: Option<Mark>, now: i64, debounce: i64) -> Result<(), &'static str> {
270    if mode != Mode::Live {
271        return Err("hook mode");
272    }
273    if lock == Lock::Held {
274        return Err("running");
275    }
276    if count <= mark.map_or(0, |m| m.covered) {
277        return Err("nothing new");
278    }
279    if mark.is_some_and(|m| now - m.started_ms < debounce) {
280        return Err("debounce");
281    }
282    Ok(())
283}
284
285/// The stop's skip in live mode: no check runs, a check that started in this turn (`turn`, main.rs
286/// `turn_id`) covered every change, and none came since. A check of an earlier turn, or one that
287/// finished after its turn's stop, never covers this turn's message.
288pub fn covered(lock: Lock, count: u64, mark: Option<Mark>, turn: u64) -> bool {
289    lock != Lock::Held && count > 0 && mark.is_some_and(|m| m.covered == count && m.turn == Some(turn))
290}
291
292/// Waits up to `max_ms` for a running check to finish; the lock's state after.
293pub fn wait_free(ctx: &Path, max_ms: u64) -> Lock {
294    let until = std::time::Instant::now() + std::time::Duration::from_millis(max_ms);
295    loop {
296        let l = lock_state(ctx, crate::learn::now_ms(), stale_ms(), alive);
297        if l != Lock::Held || std::time::Instant::now() >= until {
298            return l;
299        }
300        std::thread::sleep(std::time::Duration::from_millis(200));
301    }
302}
303
304/// How long a stop in live mode waits for a running check.
305pub const STOP_WAIT_MS: u64 = 15_000;
306
307/// The stop's wait for a running check, inside the harness's budget (expert_route.rs `stop_budget`; none:
308/// Claude Code's): at most a quarter of it. omp's 28 s gives 7 s.
309pub fn stop_wait(budget: Option<Duration>) -> Duration {
310    let wait = Duration::from_millis(STOP_WAIT_MS);
311    budget.map_or(wait, |b| wait.min(b / 4))
312}
313
314/// The stop's model call's limit, `spent` into the hook: what the harness's budget leaves but 3 s for the
315/// rest of the hook, never over the model's own limit. omp's 28 s gives at most 18 s, after a 7 s wait.
316pub fn call_limit(budget: Option<Duration>, spent: Duration) -> Duration {
317    let Some(b) = budget else { return crate::predict::timeout() };
318    crate::predict::timeout().min(b.saturating_sub(spent).saturating_sub(Duration::from_secs(3)))
319}
320
321#[cfg(test)]
322mod tests {
323    use super::*;
324
325    fn tmp(tag: &str) -> PathBuf {
326        let d = std::env::temp_dir().join(format!("borg-live-{tag}-{}", std::process::id()));
327        let _ = std::fs::remove_dir_all(&d);
328        std::fs::create_dir_all(&d).unwrap();
329        d.join("s1.ctx")
330    }
331
332    #[test]
333    fn env_beats_config_and_unknown_values_fall_back_to_the_default() {
334        assert_eq!(resolve(None, None), Resolved { mode: Mode::Live, source: "default", warning: None });
335        assert_eq!(resolve(None, Some("live")).mode, Mode::Live);
336        assert_eq!(resolve(Some("hook"), Some("live")), Resolved { mode: Mode::Hook, source: "BORG_MODE", warning: None });
337        assert_eq!(resolve(Some(" LIVE "), None).mode, Mode::Live);
338        assert_eq!(resolve(Some(""), Some("live")).source, "config", "an empty BORG_MODE is unset");
339        let r = resolve(None, Some("eager"));
340        assert_eq!(r.mode, Mode::Live);
341        assert!(r.warning.unwrap().contains("\"eager\""));
342    }
343
344    #[test]
345    fn a_check_starts_only_in_live_mode_with_new_changes_no_running_check_and_after_the_debounce() {
346        let m = |covered, started_ms| Some(Mark { covered, started_ms, turn: None });
347        assert_eq!(decide(Mode::Live, Lock::Free, 1, None, 0, 45_000), Ok(()), "the first change checks at once");
348        assert_eq!(decide(Mode::Hook, Lock::Free, 1, None, 0, 45_000), Err("hook mode"));
349        assert_eq!(decide(Mode::Live, Lock::Held, 3, m(1, 0), 100_000, 45_000), Err("running"));
350        assert_eq!(decide(Mode::Live, Lock::Stale, 3, m(1, 0), 100_000, 45_000), Ok(()), "a stale lock does not block");
351        assert_eq!(decide(Mode::Live, Lock::Free, 2, m(2, 0), 100_000, 45_000), Err("nothing new"));
352        assert_eq!(decide(Mode::Live, Lock::Free, 0, None, 0, 45_000), Err("nothing new"));
353        assert_eq!(decide(Mode::Live, Lock::Free, 3, m(2, 60_000), 100_000, 45_000), Err("debounce"));
354        assert_eq!(decide(Mode::Live, Lock::Free, 3, m(2, 55_000), 100_000, 45_000), Ok(()));
355    }
356
357    #[test]
358    fn the_lock_is_held_while_its_process_lives_and_stale_when_gone_or_too_old() {
359        let ctx = tmp("lock");
360        assert_eq!(lock_state(&ctx, 1000, 120_000, |_| true), Lock::Free);
361        let g = acquire(&ctx, 1000).expect("free lock taken");
362        assert!(acquire(&ctx, 1001).is_none(), "a second check does not run");
363        assert_eq!(lock_state(&ctx, 2000, 120_000, |_| true), Lock::Held);
364        assert_eq!(lock_state(&ctx, 2000, 120_000, |_| false), Lock::Stale, "its process is gone");
365        assert_eq!(lock_state(&ctx, 200_000, 120_000, |_| true), Lock::Stale, "older than twice the timeout");
366        drop(g);
367        assert_eq!(lock_state(&ctx, 2000, 120_000, |_| true), Lock::Free, "the guard removes it");
368        // A lock left by a process that is gone is taken over.
