dbmd_core/log.rs
1//! `log` — the append-only, month-rotating chronological log.
2//!
3//! One logical timeline: the active `log.md` at the store root plus
4//! `log/<YYYY-MM>.md` archives. [`Log::append`] rolls older months into
5//! archives on write so the active file stays current-month. [`Log::tail`] and
6//! [`Log::since`] **reverse-read from EOF**. Both read each file they touch in
7//! full — the on-disk order is not guaranteed monotonic, so neither can
8//! early-stop within a file — and select by timestamp: `tail` keeps the `n`
9//! newest, `since` keeps everything newer than the cutoff. Both cross into
10//! month archives only as far back as the requested window reaches (by the
11//! cutoff's month for `since`, by the current `n`th-newest's month for `tail`)
12//! — never the whole history.
13//!
14//! Append-only contract: there is no rewrite API. Corrective entries go on the
15//! end; out-of-order timestamps are a validate warning (`LOG_OUT_OF_ORDER`),
16//! signalling a probable rewrite.
17
18use std::collections::BTreeMap;
19use std::fs::{self, File};
20use std::io::{Read, Seek, SeekFrom};
21use std::path::{Path, PathBuf};
22
23use chrono::{DateTime, Datelike, FixedOffset, NaiveDateTime, TimeZone, Utc};
24
25use crate::store::Store;
26
27/// The on-disk header timestamp format: `YYYY-MM-DD HH:MM` (minute precision,
28/// no timezone). Parsing reattaches UTC; emitting renders the entry's own
29/// wall-clock, so a read→write→read round-trip is stable at minute precision.
30const TS_FORMAT: &str = "%Y-%m-%d %H:%M";
31
32/// The frontmatter block written when the active `log.md` is created.
33const LOG_FRONTMATTER: &str = "---\ntype: log\n---\n\n# Curator log\n";
34
35/// Block size for the backward (reverse-from-EOF) reader.
36const REVERSE_BLOCK: usize = 8 * 1024;
37
38/// A recognized `log.md` entry kind. Custom kinds are valid in the format
39/// (`dbmd validate` warns on unrecognized via `LOG_UNKNOWN_KIND`); this enum
40/// carries the recognized vocabulary plus a [`LogKind::Custom`] catch-all so an
41/// unknown kind round-trips without loss.
42#[derive(Debug, Clone, PartialEq, Eq)]
43pub enum LogKind {
44 /// A source artifact was ingested.
45 Ingest,
46 /// A file was created.
47 Create,
48 /// A file was updated.
49 Update,
50 /// A file was deleted.
51 Delete,
52 /// A file was renamed/moved.
53 Rename,
54 /// A wiki-link was added.
55 Link,
56 /// A validation pass ran.
57 Validate,
58 /// The index was rebuilt.
59 IndexRebuild,
60 /// A contradiction between sources was flagged.
61 Contradiction,
62 /// Any kind outside the recognized vocabulary, preserved verbatim.
63 Custom(String),
64}
65
66impl LogKind {
67 /// The canonical lowercase string for this kind, as it appears in a log
68 /// header (`ingest`, `index-rebuild`, …).
69 pub fn as_str(&self) -> &str {
70 match self {
71 LogKind::Ingest => "ingest",
72 LogKind::Create => "create",
73 LogKind::Update => "update",
74 LogKind::Delete => "delete",
75 LogKind::Rename => "rename",
76 LogKind::Link => "link",
77 LogKind::Validate => "validate",
78 LogKind::IndexRebuild => "index-rebuild",
79 LogKind::Contradiction => "contradiction",
80 LogKind::Custom(s) => s,
81 }
82 }
83
84 /// Parse a kind from its header token; non-canonical tokens become
85 /// [`LogKind::Custom`].
86 pub fn parse(token: &str) -> LogKind {
87 match token {
88 "ingest" => LogKind::Ingest,
89 "create" => LogKind::Create,
90 "update" => LogKind::Update,
91 "delete" => LogKind::Delete,
92 "rename" => LogKind::Rename,
93 "link" => LogKind::Link,
94 "validate" => LogKind::Validate,
95 "index-rebuild" => LogKind::IndexRebuild,
96 "contradiction" => LogKind::Contradiction,
97 other => LogKind::Custom(other.to_string()),
98 }
99 }
100
101 /// True if this is one of the recognized kinds (i.e. not
102 /// [`LogKind::Custom`]).
103 pub fn is_recognized(&self) -> bool {
104 !matches!(self, LogKind::Custom(_))
105 }
106}
107
108/// One parsed `log.md` entry: a header
109/// (`## [YYYY-MM-DD HH:MM] <kind> | <object>`) plus its body.
110#[derive(Debug, Clone, PartialEq, Eq)]
111pub struct LogEntry {
112 /// The entry timestamp from the header.
113 pub timestamp: DateTime<FixedOffset>,
114 /// The entry kind.
115 pub kind: LogKind,
116 /// The object slot — a store-relative path/wiki-link target, or `None` for
117 /// store-wide actions like `validate`.
118 pub object: Option<String>,
119 /// The free-form body (one or more lines) explaining what happened.
120 pub note: String,
121}
122
123impl LogEntry {
124 /// Render this entry as it appears on disk: the `## [...]` header line,
125 /// then the note body, then a trailing blank line so successive entries are
126 /// separated. The note is emitted with header-shaped continuation lines
127 /// **escaped** (see [`escape_note_line`]) so a note line that happens to
128 /// match the entry-header shape (`## [YYYY-MM-DD HH:MM] <kind> | <obj>`) can
129 /// never be mistaken for a real entry header on readback or on the next
130 /// rotation. The escape round-trips exactly through [`unescape_note_line`].
