Skip to main content

memstead_base/engine/mutation/
parse_recovery.rs

1//! `Engine::apply_parse_recovery` — bulk-fix path that consumes the
2//! `ParsedRelationRecovery` payload on every writable-origin
3//! `PARSED_RELATION_INVALID` warning and applies the recovery action
4//! in a single operator-initiated call.
5//!
6//! The recovery action `remove_explicit_relation` means: drop the
7//! parse-time-dropped row from the source entity's markdown. The
8//! drop is already reflected in the in-memory `entity.relationships`
9//! (the parse-time validator strips the bad row at boot / reload /
10//! attach), so re-rendering and re-writing the source entity is
11//! enough — the renderer emits `## Relationships` from
12//! `entity.relationships`, so the stale row disappears.
13//!
14//! Multiple drops from the same source entity collapse to one
15//! re-render: `entity.relationships` already excludes every dropped
16//! row, so a single re-write fixes them all. The report still lists
17//! one entry per warning so consumers see exactly which drops were
18//! recovered.
19
20use indexmap::IndexMap;
21
22use crate::entity::EntityId;
23use crate::ops::{ParseRecoveryEntry, ParseRecoveryReport, WarningHint};
24use crate::vcs::{Actor, ClientId};
25
26use super::super::{Engine, EngineError, UpdateEntityArgs};
27
28impl Engine {
29    /// Walk `load_warnings`, dispatch the `remove_explicit_relation`
30    /// recovery for every writable-origin `PARSED_RELATION_INVALID`,
31    /// and report each entry on the response. Read-only-origin
32    /// warnings cannot be acted on (the engine has no write access
33    /// to their source markdown) and surface as
34    /// `outcome: "skipped"` with `reason: "readonly_mount"`.
35    ///
36    /// Failure model: per-entry failures land on the response as
37    /// `outcome: "failed"` with the underlying engine error code in
38    /// `reason`. The bulk-fix continues past per-entry failures so a
39    /// single bad source doesn't strand the rest of the batch. Only
40    /// engine-level errors (reload failure, broken workspace state)
41    /// abort the call and propagate via `Err`.
42    ///
43    /// Idempotency: after the per-source re-renders land, the method
44    /// runs `reload_each_writable_mem` so subsequent calls to
45    /// `health` / `load_warnings` reflect the post-recovery state.
46    /// Re-running on an already-clean workspace returns an empty
47    /// `entries` list with no commits.
48    pub fn apply_parse_recovery(
49        &mut self,
50        actor: Actor,
51        client: Option<&ClientId>,
52        note: Option<&str>,
53    ) -> Result<ParseRecoveryReport, EngineError> {
54        // Recovery's answer scope is every mem's parse drops, and a
55        // deferred (lazy, unloaded) mem records no load warnings until
56        // it loads — full load first, or its drops are invisible to
57        // this pass (load-scope/answer-scope rule, flywheel W7/01).
58        self.ensure_mems_loaded(None);
59        // Snapshot every `PARSED_RELATION_INVALID` warning. We
60        // iterate the snapshot, not `self.load_warnings`, so the
61        // mid-loop `update_entity` calls (which do not touch
62        // `load_warnings`) can't introduce ordering surprises.
63        struct Drop {
64            entity_id: EntityId,
65            rel_type: String,
66            target: EntityId,
67            origin: String,
68        }
69        let drops: Vec<Drop> = self
70            .load_warnings()
71            .iter()
72            .filter_map(|w| match w {
73                WarningHint::ParsedRelationInvalid {
74                    entity_id,
75                    rel_type,
76                    target,
77                    origin,
78                    ..
79                } => Some(Drop {
80                    entity_id: entity_id.clone(),
81                    rel_type: rel_type.clone(),
82                    target: target.clone(),
83                    origin: origin.clone(),
84                }),
85                _ => None,
86            })
87            .collect();
88
89        // Group writable drops by source-entity id so each source is
90        // re-rendered at most once. Iteration over the IndexMap
91        // preserves the order the warnings appeared in.
