Skip to main content

magi/
pr_recovery.rs

1//! Explicit recovery of a task's public implementation, independent of historical candidates.
2
3use std::path::Path;
4
5use anyhow::{Context, Result, ensure};
6use serde::{Deserialize, Serialize};
7
8use crate::{
9    config::{Config, MergeMode},
10    git, land,
11};
12
13/// Explicit purpose of an existing-PR recovery; never inferred from task prose.
14#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, clap::ValueEnum)]
15#[serde(rename_all = "snake_case")]
16pub enum Disposition {
17    /// Validate, publish to the selected PR and request fresh owner-approved landing.
18    Land,
19    /// Finish after local review, without public delivery.
20    Review,
21}
22
23/// Inputs of the single configuration load used to start a recovery.
24#[derive(Debug, Clone, Serialize, Deserialize)]
25pub struct ConfigEvidence {
26    /// Ordered configuration inputs; remote cache paths include the immutable source commit.
27    pub layers: Vec<std::path::PathBuf>,
28    /// Task options applied after loading the layers.
29    pub overrides: Option<crate::queue::RunOverrides>,
30    /// Daemon merge override, only when this task inherits daemon options.
31    pub daemon_merge: Option<String>,
32}
33
34/// Durable operator selection. Only the recovery command creates this record.
35#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
36pub struct Selection {
37    /// Existing pull request in this repository.
38    pub url: String,
39    /// Task-scoped public branch.
40    pub branch: String,
41    /// Initial public publication lease, independent of the implementation source.
42    pub head: String,
43    /// Full immutable source commit; older selections used the public lease.
44    #[serde(default)]
45    pub source_head: Option<String>,
46    /// Historical run explicitly selected to prove a launcher rebase.
47    #[serde(default)]
48    pub source_run: Option<String>,
49    /// Immutable proof, recomputed before starting or resuming work.
50    #[serde(default)]
51    pub source_proof: Option<Box<SourceProof>>,
52    /// Missing on old records: requires an explicit new selection before work.
53    #[serde(default)]
54    pub disposition: Option<Disposition>,
55    /// Criteria the operator approved, retained without rewriting them.
56    pub instruction: String,
57    /// Approval generation: a later selection does not resume an earlier run.
58    pub approved_at: jiff::Timestamp,
59}
60
61mod lineage;
62pub use lineage::SourceProof;
63
64const RETAIN_SOURCES: &str = "retain-recovery-sources";
65
66/// Preserve every historical run's source artifacts before releasing a recovery task.
67/// A sidecar survives stale run-state saves and applies even to older run schemas.
68pub fn retain_sources(task: &crate::queue::Task, home: &Path) -> Result<()> {
69    for id in task.runs.iter().collect::<std::collections::BTreeSet<_>>() {
70        ensure!(
71            crate::run::is_run_id(id),
72            "invalid historical run id; refusing source retention"
73        );
74        let dir = home.join("runs").join(id);
75        ensure!(
76            dir.is_dir(),
77            "historical run {id} is missing; cannot retain its sources"
78        );
79        std::fs::write(dir.join(RETAIN_SOURCES), &task.id)
80            .with_context(|| format!("retaining sources of historical run {id}"))?;
81    }
82    Ok(())
83}
84
85pub(crate) fn sources_retained(home: &Path, id: &str) -> bool {
86    home.join("runs")
87        .join(id)
88        .join(RETAIN_SOURCES)
89        .try_exists()
90        .unwrap_or(true)
91}
92
93/// A partially created recovery is retained and linked to its task for inspection.
94#[derive(Debug)]
95pub(crate) struct StartupFailed {
96    pub run: String,
97    pub cause: String,
98}
99
100impl std::fmt::Display for StartupFailed {
101    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
102        write!(
103            f,
104            "recovery run {} could not start: {}",
105            self.run, self.cause
106        )
107    }
108}
109impl std::error::Error for StartupFailed {}
110
111impl Selection {
112    /// Immutable implementation commit, distinct from the public update lease.
113    pub fn source(&self) -> &str {
114        self.source_head.as_deref().unwrap_or(&self.head)
115    }
116
117    /// Reconcile only the delivery controls; preserve all other resolved settings.