131 fn render(&self) -> String {
132 let ts = self.timestamp.format(TS_FORMAT);
133 let mut out = String::new();
134 match &self.object {
135 Some(obj) => {
136 out.push_str(&format!("## [{}] {} | {}\n", ts, self.kind.as_str(), obj));
137 }
138 None => {
139 out.push_str(&format!("## [{}] {}\n", ts, self.kind.as_str()));
140 }
141 }
142 // Trim only the structural line terminators (`\n`/`\r`) — the trailing
143 // blank line separating entries is appended below, so a note's own
144 // trailing newlines would otherwise stack up and shift on every
145 // re-render. Spaces and tabs are legitimate note *content* and must be
146 // preserved verbatim, so the round-trip is exact: readback
147 // (`parse_entries`) trims the same `['\n', '\r']` set and no more, and a
148 // note ending in a space (`"note 0 "`) must reconstruct unchanged.
149 let note = self.note.trim_end_matches(['\n', '\r']);
150 if !note.is_empty() {
151 // Escape per line: a note line that parses as an entry header is
152 // prefixed so it is no longer at column 0 as `## [` — it stays note
153 // body on readback and on rotation, never a fabricated entry.
154 for (i, line) in note.split('\n').enumerate() {
155 if i > 0 {
156 out.push('\n');
157 }
158 out.push_str(&escape_note_line(line));
159 }
160 out.push('\n');
161 }
162 out.push('\n');
163 out
164 }
165
166 /// The `(year, month)` of this entry's wall-clock timestamp — the rotation
167 /// bucket.
168 fn year_month(&self) -> (i32, u32) {
169 (self.timestamp.year(), self.timestamp.month())
170 }
171}
172
173/// The store's chronological log: a thin handle for the append-only timeline.
174/// All methods take the [`Store`] so they resolve the active `log.md` and the
175/// `log/` archives under the store root.
176#[derive(Debug, Clone)]
177pub struct Log;
178
179impl Log {
180 /// Atomically append `entry` to the active `log.md`, creating it (with
181 /// `type: log` frontmatter) if absent. **If the active log holds entries
182 /// from a prior month, roll those older months into `log/<YYYY-MM>.md`
183 /// first** (atomic move), keeping the active file to the current month.
184 ///
185 /// **Concurrency.** `append` is a read-modify-write of the whole active file
186 /// (`write_atomic` is atomic at the file level, but the read→render→write
187 /// window is not). Two concurrent appenders — the manager and a cron-driven
188 /// background system, say — would otherwise both read the same N-entry
189 /// snapshot and each write N+1 entries, the second rename clobbering the
190 /// first and silently dropping an audit entry. We serialize the whole
191 /// read-modify-write under an advisory file lock (`flock`, held for the
192 /// duration) so concurrent appends queue instead of racing. The lock is
193 /// advisory and process-scoped; it guards the toolkit's own appends, which is
194 /// the realistic contention path.
195 pub fn append(store: &Store, entry: &LogEntry) -> crate::Result<()> {
196 let active = active_log_path(store);
197
198 // Serialize concurrent appends for the whole read-modify-write. Held
199 // until `_lock` drops at function exit (covering both the rotation and
200 // the plain-append paths). A lock failure is non-fatal: we proceed
201 // unlocked rather than refuse to log (best-effort, same posture as the
202 // pre-fix behaviour on platforms without advisory locks).
203 let _lock = AppendLock::acquire(&active);
204
205 // Read the active file's current contents (if any). The "current month"
206 // is the month of the entry being appended (the newest in the timeline);
207 // every existing entry from a strictly-earlier month rolls to archives.
208 let current_ym = entry.year_month();
209
210 if active.exists() {
211 let content = fs::read_to_string(&active)?;
212 let (header, entries) = parse_active(&content);
213
214 // Partition existing entries into prior-month (roll out) and
215 // current-or-later (keep in the active file).
216 let mut by_month: BTreeMap<(i32, u32), Vec<LogEntry>> = BTreeMap::new();
217 let mut keep: Vec<LogEntry> = Vec::new();
218 for e in entries {
219 if e.year_month() < current_ym {
220 by_month.entry(e.year_month()).or_default().push(e);
221 } else {
222 keep.push(e);
223 }
224 }
225
226 // A rotation is two non-atomic durable writes (archive append, then
227 // active trim). The marker disambiguates a crash-retry re-roll from a
228 // fresh rotation so a genuinely-distinct same-minute entry is never
229 // dropped (see `rotation_marker_path`). `recovering` is captured
230 // BEFORE we (re)write the marker, so the current attempt's archive
231 // append uses the right mode; the marker only changes what a LATER
232 // retry sees.
233 let marker = rotation_marker_path(store);
234 let recovering = marker.exists();
235
236 if !by_month.is_empty() {
237 // Roll each prior month into its archive (atomic per-file),
238 // appending to any existing archive for that month.
239 let dir = archive_dir(store);
240 fs::create_dir_all(&dir)?;
241 // Mark the rotation in-flight so a crash before the active trim
242 // is recoverable as a re-roll (deduped), not re-appended.
243 if !recovering {
244 fs::write(&marker, b"")?;
245 }
246
247 // Scope the crash-recovery dedup correctly. The marker only tells
248 // us a rotation may have been interrupted; on its own it does NOT
249 // prove these specific entries are a re-roll. A genuine interrupted
250 // rotation commits its archive appends FIRST and crashes before the
251 // active trim, so on retry every prior-month entry still in the
252 // active file already has a matching copy in its archive — the
253 // whole roll-out batch is a multiset-subset of the archives. Only
254 // then is the dedup the right thing: suppress the copies the prior
255 // attempt already wrote.