92        let mut writable_by_source: IndexMap<EntityId, Vec<usize>> = IndexMap::new();
93        let mut readonly_indices: Vec<usize> = Vec::new();
94        for (idx, drop) in drops.iter().enumerate() {
95            if drop.origin == "writable" {
96                writable_by_source
97                    .entry(drop.entity_id.clone())
98                    .or_default()
99                    .push(idx);
100            } else {
101                readonly_indices.push(idx);
102            }
103        }
104
105        let mut entries: Vec<ParseRecoveryEntry> = Vec::with_capacity(drops.len());
106        // Per-drop result slot — populated as the per-source attempts
107        // land. Keyed by the snapshot index so the final `entries`
108        // vec preserves the warning order.
109        let mut result_per_drop: Vec<Option<(String, Option<String>)>> = vec![None; drops.len()];
110
111        for idx in &readonly_indices {
112            result_per_drop[*idx] = Some((
113                ParseRecoveryEntry::OUTCOME_SKIPPED.to_string(),
114                Some(ParseRecoveryEntry::REASON_READONLY_MOUNT.to_string()),
115            ));
116        }
117
118        let mut last_commit_sha = String::new();
119        for (source_id, drop_indices) in writable_by_source {
120            let outcome = self.rewrite_for_parse_recovery(&source_id, actor, client, note);
121            match outcome {
122                Ok(commit_sha) => {
123                    if !commit_sha.is_empty() {
124                        last_commit_sha = commit_sha;
125                    }
126                    for idx in drop_indices {
127                        result_per_drop[idx] =
128                            Some((ParseRecoveryEntry::OUTCOME_REMOVED.to_string(), None));
129                    }
130                }
131                Err(err) => {
132                    let code = err.code().to_string();
133                    for idx in drop_indices {
134                        result_per_drop[idx] = Some((
135                            ParseRecoveryEntry::OUTCOME_FAILED.to_string(),
136                            Some(code.clone()),
137                        ));
138                    }
139                }
140            }
141        }
142
143        // Materialise entries in the original warning order.
144        for (idx, drop) in drops.into_iter().enumerate() {
145            let (outcome, reason) = result_per_drop[idx]
146                .take()
147                .expect("every drop should have been classified");
148            entries.push(ParseRecoveryEntry {
149                entity_id: drop.entity_id,
150                rel_type: drop.rel_type,
151                target: drop.target,
152                outcome,
153                reason,
154            });
155        }
156
157        // Reload writable mems so `load_warnings` reflects the
158        // post-recovery state — drops that re-rendered cleanly drop
159        // out; drops that failed (or read-only ones that were always
160        // out of scope) survive.
161        if !entries.is_empty() {
162            self.reload_each_writable_mem()?;
163        }
164
165        Ok(ParseRecoveryReport {
166            entries,
167            commit_sha: last_commit_sha,
168        })
169    }
170
171    /// Re-render the source entity and write it back to disk. Calls
172    /// `update_entity` seeding the first section's current body as a
173    /// rewrite anchor — section content stays identical, but the
174    /// full re-render flushes the parse-time relation drops out of
175    /// the auto-managed `## Relationships` section. Returns the
176    /// resulting `commit_sha` on success.
177    ///
178    /// Section-anchor seeding (instead of an empty payload) keeps
179    /// this internal rewrite path on the public `update_entity`
180    /// surface — the agent-boundary `EMPTY_UPDATE` guard refuses
181    /// empty payloads, and re-using the same guard avoids splitting the engine's
182    /// validation surface for one internal caller. Picking the
183    /// first section is safe because real entities always carry at
184    /// least one schema-required section (a stub would have tripped
185    /// the `StubNotUpdatable` guard before reaching here).