369        std::fs::write(lock_file(&ctx), format!("{} 1000\n", i32::MAX)).unwrap();
370        assert!(!alive(i32::MAX));
371        let g = acquire(&ctx, 2000).expect("stale lock taken over");
372        // A guard whose lock was taken over leaves the new holder's file alone.
373        std::fs::write(lock_file(&ctx), "1 2000\n").unwrap();
374        drop(g);
375        assert!(lock_file(&ctx).exists(), "not this guard's lock any more");
376    }
377
378    #[test]
379    fn a_half_written_lock_is_held_until_its_file_is_old() {
380        assert_eq!(state_of("", 1000, 2000, 120_000, |_| false), Lock::Held, "being written: never taken over");
381        assert_eq!(state_of("12", 1000, 2000, 120_000, |_| false), Lock::Held);
382        assert_eq!(state_of("", 1000, 200_000, 120_000, |_| false), Lock::Stale, "left empty long ago");
383        assert_eq!(state_of("7 1000", 0, 2000, 120_000, |_| true), Lock::Held);
384        let ctx = tmp("half");
385        std::fs::write(lock_file(&ctx), "").unwrap();
386        assert!(acquire(&ctx, crate::learn::now_ms()).is_none(), "an empty, fresh lock is someone's");
387    }
388
389    #[test]
390    fn counter_and_mark_round_trip() {
391        let ctx = tmp("mark");
392        assert_eq!((count(&ctx), read_mark(&ctx)), (0, None));
393        bump(&ctx, "Edit");
394        bump(&ctx, "Bash");
395        assert_eq!(count(&ctx), 2);
396        write_mark(&ctx, Mark { covered: 2, started_ms: 77, turn: Some(9) });
397        assert_eq!(read_mark(&ctx), Some(Mark { covered: 2, started_ms: 77, turn: Some(9) }));
398        std::fs::write(mark_file(&ctx), "2 77\n").unwrap();
399        assert_eq!(read_mark(&ctx), Some(Mark { covered: 2, started_ms: 77, turn: None }), "a mark of before turns");
400    }
401
402    #[test]
403    fn the_stop_skips_only_when_a_check_of_this_turn_covered_every_change() {
404        let m = Some(Mark { covered: 3, started_ms: 0, turn: Some(40) });
405        assert!(covered(Lock::Free, 3, m, 40));
406        assert!(!covered(Lock::Held, 3, m, 40), "a running check");
407        assert!(!covered(Lock::Free, 4, m, 40), "a change after the check");
408        assert!(!covered(Lock::Free, 3, m, 41), "the check was an earlier turn's");
409        assert!(!covered(Lock::Free, 3, Some(Mark { covered: 3, started_ms: 0, turn: None }), 40), "a mark without its turn");
410        assert!(!covered(Lock::Free, 0, None, 40));
411    }
412
413    #[test]
414    fn the_stop_fits_its_wait_and_model_call_into_the_harness_budget() {
415        let s = Duration::from_secs;
416        assert_eq!((stop_wait(None), call_limit(None, s(30))), (s(15), s(45)), "Claude Code: unchanged");
417        assert_eq!((stop_wait(Some(s(90))), call_limit(Some(s(90)), s(15))), (s(15), s(45)), "opencode: room for both");
418        assert_eq!((stop_wait(Some(s(28))), call_limit(Some(s(28)), s(7))), (s(7), s(18)), "omp");
419        assert_eq!(call_limit(Some(s(28)), s(11)), s(14), "time spent before the call comes off it");
420        assert_eq!(call_limit(Some(s(28)), s(27)), Duration::ZERO);
421        assert_eq!(stop_wait(Some(s(8))), s(2));
422    }
423
424    #[test]
425    fn a_lock_that_cannot_be_taken_still_runs_f_unlocked() {
426        // The lock file cannot be opened (a directory in its place): `locked` runs `f` unlocked.
427        let d = tmp("try-locked");
428        let file = d.join("tasks.jsonl");
429        std::fs::create_dir_all(d.join("tasks.jsonl.lock")).unwrap();
430        assert!(locked(&file, || true));
431        let _ = std::fs::remove_dir_all(d);
432    }
433
434    #[test]
435    fn appends_under_the_queue_lock_never_land_in_a_taken_file() {
436        let ctx = tmp("queue");
437        let q = ctx.with_extension("queued");
438        let writers: Vec<_> = (0..8)
439            .map(|i| {
440                let q = q.clone();
441                std::thread::spawn(move || {
442                    for j in 0..50 {
443                        locked(&q, || std::fs::OpenOptions::new().create(true).append(true).open(&q).and_then(|mut f| writeln!(f, "{i}-{j}")).unwrap());
444                    }
445                })
446            })
447            .collect();
448        let mut got = 0;
449        let taken = q.with_extension("queued.t");
450        while writers.iter().any(|w| !w.is_finished()) {
451            if locked(&q, || std::fs::rename(&q, &taken).is_ok()) {
452                got += std::fs::read_to_string(&taken).unwrap().lines().count();
453            }
454        }
455        for w in writers {
456            w.join().unwrap();
457        }
458        if std::fs::rename(&q, &taken).is_ok() {
459            got += std::fs::read_to_string(&taken).unwrap().lines().count();
460        }
461        assert_eq!(got, 400, "every line taken once");
462    }
463}