256 //
257 // If instead some prior-month entry has NO matching archive copy,
258 // no completed archive write exists to re-roll: the marker is stale
259 // (e.g. committed/synced into a `merge=union` clone after a crash
260 // stranded it) and these entries are a FRESH roll. Treating them as
261 // a re-roll would dedup a genuinely-distinct same-(minute,kind,
262 // object,note) entry against an unrelated pre-existing archive entry
263 // and, because the active file is trimmed unconditionally below,
264 // drop it from disk entirely. So we only enter recovery mode when
265 // the entire batch is already reflected in the archives. The
266 // tradeoff favors preservation: a rare genuine partial multi-month
267 // crash may re-append an already-archived entry (a visible,
268 // recoverable duplicate) rather than ever silently losing one.
269 let recovering_reroll = recovering && batch_is_archived(store, &by_month)?;
270
271 for ((y, m), month_entries) in &by_month {
272 let path = archive_path(store, *y, *m);
273 append_to_archive(&path, month_entries, recovering_reroll)?;
274 }
275
276 // Rewrite the active file to the kept (current-month) entries
277 // plus the new entry — atomically.
278 let mut body = String::new();
279 for e in &keep {
280 body.push_str(&e.render());
281 }
282 body.push_str(&entry.render());
283 let full = compose_active(&header, &body);
284 crate::fsx::write_atomic(&active, full.as_bytes())?;
285 // Rotation committed (active trimmed): clear the in-flight marker.
286 let _ = fs::remove_file(&marker);
287 return Ok(());
288 }
289
290 // No rotation needed. If a stale marker lingers (a crash that trimmed
291 // the active file but never deleted the marker), clear it so the next
292 // real rotation is treated as fresh, not stuck in recovery mode.
293 if recovering {
294 let _ = fs::remove_file(&marker);
295 }
296 // Plain atomic append of the rendered entry.
297 let mut full = content;
298 if !full.ends_with('\n') {
299 full.push('\n');
300 }
301 full.push_str(&entry.render());
302 crate::fsx::write_atomic(&active, full.as_bytes())?;
303 Ok(())
304 } else {
305 // Fresh log: frontmatter + the single entry.
306 if let Some(parent) = active.parent() {
307 fs::create_dir_all(parent)?;
308 }
309 let body = entry.render();
310 let full = compose_active(LOG_FRONTMATTER, &body);
311 crate::fsx::write_atomic(&active, full.as_bytes())?;
312 Ok(())
313 }
314 }
315
316 /// The `n` most-recent entries **by timestamp**, returned oldest→newest.
317 ///
318 /// **Out-of-order safety (mirrors [`Log::since`]).** The log is append-only
319 /// but *not* guaranteed to be in non-decreasing timestamp order on disk: a
320 /// corrective entry is appended below the entry it corrects, a
321 /// backdated/clock-skewed write lands physically after newer entries, and a
322 /// `merge=union` clone merge interleaves both sides until a later agent
323 /// reorders. Out-of-order is only a `LOG_OUT_OF_ORDER` warning, never
324 /// rejected. So the last `n` *physical* entries are **not** the `n` newest
325 /// by time — taking them would omit a genuinely-recent entry that sits
326 /// physically before an older one, and the documented curator warm-up
327 /// (`dbmd log tail 20`) would report a stale picture of what was done lately.
328 /// We therefore feed every entry of each file we touch through a bounded
329 /// newest-by-timestamp window and let it select the true top `n`.
330 ///
331 /// Bounded cost: the active `log.md` is kept to the current month by
332 /// rotation, so a full read of it is cheap and is not a whole-store walk.
333 /// Across archives we *can* prune: each `log/<YYYY-MM>.md` holds only entries
334 /// from that month (rotation buckets by the entry's own year-month), so once
335 /// the window is full, an archive whose month is strictly before the
336 /// window-minimum's month cannot contain any entry newer than the current
337 /// `n`th-newest. We cross archives newest-month-first and stop at the first
338 /// such archive.
339 pub fn tail(store: &Store, n: usize) -> crate::Result<Vec<LogEntry>> {
340 if n == 0 {
341 return Ok(Vec::new());
342 }
343
344 // A bounded window of the `n` entries with the largest timestamps. No
345 // within-file early stop: out-of-order entries mean a newer entry can
346 // sit physically before an older one, so each file is read fully.
347 let mut window = NewestWindow::new(n);
348 // Active↔archive overlap dedup, narrowly scoped AND gated on the
349 // crash-recovery marker (mirrors `since` and the write side). The only
350 // legitimate source of an active↔archive overlap is a rotation
351 // interrupted between its two non-atomic durable writes (archive append
352 // committed, active trim not), which leaves the SAME entry in both the
353 // untrimmed active file and its month archive — exactly the state the
354 // `.rotating` marker records (see `rotation_marker_path`). When the
355 // marker is ABSENT (normal operation) there is no such overlap: a
356 // backdated append that rotated as a FRESH roll and merely collides on
357 // (minute,kind,object,note) with an active entry is a genuinely-DISTINCT
358 // event the write side deliberately preserved on disk, and the reader
359 // must report it. The old code deduped unconditionally and silently
360 // dropped that distinct entry (the bug). We therefore build `active_seen`
361 // — and suppress matching archive entries below — ONLY while the marker
362 // is present. Even then it suppresses only an ARCHIVE entry that matches
363 // an ACTIVE one, never active-vs-active or archive-vs-archive.
364 let recovering = rotation_marker_path(store).exists();
365 let mut active_seen: std::collections::HashSet<EntryKey> = std::collections::HashSet::new();
366
367 // Active file: scan fully (current-month-bounded by rotation). Record
368 // every identity for overlap detection *only while recovering* (the
369 // marker is present); consider every entry regardless — a same-minute
370 // duplicate WITHIN the active file is two distinct appends.