186    fn rewrite_for_parse_recovery(
187        &mut self,
188        source_id: &EntityId,
189        actor: Actor,
190        client: Option<&ClientId>,
191        note: Option<&str>,
192    ) -> Result<String, EngineError> {
193        let entity = self
194            .store()
195            .get(source_id)
196            .ok_or_else(|| EngineError::NotFound {
197                id: source_id.to_string(),
198            })?;
199        let expected_hash = entity.content_hash.clone();
200        let mut sections: IndexMap<String, String> = IndexMap::new();
201        if let Some((key, body)) = entity.sections.iter().next() {
202            sections.insert(key.clone(), body.clone());
203        }
204        let args = UpdateEntityArgs {
205            anchors: Vec::new(),
206            id: source_id.clone(),
207            expected_hash: Some(expected_hash),
208            sections,
209            append_sections: IndexMap::new(),
210            patch_sections: IndexMap::new(),
211            metadata: IndexMap::new(),
212            metadata_unset: Vec::new(),
213            dry_run: false,
214            declare_relations: Vec::new(),
215            relations_unset: Vec::new(),
216            anchors_unset: Vec::new(),
217        };
218        let outcome = self.update_entity(args, actor, client, note)?;
219        Ok(outcome.commit_sha)
220    }
221}
222
223#[cfg(test)]
224mod tests {
225    use tempfile::TempDir;
226
227    use crate::backend::MemBackend;
228    use crate::engine::Engine;
229    use crate::engine::test_helpers::{
230        archive_mount, build_archive, cli_actor, folder_mount, write_schema_files_with_default_type,
231    };
232    use crate::ops::{ParseRecoveryEntry, WarningHint};
233    use crate::storage::{ArchiveBackend, FilesystemMemWriter};
234    use crate::workspace::{Mount, MountCapability, MountLifecycle, MountStorage};
235
236    use memstead_schema::SchemaRef;
237
238    /// Two writable parse-time drops on the same source collapse to
239    /// one re-render. Both entries land on the response as
240    /// `removed`; the on-disk markdown no longer carries the bad
241    /// rows; `load_warnings` is empty after the call.
242    #[test]
243    fn apply_parse_recovery_clears_writable_drops_in_one_call() {
244        let tmp = TempDir::new().unwrap();
245        let mem_dir = tmp.path().to_path_buf();
246        let target = "---\ntype: spec\n---\n# Target\n\n## Identity\n\nTarget body.\n";
247        let source = "---\ntype: spec\n---\n# Source\n\n## Identity\n\nSource body.\n\n## Relationships\n\n- **MADE_UP_TYPE_A**: [[specs--target]]\n- **MADE_UP_TYPE_B**: [[specs--target]]\n";
248        std::fs::write(mem_dir.join("target.md"), target).unwrap();
249        std::fs::write(mem_dir.join("source.md"), source).unwrap();
250
251        let writer = FilesystemMemWriter::new(mem_dir.clone());
252        let mut engine = Engine::from_mounts(vec![(
253            folder_mount("specs", mem_dir.clone()),
254            Box::new(writer) as Box<dyn MemBackend>,
255        )])
256        .unwrap();
257
258        let pre: Vec<_> = engine
259            .load_warnings()
260            .iter()
261            .filter(|w| matches!(w, WarningHint::ParsedRelationInvalid { .. }))
262            .collect();
263        assert_eq!(pre.len(), 2, "expected two parse-time drops, got {pre:?}");
264
265        let (actor, client) = cli_actor();
266        let report = engine
267            .apply_parse_recovery(actor, Some(&client), Some("recovery"))
268            .expect("recovery succeeds");
269
270        assert_eq!(report.entries.len(), 2);
271        for entry in &report.entries {
272            assert_eq!(
273                entry.outcome,
274                ParseRecoveryEntry::OUTCOME_REMOVED,
275                "expected both writable drops removed, got {entry:?}",
276            );
277            assert!(entry.reason.is_none());
278        }
279        assert!(!report.commit_sha.is_empty(), "recovery must commit");
280
281        let post: Vec<_> = engine
282            .load_warnings()
283            .iter()
284            .filter(|w| matches!(w, WarningHint::ParsedRelationInvalid { .. }))
285            .collect();
286        assert!(post.is_empty(), "drops must be cleared, got {post:?}");
287
288        let cleaned = std::fs::read_to_string(mem_dir.join("source.md")).unwrap();
289        assert!(
290            !cleaned.contains("MADE_UP_TYPE_A"),
291            "cleaned source: {cleaned}"
292        );
293        assert!(
294            !cleaned.contains("MADE_UP_TYPE_B"),
295            "cleaned source: {cleaned}"
296        );
297    }
298
299    /// Recover leg (criterion 5): `apply_parse_recovery` re-renders the
300    /// source entity but leaves its anchors sidecar bit-intact — the
301    /// re-render is an anchorless update, which never stages the sidecar.