118    pub fn configure(&self, config: &mut Config) -> Result<()> {
119        match self.disposition {
120            Some(Disposition::Land) => {
121                ensure!(
122                    config.graph.reviewers > 0 && config.graph.review_rounds > 0,
123                    "delivery requires at least one reviewer and review round"
124                );
125                ensure!(
126                    !config.verify.gate.is_empty()
127                        && config.verify.gate.iter().all(|s| !s.trim().is_empty()),
128                    "delivery requires nonempty verify.gate commands; configure the repository's documented gate in a machine/repository layer or recover-pr --config <complete-config>"
129                );
130                ensure!(
131                    !config.verify.e2e.is_empty()
132                        && config.verify.e2e.iter().all(|s| !s.trim().is_empty()),
133                    "delivery requires nonempty verify.e2e commands; configure the repository's required real-machine checks before recover-pr --disposition land"
134                );
135                config.merge.mode = MergeMode::Pr;
136                config.graph.land = true;
137                config.graph.land_approval = true;
138            }
139            Some(Disposition::Review) => {
140                config.merge.mode = MergeMode::None;
141                config.graph.land = false;
142            }
143            None => anyhow::bail!(
144                "recovery disposition is missing; select the public head again with recover-pr --disposition land or --disposition review (and --source-head <full-SHA> to preserve a reviewed fix)"
145            ),
146        }
147        Ok(())
148    }
149
150    /// Prove the adopted fix contains the selected public implementation.
151    pub async fn verify_source(&self, repo: &Path) -> Result<()> {
152        if self.source_proof.is_some() {
153            self.verify_source_under(repo, &crate::run::home()).await
154        } else {
155            self.verify_source_under(repo, Path::new("")).await
156        }
157    }
158
159    async fn verify_source_under(&self, repo: &Path, home: &Path) -> Result<()> {
160        let source = git::rev_parse(repo, &format!("{}^{{commit}}", self.source())).await?;
161        ensure!(
162            source == self.source(),
163            "source-head must resolve to its full commit SHA"
164        );
165        if let Some(proof) = &self.source_proof {
166            // Historical state is large; keep replay out of the caller's future storage.
167            let actual = Box::pin(lineage::prove(self, repo, home, &proof.current_base)).await?;
168            ensure!(
169                &actual == proof.as_ref(),
170                "saved source proof changed; select the source again"
171            );
172        } else {
173            ensure!(
174                git::is_ancestor(repo, &self.head, &source).await,
175                "source-head does not preserve the selected public implementation by ancestry; supply --source-run <retained-run-id> for a recorded launcher rebase"
176            );
177            ensure!(
178                self.source_run.is_none(),
179                "source-run requires a saved source proof; select again"
180            );
181        }
182        Ok(())
183    }
184
185    /// Select and persist evidence without changing the implementation or public lease.
186    pub async fn prepare_source(
187        &mut self,
188        repo: &Path,
189        home: &Path,
190        current_base: &str,
191    ) -> Result<()> {
192        ensure!(
193            self.source_proof.is_none(),
194            "source proof is already selected; refusing to replace saved evidence"
195        );
196        if self.source_run.is_some() {
197            self.source_proof = Some(Box::new(
198                Box::pin(lineage::prove(self, repo, home, current_base)).await?,
199            ));
200            Ok(())
201        } else {
202            self.verify_source(repo).await
203        }
204    }
205
206    /// Confirm task identity and immutable criteria before starting work.
207    pub fn check_task(&self, task: &str, instruction: &str) -> Result<()> {
208        ensure!(
209            self.branch == format!("magi/task-{task}"),
210            "recovery head belongs to another task"
211        );
212        ensure!(
213            self.instruction == instruction,
214            "recovery criteria changed; select the public head again"
215        );
216        ensure!(
217            [self.head.as_str(), self.source()]
218                .iter()
219                .all(|sha| sha.len() == 40 && sha.bytes().all(|b| b.is_ascii_hexdigit())),
220            "recovery requires a full commit SHA"
221        );
222        Ok(())
223    }
224
225    /// Read-only forge validation plus fetch; no local branch is moved or pushed.