371 let active = active_log_path(store);
372 if active.exists() {
373 reverse_collect(&active, |e| {
374 if recovering {
375 active_seen.insert(entry_key(&e));
376 }
377 window.consider(e);
378 false
379 })?;
380 }
381
382 // Archives, newest-month-first. Once the window is full, an archive
383 // whose month is strictly before the window-minimum's month holds only
384 // entries older than the current cutoff, so it (and every older archive)
385 // is skippable.
386 for archive in list_archives_desc(store)? {
387 if let (true, Some(cutoff_ym), Some(arch_ym)) = (
388 window.is_full(),
389 window.min_year_month(),
390 archive_year_month(&archive),
391 ) {
392 if arch_ym < cutoff_ym {
393 break;
394 }
395 }
396 reverse_collect(&archive, |e| {
397 // Suppress only a crash-retry active↔archive overlap, and only
398 // when recovering (marker present). `active_seen` is empty
399 // otherwise, so this never suppresses in normal operation — a
400 // distinct same-(minute,kind,object,note) archive entry survives
401 // even when an active entry collides on those fields. Archives
402 // are never deduped against each other.
403 if !active_seen.contains(&entry_key(&e)) {
404 window.consider(e);
405 }
406 false
407 })?;
408 }
409
410 Ok(window.into_sorted())
411 }
412
413 /// Entries strictly newer than `time`, reverse-scanning active → archives.
414 ///
415 /// **No within-file early stop.** The log is append-only but *not*
416 /// guaranteed to be in non-decreasing timestamp order on disk: a corrective
417 /// entry is appended below the entry it corrects (SPEC: "if a finding is
418 /// wrong, append a corrective entry below it"), a backdated/clock-skewed
419 /// write lands physically after newer entries, and a `merge=union` clone
420 /// merge interleaves both sides until a later agent reorders. Out-of-order
421 /// is only a `LOG_OUT_OF_ORDER` warning, never rejected. So a newer entry
422 /// can sit physically *before* an older one; stopping at the first
423 /// older-than-`time` entry would silently drop those — the documented
424 /// curator warm-up (`dbmd log since <ts>`) would miss real recent work.
425 /// We therefore read every entry of each file we touch.
426 ///
427 /// Bounded cost: the active `log.md` is kept to the current month by
428 /// rotation, so a full read of it is cheap (the same read `tail` does for a
429 /// large `n`) and is not a whole-store walk. Across archives we *can* stop:
430 /// each `log/<YYYY-MM>.md` holds only entries from that month (rotation
431 /// buckets by the entry's own year-month), so an archive whose month is
432 /// strictly before `time`'s month cannot contain any entry newer than
433 /// `time`. We cross archives newest-month-first and stop at the first whose
434 /// month is entirely at or before `time`'s.
435 pub fn since(store: &Store, time: DateTime<FixedOffset>) -> crate::Result<Vec<LogEntry>> {
436 let mut collected: Vec<LogEntry> = Vec::new();
437 // Active↔archive overlap dedup, narrowly scoped AND gated on the
438 // crash-recovery marker (mirrors `tail` and the write side). An overlap
439 // (the SAME entry in both the untrimmed active file and the archive)
440 // arises ONLY from a rotation interrupted between its two non-atomic
441 // durable writes — exactly the state the `.rotating` marker records (see
442 // `rotation_marker_path`). When the marker is ABSENT (normal operation)
443 // there is no overlap to mask: a backdated append that rotated as a FRESH
444 // roll and merely collides on (minute,kind,object,note) with an active
445 // entry is a genuinely-DISTINCT event the write side preserved on disk,
446 // and the reader must report it. Deduping unconditionally silently
447 // dropped that distinct entry (the bug). We therefore record ACTIVE
448 // identities — and suppress matching archive entries below — ONLY while
449 // recovering; even then it suppresses an ARCHIVE entry against an ACTIVE
450 // one, never active-vs-active or archive-vs-archive.
451 let recovering = rotation_marker_path(store).exists();
452 let mut active_seen: std::collections::HashSet<EntryKey> = std::collections::HashSet::new();
453
454 // Active file: scan fully, no early stop (out-of-order safe). Collect
455 // every in-window entry (a same-minute duplicate within the active file
456 // is two distinct appends), recording identities for overlap detection
457 // only while recovering (the marker is present).
458 let active = active_log_path(store);
459 if active.exists() {
460 reverse_collect(&active, |e| {
461 if e.timestamp > time {
462 if recovering {
463 active_seen.insert(entry_key(&e));
464 }
465 collected.push(e);
466 }
467 false
468 })?;
469 }
470
471 // The cutoff's own (year, month): any archive strictly before it holds
472 // only older entries and is skippable. Archive months are bucketed on
473 // the UTC calendar (on-disk timestamps are offset-free and re-read as
474 // UTC; rotation buckets by the entry's UTC year-month), so the pruning
475 // calendar must be UTC too. A non-UTC `since` offset (advertised in the
476 // CLI hint, e.g. `…T00:30:00+07:00`) whose local month differs from its
477 // UTC month would otherwise prune away an archive holding entries that
478 // are strictly newer than `time` — `time.year()/.month()` read the
479 // offset-LOCAL calendar, not UTC.
480 let cutoff_utc = time.with_timezone(&Utc);
481 let cutoff_ym = (cutoff_utc.year(), cutoff_utc.month());
482
483 for archive in list_archives_desc(store)? {
484 // Archives are newest-month-first; once a month is strictly before
485 // the cutoff's month, every remaining (older) archive is too.
486 if let Some(arch_ym) = archive_year_month(&archive) {
487 if arch_ym < cutoff_ym {
488 break;
489 }
490 }
491 // Scan this archive fully — within a month, entries may still be
492 // out of order, so no within-file early stop.
