1use std::path::{Path, PathBuf};
28
29use anyhow::Result;
30
31use crate::git;
32use crate::run::{Liveness, RunState, RunStatus};
33
34#[derive(Debug, Clone)]
36pub struct Takeover {
37 pub earlier: Vec<String>,
40 pub home: PathBuf,
42 pub choice: Option<crate::reconcile::Choice>,
46 pub task_instruction: Option<String>,
50}
51
52#[derive(Debug, Clone, PartialEq, Eq)]
60pub enum Refused {
61 Foreign {
63 branch: String,
65 path: String,
67 why: String,
69 },
70 Unsafe {
72 branch: String,
74 path: String,
76 why: String,
78 },
79 ReleaseFailed {
81 branch: String,
83 path: String,
85 run: String,
87 },
88}
89
90impl std::fmt::Display for Refused {
91 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
92 match self {
93 Refused::Foreign { branch, path, why } => write!(
94 f,
95 "branch `{branch}` is checked out in {path}, which magi will not remove by \
96 itself: {why}. Remove that worktree (`git worktree remove`) if it is not \
97 needed, and try again."
98 ),
99 Refused::Unsafe { branch, path, why } => write!(
100 f,
101 "branch `{branch}` is checked out in {path}: {why}. Commit or discard the work \
102 there and remove that worktree (`git worktree remove`), or say the run may be \
103 discarded, and try again."
104 ),
105 Refused::ReleaseFailed { branch, path, run } => write!(
106 f,
107 "branch `{branch}` is checked out in {path} by run {run}, and releasing that \
108 worktree failed or found it changed (git refuses to remove a worktree with \
109 uncommitted changes); it was left as it was"
110 ),
111 }
112 }
113}
114
115impl std::error::Error for Refused {}
116
117#[derive(Debug, Clone)]
120pub struct Released {
121 pub old_id: String,
123 index: usize,
124 path: PathBuf,
125 home: PathBuf,
126 successor: String,
127 tip: String,
130 pub audit: String,
132}
133
134impl Released {
135 pub async fn restore(&self, repo: &Path, branch: &str) {
139 let Ok(_owner) = crate::pr_head::owner_lock(&self.home, &self.old_id) else {
140 tracing::warn!("could not lock run {} for handover rollback", self.old_id);
141 return;
142 };
143 let Ok(mut state) = RunState::load_under(&self.old_id, &self.home) else {
144 return;
145 };
146 let claimed =
147 crate::daemon::is_working_on_other(&self.home, &self.old_id, jiff::Timestamp::now())
148 .unwrap_or(true);
149 if state.released_to.as_deref() != Some(self.successor.as_str())
150 || state.status == RunStatus::Superseded
151 || state.pr_taken_over_by.is_some()
152 || state.liveness_under(&self.home, claimed) != Liveness::Dead
153 {
154 tracing::warn!(
155 "run {} changed ownership; handover rollback refused",
156 self.old_id
157 );
158 return;
159 }
160 let path = self.path.to_string_lossy().to_string();
161 let refname = format!("refs/heads/{branch}");
164 if git::rev_parse(repo, &refname).await.ok().as_deref() != Some(self.tip.as_str())
165 && let Err(e) = git::git(repo, &["branch", "-f", branch, &self.tip]).await
166 {
167 tracing::warn!("could not put `{branch}` back at {}: {e:#}", self.tip);
168 return;
169 }
170 if let Err(e) = git::git(repo, &["worktree", "add", &path, branch]).await {
171 tracing::warn!(
172 "could not put run {}'s worktree back at {path}: {e:#}",
173 self.old_id
174 );
175 return;
176 }
177 if let Some(c) = state.candidates.get_mut(self.index) {
178 c.folded = false;
179 }
180 state.released_to = None;
181 state.released_branches.retain(|b| b != branch);
182 state.events.pop();
183 let put_back = state.save_under(&self.home);
184 if let Err(e) = put_back {
185 tracing::warn!("could not unmark run {}: {e:#}", self.old_id);
186 }
187 }
188}
189
190#[derive(Debug, Clone)]
192pub struct Holder {
193 pub run: String,
195 pub status: RunStatus,
197 pub liveness: Liveness,
199 pub driver_unproven: bool,