226    pub async fn verify(&self, repo: &Path, remote: &str, base: &str, lease: &str) -> Result<()> {
227        let pr_repo = land::slug_of_pr_url(&self.url).context("invalid recovery PR URL")?;
228        for args in [
229            vec!["remote", "get-url", "--all", remote],
230            vec!["remote", "get-url", "--push", "--all", remote],
231        ] {
232            let urls = git::git(repo, &args).await?;
233            ensure!(
234                !urls.trim().is_empty(),
235                "recovery remote has no repository URL"
236            );
237            for url in urls.lines() {
238                let remote_repo =
239                    remote_slug(url).context("cannot prove recovery remote repository identity")?;
240                land::ensure_same_repo(&remote_repo, &pr_repo)?;
241            }
242        }
243        let found = land::find_open_pr(repo, &self.branch, base).await?;
244        ensure!(
245            matches!(found, land::OpenPr::One { ref url, .. } if url == &self.url),
246            "selected PR is not the unique open task PR in this repository"
247        );
248        let (ok, raw) = land::gh(
249            repo,
250            &[
251                "pr".into(),
252                "view".into(),
253                self.url.clone(),
254                "--json".into(),
255                "headRefName,headRefOid,baseRefName,state,isCrossRepository".into(),
256            ],
257        )
258        .await?;
259        ensure!(ok, "could not verify selected PR: {raw}");
260        validate_view(&raw, &self.branch, base, lease)?;
261        let fetched = git::fetch(repo, remote, &self.branch).await?;
262        ensure!(
263            fetched.ok(),
264            "fetch selected public head failed: {}",
265            fetched.stderr
266        );
267        let fetched_head =
268            git::rev_parse(repo, &format!("refs/remotes/{remote}/{}", self.branch)).await?;
269        ensure!(
270            fetched_head == lease,
271            "public head moved; select and approve its new head before recovery"
272        );
273        ensure!(
274            git::remote_head(repo, remote, &self.branch)
275                .await?
276                .as_deref()
277                == Some(lease),
278            "public head lease changed during recovery"
279        );
280        Ok(())
281    }
282}
283
284// Require a recognizable forge URL rather than guessing through local paths or SSH aliases.
285fn remote_slug(url: &str) -> Option<String> {
286    let url = url
287        .trim()
288        .trim_end_matches('/')
289        .strip_suffix(".git")
290        .unwrap_or(url.trim().trim_end_matches('/'));
291    let (host, path) = if let Some(rest) = url
292        .strip_prefix("https://")
293        .or_else(|| url.strip_prefix("http://"))
294        .or_else(|| url.strip_prefix("ssh://"))
295    {
296        let (authority, path) = rest.split_once('/')?;
297        (authority.rsplit('@').next()?, path)
298    } else {
299        let (authority, path) = url.split_once(':')?;
300        let (_, host) = authority.split_once('@')?;
301        (host, path)
302    };
303    let mut parts = path.split('/');
304    let owner = parts.next()?;
305    let repo = parts.next()?;
306    if host.is_empty() || owner.is_empty() || repo.is_empty() || parts.next().is_some() {
307        return None;
308    }
309    Some(format!("{host}/{owner}/{repo}"))
310}
311
312fn validate_view(raw: &str, branch: &str, base: &str, lease: &str) -> Result<()> {
313    #[derive(Deserialize)]
314    #[serde(rename_all = "camelCase")]
315    struct View {
316        head_ref_name: String,
317        head_ref_oid: String,
318        base_ref_name: String,
319        state: String,
320        is_cross_repository: bool,
321    }
322    let view: View = serde_json::from_str(raw).context("selected public head metadata")?;
323    ensure!(
324        view.state == "OPEN"
325            && !view.is_cross_repository
326            && view.head_ref_name == branch
327            && view.base_ref_name == base,
328        "selected PR ownership or base changed"
329    );
330    ensure!(
331        view.head_ref_oid == lease,
332        "selected public head moved; refusing recovery"
333    );
334    Ok(())
335}
336
337#[cfg(test)]
338mod tests {
339    use super::*;
340
341    #[test]
342    fn recovery_uses_resolved_layers_and_preserves_unrelated_overrides() {
343        let dir = tempfile::tempdir().unwrap();
344        let machine = dir.path().join("machine.toml");
345        let repo = dir.path().join("repository.toml");
346        std::fs::write(&machine, "[verify]\ngate = ['documented gate']\ne2e = ['real checks']\n[graph]\nland = false\nland_approval = false\n").unwrap();