493 reverse_collect(&archive, |e| {
494 // Suppress only a crash-retry active↔archive overlap, and only
495 // when recovering (marker present). `active_seen` is empty
496 // otherwise, so a distinct same-(minute,kind,object,note) archive
497 // entry survives in normal operation even when an active entry
498 // collides on those fields.
499 if e.timestamp > time && !active_seen.contains(&entry_key(&e)) {
500 collected.push(e);
501 }
502 false
503 })?;
504 }
505
506 collected.reverse();
507 Ok(collected)
508 }
509
510 /// The timestamp of the most recent `validate` entry — the default `since`
511 /// window for working-set validation ([`crate::validate::validate_working_set`]).
512 pub fn last_validate_at(store: &Store) -> crate::Result<Option<DateTime<FixedOffset>>> {
513 let mut found: Option<DateTime<FixedOffset>> = None;
514
515 let active = active_log_path(store);
516 if active.exists() {
517 reverse_collect(&active, |e| {
518 if e.kind == LogKind::Validate {
519 found = Some(e.timestamp);
520 true
521 } else {
522 false
523 }
524 })?;
525 }
526
527 if found.is_none() {
528 for archive in list_archives_desc(store)? {
529 reverse_collect(&archive, |e| {
530 if e.kind == LogKind::Validate {
531 found = Some(e.timestamp);
532 true
533 } else {
534 false
535 }
536 })?;
537 if found.is_some() {
538 break;
539 }
540 }
541 }
542
543 Ok(found)
544 }
545
546 /// Parse a single entry header (`## [YYYY-MM-DD HH:MM] <kind> | <object>`)
547 /// into its timestamp, kind, and object. Returns `None` if the line isn't a
548 /// well-formed entry header.
549 pub fn parse_header(line: &str) -> Option<(DateTime<FixedOffset>, LogKind, Option<String>)> {
550 let line = line.trim_end_matches(['\n', '\r']);
551 let rest = line.strip_prefix("## [")?;
552 let close = rest.find(']')?;
553 let ts_str = &rest[..close];
554 let timestamp = parse_timestamp(ts_str)?;
555
556 // Everything after the closing bracket: ` <kind> | <object>` or
557 // ` <kind>`.
558 let after = rest[close + 1..].trim();
559 if after.is_empty() {
560 return None;
561 }
562
563 let (kind_str, object) = match after.split_once('|') {
564 Some((k, o)) => {
565 let obj = o.trim();
566 let obj = if obj.is_empty() {
567 None
568 } else {
569 Some(obj.to_string())
570 };
571 (k.trim(), obj)
572 }
573 None => (after, None),
574 };
575
576 if kind_str.is_empty() {
577 return None;
578 }
579
580 Some((timestamp, LogKind::parse(kind_str), object))
581 }
582}
583
584// ── Internal helpers ────────────────────────────────────────────────────────
585
586/// A bounded window of the `n` entries with the largest timestamps, fed by a
587/// **reverse (newest-physical-first) scan** and used by [`Log::tail`].
588///
589/// Why this exists: the last `n` *physical* entries are the `n` newest only
590/// when the log is in non-decreasing time order. That's the append-only
591/// contract, not a guarantee — a backdated, clock-skewed, or merge-interleaved
592/// entry violates it (and trips the `LOG_OUT_OF_ORDER` validate warning). The
593/// window decouples `tail` from that assumption: it keeps the `n` largest
594/// timestamps seen regardless of the order they arrive in, so the caller can
595/// read each file fully (no fragile within-file early stop) and still get the
596/// true top `n`.
597///
598/// Tie-break: entries sharing a timestamp at the window boundary are ordered by
599/// **physical recency** — the one appended later (encountered earlier in the
600/// reverse scan, i.e. a smaller `arrival`) wins. "Newest" means most-recently
601/// recorded.
602struct NewestWindow {
603 cap: usize,
604 /// Min-by-(timestamp, then physical-oldest) heap: the root is always the
605 /// next entry to evict once the window is full.
606 heap: std::collections::BinaryHeap<WindowItem>,
607 /// Count of entries fed in, in reverse-scan order, used as the tie-break
608 /// key (0 = newest physical).
609 next_arrival: u64,
610}
611
612impl NewestWindow {
613 fn new(cap: usize) -> Self {
614 NewestWindow {
615 cap,
616 heap: std::collections::BinaryHeap::with_capacity(cap),
617 next_arrival: 0,
618 }
619 }
620
621 /// Offer one entry from the scan. If the window isn't full it's kept; once
622 /// full, it's kept (evicting the current minimum) iff its timestamp is `>=`
623 /// the window minimum. Equal-timestamp boundary entries resolve by physical
624 /// recency (see the type doc).
625 fn consider(&mut self, entry: LogEntry) {
626 let arrival = self.next_arrival;
627 self.next_arrival += 1;
628
629 if self.heap.len() < self.cap {
630 self.heap.push(WindowItem { entry, arrival });
631 return;
632 }
633
634 // Window full. The heap root is the current minimum (oldest-by-
635 // timestamp held; on a tie, the oldest-physical).
636 let root = self.heap.peek().expect("full window has a root");
637 if entry.timestamp > root.entry.timestamp {
638 // Strictly newer than the window minimum: it belongs; evict the min.
639 self.heap.pop();
640 self.heap.push(WindowItem { entry, arrival });
641 }
642 // On `<=` we keep the window as-is. `<` is plainly too old. `==` is the
643 // tie case: the scan is newest-physical-first, so this entry is
644 // physically *older* than the held one of equal timestamp, and the
645 // tie-break keeps the physically-newer (most-recently-recorded) entry —
646 // so the incoming one is dropped.