202 pub dirty: bool,
204 pub dirty_files: Vec<String>,
206 pub driver_pid: Option<u32>,
208 pub head: String,
210 pub tip: String,
212}
213
214#[derive(Debug, Clone, PartialEq, Eq)]
216pub enum Owner {
217 EarlierAttempt,
219 MagiRun,
222 Foreign(String),
224}
225
226#[derive(Debug, Clone, PartialEq, Eq)]
228pub enum Decision {
229 Release,
231 Refuse(String),
233}
234
235fn short(sha: &str) -> String {
236 sha.chars().take(7).collect()
237}
238
239pub fn decide(owner: &Owner, holder: &Holder) -> Decision {
245 if let Owner::Foreign(why) = owner {
246 return Decision::Refuse(why.clone());
247 }
248 let mut why = Vec::new();
249 if holder.liveness == Liveness::Live {
250 why.push(format!(
251 "run {} is being worked on right now{}",
252 crate::run::short_of(&holder.run),
253 holder
254 .driver_pid
255 .map(|p| format!(" (driver pid {p})"))
256 .unwrap_or_default()
257 ));
258 } else if holder.liveness == Liveness::Unknown || holder.driver_unproven {
259 why.push("its driver process could not be shown to be gone".to_owned());
260 }
261 if holder.status == RunStatus::Landing {
262 why.push("that run is in `landing`, possibly waiting on an approval".to_owned());
263 }
264 if holder.dirty {
265 let files = holder.dirty_files.join(", ");
266 why.push(format!("its worktree has uncommitted changes ({files})"));
267 }
268 if holder.head != holder.tip {
269 why.push("its HEAD is not at the branch tip".to_owned());
270 }
271 if why.is_empty() {
272 return Decision::Release;
273 }
274 let kind = match owner {
275 Owner::EarlierAttempt => "an earlier attempt at this task",
276 _ => "a magi run of another task",
277 };
278 Decision::Refuse(format!(
279 "run {} (status `{}`, worktree {}, HEAD {}, branch tip {}) is {kind} and still has the \
280 branch checked out, so it was not released automatically: {}",
281 crate::run::short_of(&holder.run),
282 holder.status.as_str(),
283 if holder.dirty { "dirty" } else { "clean" },
284 short(&holder.head),
285 short(&holder.tip),
286 why.join("; ")
287 ))
288}
289
290async fn inspect(
292 repo: &Path,
293 branch: &str,
294 path: &Path,
295 state: &RunState,
296 home: &Path,
297) -> Result<Holder> {
298 let claimed = crate::daemon::is_working_on_other(home, &state.id, jiff::Timestamp::now())?;
299 let liveness = state.liveness_under(home, claimed);
300 let porcelain = git::git(path, &["status", "--porcelain", "--untracked-files=normal"]).await?;
303 Ok(Holder {
304 run: state.id.clone(),
305 status: state.status,
306 liveness,
307 driver_unproven: liveness == Liveness::Unknown,
308 dirty: !porcelain.trim().is_empty(),
309 dirty_files: porcelain
310 .lines()
311 .take(5)
312 .map(|l| l.get(3..).unwrap_or(l).trim().to_owned())
313 .collect(),
314 driver_pid: state.driver_pid,
315 head: git::rev_parse(path, "HEAD").await?,
316 tip: git::rev_parse(repo, &format!("refs/heads/{branch}")).await?,
317 })
318}
319
320fn same_path(a: &Path, b: &Path) -> bool {
321 let canon = |p: &Path| std::fs::canonicalize(p).unwrap_or_else(|_| p.to_path_buf());
322 canon(a) == canon(b)
323}
324
325fn foreign(branch: &str, path: &Path, why: &str) -> anyhow::Error {
326 Refused::Foreign {
327 branch: branch.to_owned(),
328 path: path.display().to_string(),
329 why: why.to_owned(),
330 }
331 .into()
332}
333
334fn find_magi_owner(
339 path: &Path,
340 home: &Path,
341) -> std::result::Result<Option<(RunState, usize)>, String> {
342 let Some(bay) = path.parent() else {
343 return Ok(None);
344 };
345 let Some(bay_name) = bay.file_name().and_then(|n| n.to_str()) else {
346 return Ok(None);
347 };
348 let mut found = Vec::new();
349 for id in crate::run::list_ids_in(&home.join("runs")) {
350 if crate::run::short_of(&id) != bay_name {
351 continue;