302    #[test]
303    fn apply_parse_recovery_leaves_anchors_bit_intact() {
304        let tmp = TempDir::new().unwrap();
305        let mem_dir = tmp.path().to_path_buf();
306        let target = "---\ntype: spec\n---\n# Target\n\n## Identity\n\nTarget body.\n";
307        let source = "---\ntype: spec\n---\n# Source\n\n## Identity\n\nSource body.\n\n## Relationships\n\n- **MADE_UP_TYPE**: [[specs--target]]\n";
308        std::fs::write(mem_dir.join("target.md"), target).unwrap();
309        std::fs::write(mem_dir.join("source.md"), source).unwrap();
310
311        // Seed an anchors sidecar for the source entity directly on disk (the
312        // recovery path must not touch it).
313        std::fs::create_dir_all(mem_dir.join(".memstead")).unwrap();
314        let sidecar_path = mem_dir.join(".memstead").join("anchors.json");
315        let sidecar = br#"{"version":1,"entities":{"specs--source":[{"artifact":"src/lib.rs","grain":"file","class":"anchored","hash_stability":"stable","hash":"h1"}]}}"#;
316        std::fs::write(&sidecar_path, sidecar).unwrap();
317        let before = std::fs::read(&sidecar_path).unwrap();
318
319        let writer = FilesystemMemWriter::new(mem_dir.clone());
320        let mut engine = Engine::from_mounts(vec![(
321            folder_mount("specs", mem_dir.clone()),
322            Box::new(writer) as Box<dyn MemBackend>,
323        )])
324        .unwrap();
325
326        let (actor, client) = cli_actor();
327        let report = engine
328            .apply_parse_recovery(actor, Some(&client), Some("recovery"))
329            .expect("recovery succeeds");
330        assert_eq!(report.entries.len(), 1);
331        assert_eq!(
332            report.entries[0].outcome,
333            ParseRecoveryEntry::OUTCOME_REMOVED
334        );
335
336        // The sidecar file is byte-identical, and the anchor still resolves.
337        let after = std::fs::read(&sidecar_path).unwrap();
338        assert_eq!(before, after, "recovery must leave anchors bit-intact");
339        let anchors = engine.entity_anchors(&crate::EntityId::new("specs", "source"));
340        assert_eq!(anchors.len(), 1);
341        assert_eq!(anchors[0].artifact, "src/lib.rs");
342    }
343
344    /// Read-only-origin drops are reported as `skipped` with
345    /// `reason: "readonly_mount"`. The engine cannot rewrite an
346    /// archive, so the warning survives the call.
347    #[test]
348    fn apply_parse_recovery_skips_readonly_origin_drops() {
349        let tmp = TempDir::new().unwrap();
350        let target = "---\ntype: spec\n---\n# Target\n\n## Identity\n\nTarget.\n";
351        let source = "---\ntype: spec\n---\n# Source\n\n## Identity\n\nSource.\n\n## Relationships\n\n- **MADE_UP**: [[external--target]]\n";
352        let archive_path = build_archive(
353            tmp.path(),
354            "ext",
355            &[
356                ("target.md", target.as_bytes()),
357                ("source.md", source.as_bytes()),
358            ],
359        );
360
361        let mut engine = Engine::from_mounts(vec![(
362            archive_mount("external", archive_path.clone()),
363            Box::new(ArchiveBackend::new(archive_path)),
364        )])
365        .unwrap();
366
367        let (actor, client) = cli_actor();
368        let report = engine
369            .apply_parse_recovery(actor, Some(&client), None)
370            .expect("recovery succeeds");
371
372        assert_eq!(report.entries.len(), 1);
373        let entry = &report.entries[0];
374        assert_eq!(entry.outcome, ParseRecoveryEntry::OUTCOME_SKIPPED);
375        assert_eq!(
376            entry.reason.as_deref(),
377            Some(ParseRecoveryEntry::REASON_READONLY_MOUNT),
378        );
379        assert!(
380            report.commit_sha.is_empty(),
381            "readonly path commits nothing"
382        );
383
384        let post: Vec<_> = engine
385            .load_warnings()
386            .iter()
387            .filter(|w| matches!(w, WarningHint::ParsedRelationInvalid { .. }))
388            .collect();
389        assert_eq!(post.len(), 1, "readonly drop must persist, got {post:?}");
390    }
391
392    /// Re-running on an already-clean workspace returns an empty
393    /// entries list and produces no commits.