647 }
648
649 /// Whether the window already holds its full `cap` entries.
650 fn is_full(&self) -> bool {
651 self.heap.len() >= self.cap
652 }
653
654 /// The `(year, month)` of the window's current minimum (oldest kept) entry,
655 /// or `None` when the window is empty. Used to prune older archives: an
656 /// archive month strictly before this can't beat the current cutoff.
657 fn min_year_month(&self) -> Option<(i32, u32)> {
658 self.heap
659 .peek()
660 .map(|item| (item.entry.timestamp.year(), item.entry.timestamp.month()))
661 }
662
663 /// The held entries, oldest→newest (chronological), ties broken
664 /// oldest-physical→newest-physical.
665 fn into_sorted(self) -> Vec<LogEntry> {
666 let mut items: Vec<WindowItem> = self.heap.into_vec();
667 // Ascending by timestamp; on a tie, oldest-physical (larger arrival)
668 // first so the most-recently-recorded entry sorts last.
669 items.sort_by(|a, b| {
670 a.entry
671 .timestamp
672 .cmp(&b.entry.timestamp)
673 .then(b.arrival.cmp(&a.arrival))
674 });
675 items.into_iter().map(|i| i.entry).collect()
676 }
677}
678
679/// One slot in [`NewestWindow`]'s heap. `Ord` is defined so the heap is a
680/// **min-heap on `(timestamp, physical-oldest)`**: `BinaryHeap` is a max-heap,
681/// so the root (max under this `Ord`) is the eviction candidate — the smallest
682/// timestamp, and on a tie the oldest-physical (largest `arrival`).
683struct WindowItem {
684 entry: LogEntry,
685 arrival: u64,
686}
687
688impl PartialEq for WindowItem {
689 fn eq(&self, other: &Self) -> bool {
690 self.entry.timestamp == other.entry.timestamp && self.arrival == other.arrival
691 }
692}
693impl Eq for WindowItem {}
694
695impl Ord for WindowItem {
696 fn cmp(&self, other: &Self) -> std::cmp::Ordering {
697 // Reverse on timestamp so the *smallest* timestamp is the heap max
698 // (eviction candidate). On equal timestamps, the larger `arrival`
699 // (older physical) is the heap max so it is evicted first.
700 other
701 .entry
702 .timestamp
703 .cmp(&self.entry.timestamp)
704 .then(self.arrival.cmp(&other.arrival))
705 }
706}
707impl PartialOrd for WindowItem {
708 fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
709 Some(self.cmp(other))
710 }
711}
712
713/// An advisory, exclusive lock serializing concurrent [`Log::append`] calls.
714///
715/// Held on a dedicated sibling lock file (`<active>.lock`) rather than on
716/// `log.md` itself: `write_atomic` replaces the active file by `rename`, so the
717/// active inode changes under us and a lock on its fd would not cover the new
718/// file. The lock file is stable, so the lock spans the whole read-modify-write.
719///
720/// On Unix this is `flock(LOCK_EX)`, released on drop (or implicitly when the
721/// process exits / the fd closes, so a crash never strands the lock). The
722/// lock file is created if absent and intentionally left on disk between runs
723/// (locking it does not depend on its contents). On non-Unix targets the lock
724/// is a no-op — db.md's append surface is Unix-targeted, and a missing advisory
725/// lock degrades to the pre-fix last-writer-wins, never to incorrectness of a
726/// single writer.
727struct AppendLock {
728 #[cfg(unix)]
729 file: Option<File>,
730}
731
732impl AppendLock {
733 /// Acquire the exclusive append lock for the store whose active log is
734 /// `active`. Best-effort: any failure to open or lock the lock file yields
735 /// an unlocked guard (we log rather than refuse to log). Blocks until the
736 /// lock is granted when another appender holds it.
737 fn acquire(active: &Path) -> AppendLock {
738 #[cfg(unix)]
739 {
740 let file = Self::open_and_lock(active);
741 AppendLock { file }
742 }
743 #[cfg(not(unix))]
744 {
745 let _ = active;
746 AppendLock {}
747 }
748 }
749
750 #[cfg(unix)]
751 fn open_and_lock(active: &Path) -> Option<File> {
752 use std::os::unix::io::AsRawFd;
753
754 // The lock file lives beside the active log; ensure its parent exists
755 // (the fresh-log path may run before `log.md`'s directory is created).
756 if let Some(parent) = active.parent() {
757 let _ = fs::create_dir_all(parent);
758 }
759 let lock_path = lock_path_for(active);
760 let file = std::fs::OpenOptions::new()
761 .create(true)
762 .truncate(false)
763 .write(true)
764 .open(&lock_path)
765 .ok()?;
766
767 // Blocking exclusive advisory lock. `flock` is in libc, which every Rust
768 // binary links, so the bare `extern "C"` declaration needs no crate dep.
769 let rc = unsafe { flock(file.as_raw_fd(), LOCK_EX) };
770 if rc != 0 {
771 // Could not lock (e.g. a filesystem without flock support): proceed
772 // unlocked rather than fail the append.
773 return None;
774 }
775 Some(file)
776 }
777}
778
779#[cfg(unix)]
780impl Drop for AppendLock {
781 fn drop(&mut self) {
782 use std::os::unix::io::AsRawFd;
783 if let Some(file) = &self.file {
784 // Release explicitly; the fd close on drop would also release it.
785 unsafe { flock(file.as_raw_fd(), LOCK_UN) };
786 }
787 }
788}
789
790#[cfg(unix)]
791extern "C" {
792 fn flock(fd: std::os::raw::c_int, operation: std::os::raw::c_int) -> std::os::raw::c_int;
793}
794
795/// `flock` operation: exclusive lock (`LOCK_EX`), blocking.