352 }
353 let Ok(state) = RunState::load_under(&id, home) else {
354 return Err(format!("run {bay_name}'s record could not be read"));
355 };
356 if !same_path(&state.worktree_root(), bay) {
357 continue;
358 }
359 if let Some(i) = state
360 .candidates
361 .iter()
362 .position(|c| !c.folded && same_path(&c.worktree, path))
363 {
364 found.push((state, i));
365 }
366 }
367 match found.len() {
368 0 => Ok(None),
369 1 => Ok(found.pop()),
370 _ => Err(format!("more than one run record claims it ({bay_name})")),
371 }
372}
373
374pub(crate) fn landing_checkpoint(state: &RunState) -> bool {
377 state.status.resumable()
378 && state.config.graph.land
379 && state.pr.as_ref().is_some_and(|pr| pr.state == "open")
380 && state
381 .merge
382 .as_ref()
383 .is_some_and(|m| m.mode == crate::config::MergeMode::Pr && !m.empty)
384}
385
386pub(crate) fn can_resume(state: &RunState) -> bool {
391 let review = crate::graph::review_conclusion(&state.reviews, state.config.graph.review_rounds);
392 state.status.resumable()
393 && !state.released()
394 && state.retired_to.is_none()
395 && state.pr_taken_over_by.is_none()
396 && (resume_checkpoint(state)
397 || crate::run::try_home().is_some_and(|home| has_resume_checkpoint(state, &home))
398 || ((state.config.graph.review_rounds == 0
399 || state.reviews.iter().any(|round| round.clean))
400 && (!state.gate_ran || state.resource_block.is_some()))
401 || (state.resource_block.is_some() && review == Some(RunStatus::Gating))
402 || (review.is_none()
403 && state.reviews.last().is_some_and(|round| {
404 round.e2e_status() == crate::run::E2eStatus::ResourceBlocked
405 }))
406 || !(state.status == RunStatus::Blocked
407 && state.reviews.len() >= state.config.graph.review_rounds))
408}
409
410fn resume_checkpoint(state: &RunState) -> bool {
411 state.status == RunStatus::Landing
412 || landing_checkpoint(state)
413 || state.pr_publication_pending
414 || state.land_approval.is_some()
415 || state.merge_decision.is_some()
416 || state.land_armed_head.is_some()
417 || !state.merge_repairs.is_empty()
418}
419
420pub(crate) fn has_resume_checkpoint(state: &RunState, home: &Path) -> bool {
424 state.status.resumable()
425 && (resume_checkpoint(state)
426 || crate::ask::Questions::at(home.join("questions"))
427 .list()
428 .iter()
429 .any(|q| q.run == state.id && q.status.open()))
430}
431
432fn locked_holder(
435 home: &Path,
436 id: &str,
437 path: &Path,
438) -> Result<(RunState, usize, crate::pr_head::OwnerLock)> {
439 let owner = crate::pr_head::owner_lock(home, id)?;
440 let state = RunState::load_under(id, home)?;
441 let index = state
442 .candidates
443 .iter()
444 .position(|c| !c.folded && same_path(&c.worktree, path))
445 .ok_or_else(|| anyhow::anyhow!("worktree holder changed; retry takeover"))?;
446 anyhow::ensure!(
447 state.released_to.is_none(),
448 "run already handed over its worktree; retry takeover"
449 );
450 anyhow::ensure!(
451 path.parent()
452 .is_some_and(|p| same_path(&state.worktree_root(), p)),
453 "worktree holder root changed; retry takeover"
454 );
455 Ok((state, index, owner))
456}
457
458pub async fn release(
466 repo: &Path,
467 branch: &str,
468 new_run: &str,
469 takeover: &Takeover,
470) -> Result<Option<Released>> {
471 let Some(path) = git::worktree_holding(repo, branch).await? else {
472 return Ok(None);
473 };
474 let mut owner = None;
477 for id in &takeover.earlier {
478 let Ok(state) = RunState::load_under(id, &takeover.home) else {
479 continue;
480 };
481 if let Some(i) = state
482 .candidates
483 .iter()
484 .position(|c| !c.folded && same_path(&c.worktree, &path))
485 {
486 if !path
488 .parent()
489 .is_some_and(|b| same_path(&state.worktree_root(), b))
490 {
491 return Err(foreign(