347        std::fs::write(&repo, "[merge]\nmode = 'none'\nbase = 'trunk'\n").unwrap();
348        let mut config = Config::load_layers(&[machine, repo]).unwrap();
349        let overrides = crate::queue::RunOverrides {
350            merge: Some("none".into()),
351            reviewers: Some(2),
352            ..Default::default()
353        };
354        overrides.apply(&mut config);
355        let mut selection = Selection {
356            url: "url".into(),
357            branch: "magi/task-t".into(),
358            head: "a".repeat(40),
359            source_head: None,
360            source_run: None,
361            source_proof: None,
362            disposition: Some(Disposition::Land),
363            instruction: "criteria".into(),
364            approved_at: jiff::Timestamp::now(),
365        };
366        selection.configure(&mut config).unwrap();
367        assert_eq!(config.merge.mode, MergeMode::Pr);
368        assert_eq!(config.graph.reviewers, 2);
369        assert_eq!(config.merge.base.as_deref(), Some("trunk"));
370        assert_eq!(config.verify.gate, ["documented gate"]);
371        assert_eq!(config.verify.e2e, ["real checks"]);
372        assert!(config.graph.land && config.graph.land_approval);
373        let explicit = dir.path().join("explicit.toml");
374        std::fs::write(&explicit, "[merge]\nbase = 'other'\n").unwrap();
375        let (mut replacement, paths) = Config::discover(dir.path(), Some(&explicit)).unwrap();
376        assert_eq!(paths, [explicit]);
377        assert!(
378            selection
379                .configure(&mut replacement)
380                .unwrap_err()
381                .to_string()
382                .contains("verify.gate")
383        );
384        replacement.verify.gate = vec![" ".into()];
385        assert!(selection.configure(&mut replacement).is_err());
386        selection.disposition = Some(Disposition::Review);
387        selection.configure(&mut replacement).unwrap();
388        assert_eq!(replacement.merge.mode, MergeMode::None);
389        assert!(!replacement.graph.land);
390    }
391
392    #[test]
393    fn selected_source_requires_the_same_fetch_and_push_repository() {
394        for remote in [
395            "https://example.test/team/project.git",
396            "git@example.test:team/project.git",
397            "ssh://git@example.test/team/project.git",
398        ] {
399            assert_eq!(
400                remote_slug(remote).as_deref(),
401                Some("example.test/team/project")
402            );
403        }
404        assert!(remote_slug("/tmp/another.git").is_none());
405        assert!(
406            land::ensure_same_repo(
407                &remote_slug("git@example.test:team/other.git").unwrap(),
408                "example.test/team/project"
409            )
410            .is_err()
411        );
412        assert!(
413            land::ensure_same_repo(
414                &remote_slug("git@other.test:team/project.git").unwrap(),
415                "example.test/team/project"
416            )
417            .is_err()
418        );
419    }
420
421    #[test]
422    fn selected_head_refuses_moved_or_foreign_public_heads() {
423        let raw = r#"{"headRefName":"magi/task-t","headRefOid":"aaa","baseRefName":"main","state":"OPEN","isCrossRepository":false}"#;
424        assert!(validate_view(raw, "magi/task-t", "main", "aaa").is_ok());
425        assert!(validate_view(raw, "magi/task-t", "main", "bbb").is_err());
426        assert!(validate_view(raw, "magi/task-other", "main", "aaa").is_err());
427        assert!(validate_view(raw, "magi/task-t", "other", "aaa").is_err());
428        assert!(
429            validate_view(&raw.replace("false", "true"), "magi/task-t", "main", "aaa").is_err()
430        );
431        assert!(
432            validate_view(&raw.replace("OPEN", "CLOSED"), "magi/task-t", "main", "aaa").is_err()
433        );
434    }
435
436    #[test]
437    fn selected_head_is_bound_to_authored_criteria_and_task() {
438        let selected = Selection {
439            url: "url".into(),
440            branch: "magi/task-t".into(),
441            head: "a".repeat(40),
442            source_head: None,
443            source_run: None,
444            source_proof: None,
445            disposition: Some(Disposition::Review),
446            instruction: "approved".into(),
447            approved_at: jiff::Timestamp::now(),
448        };
449        assert!(selected.check_task("t", "approved").is_ok());
450        assert!(
451            selected
452                .check_task("t", "prose mentioning another candidate")
453                .is_err()
454        );
455        assert!(selected.check_task("other", "approved").is_err());
456    }
457}