796#[cfg(unix)]
797const LOCK_EX: std::os::raw::c_int = 2;
798/// `flock` operation: unlock (`LOCK_UN`).
799#[cfg(unix)]
800const LOCK_UN: std::os::raw::c_int = 8;
801
802/// The advisory-lock sibling path for an active log file (`<name>.lock`).
803#[cfg(unix)]
804fn lock_path_for(active: &Path) -> PathBuf {
805 let mut name = active
806 .file_name()
807 .map(|s| s.to_os_string())
808 .unwrap_or_else(|| std::ffi::OsString::from("log.md"));
809 name.push(".lock");
810 match active.parent() {
811 Some(parent) => parent.join(name),
812 None => PathBuf::from(name),
813 }
814}
815
816/// The active `log.md` path under the store root.
817fn active_log_path(store: &Store) -> PathBuf {
818 store.root.join("log.md")
819}
820
821/// The `log/` archive directory under the store root.
822fn archive_dir(store: &Store) -> PathBuf {
823 store.root.join("log")
824}
825
826/// The `log/<YYYY-MM>.md` archive path for a given month.
827fn archive_path(store: &Store, year: i32, month: u32) -> PathBuf {
828 archive_dir(store).join(format!("{:04}-{:02}.md", year, month))
829}
830
831/// The crash-recovery marker for an in-progress rotation.
832///
833/// Its **presence** at the start of [`Log::append`] means a prior rotation
834/// appended prior-month entries to their archives but may not have trimmed the
835/// active file (a crash, or an active-rewrite error, between the two non-atomic
836/// durable writes). The retry must then DEDUP the re-rolled entries against the
837/// archive so it adds nothing.
838///
839/// Its **absence** means a fresh rotation: every prior-month entry being rolled
840/// is genuinely new to its archive and is appended UNCONDITIONALLY. This is the
841/// load-bearing distinction — a content-only dedup cannot tell an idempotent
842/// re-roll of one physical entry from a genuinely-distinct same-minute repeat
843/// (on-disk headers are minute-precision, so two real appends to the same object
844/// in the same minute with the same note render byte-identically). Gating the
845/// dedup on "are we recovering a crashed rotation?" lets a backdated duplicate
846/// survive while still suppressing a true re-roll.
847///
848/// Lives in `log/` (toolkit-managed; a dotfile, so never walked, indexed, or
849/// validated as content — `list_archives_desc` matches only `YYYY-MM.md`).
850fn rotation_marker_path(store: &Store) -> PathBuf {
851 archive_dir(store).join(".rotating")
852}
853
854/// Parse a `YYYY-MM-DD HH:MM` header timestamp, reattaching UTC. `None` on any
855/// malformed shape.
856fn parse_timestamp(s: &str) -> Option<DateTime<FixedOffset>> {
857 let naive = NaiveDateTime::parse_from_str(s.trim(), TS_FORMAT).ok()?;
858 let utc = FixedOffset::east_opt(0)?;
859 utc.from_local_datetime(&naive).single()
860}
861
862/// Split a `log.md` / archive file into its leading frontmatter+heading block
863/// (everything up to and including the line before the first `## [` header) and
864/// its parsed entries. If there are no entries, the whole content is the header
865/// block.
866fn parse_active(content: &str) -> (String, Vec<LogEntry>) {
867 match find_first_header(content) {
868 Some(idx) => {
869 let header = content[..idx].to_string();
870 let entries = parse_entries(&content[idx..]);
871 (header, entries)
872 }
873 None => (content.to_string(), Vec::new()),
874 }
875}
876
877/// Byte offset of the first **valid** entry header — a `## [` line-start that
878/// [`Log::parse_header`] accepts — or `None`.
879///
880/// Crucially this skips `## [`-SHAPED lines that `parse_header` REJECTS (a
881/// merge-orphaned note, an exporter-malformed line) appearing before the first
882/// real entry: everything up to the first valid header becomes the preserved
883/// `header` block in [`parse_active`], so a rotation re-emits it verbatim.
884/// Returning the first `## [`-shaped line instead (as this once did) put those
885/// pre-entry lines into the entries region, where [`parse_entries`] — which
886/// opens an entry only on a parseable header — dropped them on the floor,
887/// silently erasing append-only content on the next rotation.
888fn find_first_header(content: &str) -> Option<usize> {
889 let mut offset = 0usize;
890 for line in content.split_inclusive('\n') {
891 let line_str = line.trim_end_matches(['\r', '\n']);
892 if line_str.starts_with("## [") && Log::parse_header(line_str).is_some() {
893 return Some(offset);
894 }
895 offset += line.len();
896 }
897 None
898}
899
900/// Whether `line` is a note line that — left unescaped — could be mistaken for
901/// an entry header. It is *header-ambiguous* when it is a (possibly empty) run
902/// of leading backslashes followed by a string that [`Log::parse_header`]
903/// accepts. The escape (one leading backslash) and only the escape is added to,
904/// or stripped from, such lines, so the transform is fully reversible:
905/// `## [..]` (a real header shape in note text) ⇄ `\## [..]`, and a literal
906/// `\## [..]` a note already contains ⇄ `\\## [..]`.
907fn is_header_ambiguous(line: &str) -> bool {
908 let stripped = line.trim_start_matches('\\');
909 // Only treat it as ambiguous if some backslashes were the *only* prefix and
910 // the remainder is a valid header — a backslash run that does not lead into
911 // a header (e.g. `\not a header`) is ordinary note text, left untouched.