492 branch,
493 &path,
494 &format!(
495 "run {} records it, but it is outside that run's worktree root, so \
496 magi did not make it",
497 crate::run::short_of(id)
498 ),
499 ));
500 }
501 owner = Some((state, i, Owner::EarlierAttempt));
502 break;
503 }
504 }
505 if owner.is_none() {
506 owner = match find_magi_owner(&path, &takeover.home) {
507 Ok(found) => found.map(|(s, i)| (s, i, Owner::MagiRun)),
508 Err(why) => return Err(foreign(branch, &path, &why)),
509 };
510 }
511 let Some((state, _, kind)) = owner else {
512 return Err(foreign(
513 branch,
514 &path,
515 &format!(
516 "{} is not a candidate worktree any magi run recorded (made by hand, or by \
517 something other than magi)",
518 path.display()
519 ),
520 ));
521 };
522
523 let (mut state, index, _owner) = locked_holder(&takeover.home, &state.id, &path)?;
524 if has_resume_checkpoint(&state, &takeover.home) {
525 return Err(Refused::Unsafe {
526 branch: branch.to_owned(),
527 path: path.display().to_string(),
528 why: format!("run {} has a resumable publication, landing or decision checkpoint; resume that run instead of releasing its worktree", state.short()),
529 }.into());
530 }
531 let holder = inspect(repo, branch, &path, &state, &takeover.home).await?;
532 if let Decision::Refuse(why) = decide(&kind, &holder) {
533 return Err(Refused::Unsafe {
534 branch: branch.to_owned(),
535 path: path.display().to_string(),
536 why,
537 }
538 .into());
539 }
540 let audit = format!(
541 "run {} (status `{}`, no driver, clean, HEAD {} = branch tip, branch `{branch}` kept)",
542 crate::run::short_of(&holder.run),
543 holder.status.as_str(),
544 short(&holder.head)
545 );
546
547 let old_id = state.id.clone();
548 state.candidates[index].folded = true;
549 state.released_to = Some(new_run.to_owned());
550 if !state.released_branches.iter().any(|b| b == branch) {
551 state.released_branches.push(branch.to_owned());
552 }
553 state.event(
554 "release",
555 format!(
556 "worktree of `{branch}` released to run {}: {audit}; this run can no longer be \
557 resumed from here",
558 crate::run::short_of(new_run)
559 ),
560 );
561 state.save_under(&takeover.home)?;
562
563 let again = match RunState::load_under(&old_id, &takeover.home) {
571 Ok(fresh) => inspect(repo, branch, &path, &fresh, &takeover.home).await,
572 Err(e) => Err(e),
573 };
574 let safe = matches!(&again, Ok(h) if decide(&kind, h) == Decision::Release);
575 let removed = safe
576 && git::worktree_remove_clean(repo, &path)
577 .await
578 .unwrap_or(false);
579 if !removed || path.exists() {
580 state = RunState::load_under(&old_id, &takeover.home)?;
581 state.candidates[index].folded = false;
582 state.released_to = None;
583 state.released_branches.retain(|b| b != branch);
584 state.events.pop();
585 state.event(
586 "release",
587 format!(
588 "release of `{branch}` to run {} was undone: {audit}",
589 crate::run::short_of(new_run)
590 ),
591 );
592 state.save_under(&takeover.home)?;
593 return Err(Refused::ReleaseFailed {
594 branch: branch.to_owned(),
595 path: path.display().to_string(),
596 run: crate::run::short_of(&old_id).to_owned(),
597 }
598 .into());
599 }
600 Ok(Some(Released {
601 old_id,
602 index,
603 path,
604 home: takeover.home.clone(),
605 successor: new_run.to_owned(),
606 tip: holder.tip,
607 audit,
608 }))
609}
610
611#[cfg(test)]
612mod tests {
613 #[test]
614 fn refused_display_is_the_english_text() {
615 let r = Refused::Foreign {
616 branch: "b".into(),
617 path: "/w".into(),
618 why: "w".into(),
619 };
620 assert_eq!(
621 r.to_string(),
622 "branch `b` is checked out in /w, which magi will not remove by itself: w. Remove \
623 that worktree (`git worktree remove`) if it is not needed, and try again."