394    #[test]
395    fn apply_parse_recovery_is_idempotent_after_clean_state() {
396        let tmp = TempDir::new().unwrap();
397        let mem_dir = tmp.path().to_path_buf();
398        let target = "---\ntype: spec\n---\n# Target\n\n## Identity\n\nTarget.\n";
399        let source = "---\ntype: spec\n---\n# Source\n\n## Identity\n\nSource.\n\n## Relationships\n\n- **MADE_UP_TYPE**: [[specs--target]]\n";
400        std::fs::write(mem_dir.join("target.md"), target).unwrap();
401        std::fs::write(mem_dir.join("source.md"), source).unwrap();
402
403        let writer = FilesystemMemWriter::new(mem_dir.clone());
404        let mut engine = Engine::from_mounts(vec![(
405            folder_mount("specs", mem_dir),
406            Box::new(writer) as Box<dyn MemBackend>,
407        )])
408        .unwrap();
409
410        let (actor, client) = cli_actor();
411        let first = engine
412            .apply_parse_recovery(actor, Some(&client), None)
413            .expect("first recovery succeeds");
414        assert_eq!(first.entries.len(), 1);
415        assert_eq!(
416            first.entries[0].outcome,
417            ParseRecoveryEntry::OUTCOME_REMOVED
418        );
419        assert!(!first.commit_sha.is_empty());
420
421        let second = engine
422            .apply_parse_recovery(actor, Some(&client), None)
423            .expect("second recovery succeeds");
424        assert!(
425            second.entries.is_empty(),
426            "second call must be no-op, got {:?}",
427            second.entries
428        );
429        assert!(second.commit_sha.is_empty());
430    }
431
432    /// Mixed writable + readonly drops land in a single report.
433    #[test]
434    fn apply_parse_recovery_reports_per_warning_across_origins() {
435        let tmp = TempDir::new().unwrap();
436
437        let writable_dir = tmp.path().join("writable");
438        std::fs::create_dir_all(&writable_dir).unwrap();
439        let w_target = "---\ntype: spec\n---\n# WT\n\n## Identity\n\nwt\n";
440        let w_source = "---\ntype: spec\n---\n# WS\n\n## Identity\n\nws\n\n## Relationships\n\n- **MADE_UP_A**: [[specs--target]]\n- **MADE_UP_B**: [[specs--target]]\n";
441        std::fs::write(writable_dir.join("target.md"), w_target).unwrap();
442        std::fs::write(writable_dir.join("source.md"), w_source).unwrap();
443
444        let r_target = "---\ntype: spec\n---\n# RT\n\n## Identity\n\nrt\n";
445        let r_source = "---\ntype: spec\n---\n# RS\n\n## Identity\n\nrs\n\n## Relationships\n\n- **MADE_UP_RO**: [[external--target]]\n";
446        let archive_path = build_archive(
447            tmp.path(),
448            "ext",
449            &[
450                ("target.md", r_target.as_bytes()),
451                ("source.md", r_source.as_bytes()),
452            ],
453        );
454
455        let writer = FilesystemMemWriter::new(writable_dir.clone());
456        let mut engine = Engine::from_mounts(vec![
457            (
458                folder_mount("specs", writable_dir),
459                Box::new(writer) as Box<dyn MemBackend>,
460            ),
461            (
462                archive_mount("external", archive_path.clone()),
463                Box::new(ArchiveBackend::new(archive_path)),
464            ),
465        ])
466        .unwrap();
467
468        let (actor, client) = cli_actor();
469        let report = engine
470            .apply_parse_recovery(actor, Some(&client), None)
471            .expect("recovery succeeds");
472
473        assert_eq!(report.entries.len(), 3);
474        let removed: Vec<_> = report
475            .entries
476            .iter()
477            .filter(|e| e.outcome == ParseRecoveryEntry::OUTCOME_REMOVED)
478            .collect();
479        let skipped: Vec<_> = report
480            .entries
481            .iter()
482            .filter(|e| e.outcome == ParseRecoveryEntry::OUTCOME_SKIPPED)
483            .collect();
484        assert_eq!(removed.len(), 2);
485        assert_eq!(skipped.len(), 1);
486        assert_eq!(
487            skipped[0].reason.as_deref(),
488            Some(ParseRecoveryEntry::REASON_READONLY_MOUNT),
489        );
490        assert!(!report.commit_sha.is_empty());
491    }
492
493    /// A drop whose source still has an unresolved body wiki-link to
494    /// the dropped target lands as `failed` with
495    /// `WIKILINK_WITHOUT_RELATION`. The strict validator refuses to
496    /// leave a body wiki-link unbacked by any relation; the
497    /// operator's recovery is to also remove the body reference.