624 );
625 }
626
627 use super::*;
628
629 fn holder() -> Holder {
630 Holder {
631 run: "20260901-000000-f82f".to_owned(),
632 status: RunStatus::Gating,
633 liveness: Liveness::Dead,
634 driver_unproven: false,
635 dirty: false,
636 dirty_files: Vec::new(),
637 driver_pid: None,
638 head: "a".repeat(40),
639 tip: "a".repeat(40),
640 }
641 }
642
643 fn recorded_holder(home: &Path) -> (RunState, PathBuf) {
644 let mut state = RunState::new(
645 "repo".into(),
646 "main".into(),
647 "base".into(),
648 "task".into(),
649 crate::config::Config::default(),
650 );
651 state.config.graph.worktree_root = Some(home.join("worktrees"));
652 let path = state.worktree_root().join("cand-A");
653 std::fs::create_dir_all(&path).unwrap();
654 state.driver_exited = true;
655 state.candidates.push(crate::run::Candidate {
656 index: 0,
657 label: 'A',
658 agent: "existing".into(),
659 branch: "branch".into(),
660 worktree: path.clone(),
661 summary: String::new(),
662 stat: String::new(),
663 files: 1,
664 commits: 1,
665 empty: false,
666 failed: None,
667 verified_noop: None,
668 duration_ms: 0,
669 folded: false,
670 });
671 state.save_under(home).unwrap();
672 (state, path)
673 }
674
675 #[test]
676 fn handover_reloads_a_driver_claim_after_the_scan_and_excludes_new_claims() {
677 let home = tempfile::tempdir().unwrap();
678 let (mut state, path) = recorded_holder(home.path());
679 let scanned = state.clone();
680 state.driver_exited = false;
682 state.driver_pid = Some(std::process::id());
683 state.driver_started_at = crate::proc::process_started_at(std::process::id());
684 state.begin_pass();
685 state.save_under(home.path()).unwrap();
686 let mut driver = crate::driver::DriverGuard::acquire(home.path(), &state).unwrap();
687 driver.record(&state).unwrap();
688 let (fresh, _, owner) = locked_holder(home.path(), &scanned.id, &path).unwrap();
689 assert!(!fresh.driver_exited);
690 assert_eq!(fresh.passes, state.passes);
691 assert_ne!(fresh.liveness_under(home.path(), false), Liveness::Dead);
692 assert!(crate::pr_head::owner_lock(home.path(), &state.id).is_err());
693 assert!(path.exists());
694 assert!(
695 RunState::load_under(&state.id, home.path())
696 .unwrap()
697 .released_to
698 .is_none()
699 );
700 drop(owner);
701 assert!(crate::pr_head::owner_lock(home.path(), &state.id).is_ok());
702 }
703
704 #[test]
705 fn handover_revalidates_candidate_ownership_under_the_claim_lock() {
706 let home = tempfile::tempdir().unwrap();
707 let (mut state, path) = recorded_holder(home.path());
708 state.candidates[0].folded = true;
709 state.released_to = Some("next".into());
710 state.save_under(home.path()).unwrap();
711 assert!(locked_holder(home.path(), &state.id, &path).is_err());
712 assert!(path.exists());
713 }
714
715 #[tokio::test]
716 async fn a_stale_restore_does_not_undo_a_later_handover() {
717 let home = tempfile::tempdir().unwrap();
718 let (mut state, path) = recorded_holder(home.path());
719 state.candidates[0].folded = true;
720 state.released_to = Some("later".into());
721 state.save_under(home.path()).unwrap();