498    #[test]
499    fn apply_parse_recovery_reports_failed_for_unbacked_body_link() {
500        let tmp = TempDir::new().unwrap();
501        let schemas_dir = tmp.path().join("schemas");
502        std::fs::create_dir_all(&schemas_dir).unwrap();
503        let manifest = r#"name: link-test
504version: 0.1.0
505description: schema for wikilink-blocker test
506when_to_use: tests
507types:
508  - doc
509relationships:
510  mode: strict
511  definitions:
512    - name: MENTIONS
513      description: doc references doc
514      default_weight: 1.0
515    - name: _default
516      description: fallback
517      default_weight: 1.0
518community:
519  resolution: 1.0
520  seed: 42
521"#;
522        write_schema_files_with_default_type(&schemas_dir, "link-test", manifest, &["doc"]);
523
524        let mem_dir = tmp.path().join("mem");
525        std::fs::create_dir_all(&mem_dir).unwrap();
526        let target = "---\ntype: doc\n---\n# Target\n\n## Body\n\nbody\n";
527        let source = "---\ntype: doc\n---\n# Source\n\n## Body\n\nrefer to [[specs--target]] here\n\n## Relationships\n\n- **BADTYPE**: [[specs--target]]\n";
528        std::fs::write(mem_dir.join("target.md"), target).unwrap();
529        std::fs::write(mem_dir.join("source.md"), source).unwrap();
530
531        let writer = FilesystemMemWriter::new(mem_dir.clone());
532        let pin = SchemaRef::new("link-test", semver::Version::new(0, 1, 0));
533        let mount = Mount {
534            mem: "specs".to_string(),
535            schema: Some(pin),
536            storage: MountStorage::Folder {
537                path: mem_dir.clone(),
538            },
539            capability: MountCapability::Write,
540            lifecycle: MountLifecycle::Eager,
541            cross_linkable: true,
542            migration_target: None,
543        };
544        let mut engine = Engine::from_mounts_with_schemas_dir(
545            vec![(mount, Box::new(writer) as Box<dyn MemBackend>)],
546            Some(&schemas_dir),
547        )
548        .unwrap();
549
550        let (actor, client) = cli_actor();
551        let report = engine
552            .apply_parse_recovery(actor, Some(&client), None)
553            .expect("recovery returns Ok even when entries fail");
554
555        assert_eq!(report.entries.len(), 1);
556        let entry = &report.entries[0];
557        assert_eq!(entry.outcome, ParseRecoveryEntry::OUTCOME_FAILED);
558        assert_eq!(
559            entry.reason.as_deref(),
560            Some("WIKILINK_WITHOUT_RELATION"),
561            "expected the strict validator's typed code, got {:?}",
562            entry.reason,
563        );
564        let unchanged = std::fs::read_to_string(mem_dir.join("source.md")).unwrap();
565        assert!(
566            unchanged.contains("BADTYPE"),
567            "source must be unchanged on failure"
568        );
569
570        let post: Vec<_> = engine
571            .load_warnings()
572            .iter()
573            .filter(|w| matches!(w, WarningHint::ParsedRelationInvalid { .. }))
574            .collect();
575        assert_eq!(post.len(), 1, "failed drop must persist, got {post:?}");
576    }
577}