722 let released = Released {
723 old_id: state.id.clone(),
724 index: 0,
725 path,
726 home: home.path().to_owned(),
727 successor: "original".into(),
728 tip: "tip".into(),
729 audit: String::new(),
730 };
731 released.restore(Path::new("nonexistent"), "branch").await;
732 let disk = RunState::load_under(&state.id, home.path()).unwrap();
733 assert_eq!(disk.released_to.as_deref(), Some("later"));
734 assert!(disk.candidates[0].folded);
735 }
736
737 #[test]
738 fn a_clean_superseded_stale_run_is_released() {
739 assert_eq!(decide(&Owner::EarlierAttempt, &holder()), Decision::Release);
740 let dead = Holder {
741 liveness: Liveness::Dead,
742 ..holder()
743 };
744 assert_eq!(decide(&Owner::EarlierAttempt, &dead), Decision::Release);
745 }
746
747 #[test]
748 fn a_foreign_worktree_is_refused_with_its_reason() {
749 let owner = Owner::Foreign("it is a foreign path".to_owned());
750 assert_eq!(
751 decide(&owner, &holder()),
752 Decision::Refuse("it is a foreign path".to_owned())
753 );
754 }
755
756 #[test]
757 fn landing_is_protected_even_for_an_earlier_attempt() {
758 let landing = Holder {
759 status: RunStatus::Landing,
760 ..holder()
761 };
762 assert!(matches!(
763 decide(&Owner::MagiRun, &landing),
764 Decision::Refuse(_)
765 ));
766 assert!(matches!(
767 decide(&Owner::EarlierAttempt, &landing),
768 Decision::Refuse(_)
769 ));
770 }
771
772 #[test]
773 fn a_dirty_worktree_is_refused_and_says_why() {
774 let dirty = Holder {
775 dirty: true,
776 dirty_files: vec!["scratch.txt".to_owned()],
777 ..holder()
778 };
779 let Decision::Refuse(why) = decide(&Owner::EarlierAttempt, &dirty) else {
780 panic!("dirty must be refused");
781 };
782 assert!(
783 why.contains("uncommitted") && why.contains("scratch.txt"),
784 "{why}"
785 );
786 assert!(why.contains("f82f") && why.contains("gating") && why.contains("dirty"));
787 }
788
789 #[test]
790 fn a_live_run_is_refused() {
791 let live = Holder {
792 liveness: Liveness::Live,
793 driver_pid: Some(4242),
794 ..holder()
795 };
796 let Decision::Refuse(why) = decide(&Owner::EarlierAttempt, &live) else {
797 panic!("live must be refused");
798 };
799 assert!(
800 why.contains("right now") && why.contains("f82f") && why.contains("4242"),
801 "{why}"
802 );
803 }
804
805 #[test]
806 fn unknown_without_a_recorded_pid_is_not_proof_of_a_stopped_driver() {
807 let holder = Holder {
808 liveness: Liveness::Unknown,
809 ..holder()
810 };
811 assert!(matches!(
812 decide(&Owner::EarlierAttempt, &holder),
813 Decision::Refuse(_)
814 ));
815 }
816
817 #[test]
818 fn a_driver_that_could_not_be_shown_dead_is_refused() {
819 let unproven = Holder {
820 driver_unproven: true,
821 ..holder()
822 };
823 let Decision::Refuse(why) = decide(&Owner::EarlierAttempt, &unproven) else {
824 panic!("an unproven driver must be refused");
825 };
826 assert!(why.contains("driver"), "{why}");
827 }
828
829 #[test]
830 fn a_head_off_the_tip_is_refused() {
831 let off = Holder {
832 head: "b".repeat(40),
833 ..holder()
834 };
835 assert!(matches!(
836 decide(&Owner::EarlierAttempt, &off),
837 Decision::Refuse(_)
838 ));
839 }
840}