912 Log::parse_header(stripped).is_some()
913}
914
915/// Escape one note line for on-disk emission so it can never be parsed as an
916/// entry header (the [write-path fix] for header-shaped notes corrupting the
917/// append-only log). A header-ambiguous line is prefixed with a single
918/// backslash, moving its `## [` off column 0; every other line is emitted
919/// verbatim. Reversed exactly by [`unescape_note_line`].
920fn escape_note_line(line: &str) -> std::borrow::Cow<'_, str> {
921 if is_header_ambiguous(line) {
922 std::borrow::Cow::Owned(format!("\\{line}"))
923 } else {
924 std::borrow::Cow::Borrowed(line)
925 }
926}
927
928/// Reverse [`escape_note_line`]: strip exactly one leading backslash from a
929/// header-ambiguous on-disk note line, restoring the literal the author wrote.
930/// A line that is not header-ambiguous (including a genuine `\not a header`) is
931/// returned untouched, so the round-trip is lossless for arbitrary note text.
932fn unescape_note_line(line: &str) -> std::borrow::Cow<'_, str> {
933 if let Some(rest) = line.strip_prefix('\\') {
934 if is_header_ambiguous(line) {
935 return std::borrow::Cow::Borrowed(rest);
936 }
937 }
938 std::borrow::Cow::Borrowed(line)
939}
940
941/// Parse every entry in a slice that begins at (or before, header-block
942/// included) a sequence of `## [` headers. Headers that fail to parse are
943/// skipped (their body folds into the previous valid entry's note is avoided —
944/// they simply start no new entry).
945fn parse_entries(text: &str) -> Vec<LogEntry> {
946 let mut entries: Vec<LogEntry> = Vec::new();
947 let mut cur_header: Option<(DateTime<FixedOffset>, LogKind, Option<String>)> = None;
948 let mut cur_note: Vec<&str> = Vec::new();
949
950 let flush = |entries: &mut Vec<LogEntry>,
951 header: &mut Option<(DateTime<FixedOffset>, LogKind, Option<String>)>,
952 note: &mut Vec<&str>| {
953 if let Some((timestamp, kind, object)) = header.take() {
954 // Reverse the per-line header escape `render` applies so an escaped
955 // header-shaped note line round-trips back to its literal form.
956 let joined = note
957 .iter()
958 .map(|line| unescape_note_line(line))
959 .collect::<Vec<_>>()
960 .join("\n");
961 let note_str = joined.trim_matches(['\n', '\r']).to_string();
962 entries.push(LogEntry {
963 timestamp,
964 kind,
965 object,
966 note: note_str,
967 });
968 }
969 note.clear();
970 };
971
972 for line in text.lines() {
973 if line.starts_with("## [") {
974 if let Some(parsed) = Log::parse_header(line) {
975 // Close the previous entry, start a new one.
976 flush(&mut entries, &mut cur_header, &mut cur_note);
977 cur_header = Some(parsed);
978 continue;
979 }
980 // Unparseable `## [` line: treat as body of the current entry.
981 }
982 if cur_header.is_some() {
983 cur_note.push(line);
984 }
985 }
986 flush(&mut entries, &mut cur_header, &mut cur_note);
987 entries
988}
989
990/// Recompose an active/archive file from a header block and an entry body.
991fn compose_active(header: &str, body: &str) -> String {
992 let mut out = String::new();
993 out.push_str(header);
994 if !header.is_empty() && !header.ends_with('\n') {
995 out.push('\n');
996 }
997 // Exactly one blank line between the heading block and the first entry.
998 if !header.is_empty() && !out.ends_with("\n\n") {
999 out.push('\n');
1000 }
1001 out.push_str(body);
1002 out
1003}
1004
1005/// Append entries to a month archive, creating it with `type: log` frontmatter
1006/// if absent. Atomic (temp-file rename). Entries are appended in the given
1007/// order (callers pass them already chronological within the month).
1008///
1009/// **`recovering` — the re-roll gate.** Rotation in [`Log::append`] is two
1010/// non-atomic durable writes: roll prior-month entries into the archive, then
1011/// rewrite (trim) the active file. If the process crashes or the active rewrite
1012/// errors *after* the archive write commits, the prior-month entries remain in
1013/// the still-untrimmed active file and the agent's retry re-rolls them here. A
1014/// naive concatenate would then duplicate every entry, amplifying on each retry.
1015///
1016/// We CANNOT dedup that away by content alone: on-disk headers are
1017/// minute-precision, so two genuinely-distinct appends to the same object in the
1018/// same minute with the same note render byte-identically — indistinguishable
1019/// from a re-roll of one physical entry. Deduping unconditionally therefore
1020/// silently destroyed a legitimately-distinct backdated duplicate (the bug).
1021///
1022/// So the caller passes `recovering`: `true` only when an in-progress-rotation
1023/// marker was found (a crash-retry), where we dedup the incoming batch against
1024/// the archive **by multiplicity** (skip an incoming entry only while the
1025/// archive still holds an unconsumed copy of its identity) so a re-roll of the
1026/// SAME physical entries adds nothing. On a fresh rotation (`false`) every entry
1027/// is genuinely new to the archive and is appended unconditionally, so a
1028/// distinct same-minute repeat survives.
1029fn append_to_archive(path: &Path, entries: &[LogEntry], recovering: bool) -> crate::Result<()> {
1030 if path.exists() {
1031 let existing = fs::read_to_string(path)?;
1032
1033 let mut body = String::new();
1034 if recovering {
1035 // Crash-retry: the prior (crashed) attempt may already have appended
1036 // some/all of these. Dedup by MULTIPLICITY, not set-membership, so a
1037 // partial-then-retried roll converges exactly and a re-roll of the
1038 // full batch is a no-op.
1039 let (_header, existing_entries) = parse_active(&existing);
1040 let mut remaining: std::collections::HashMap<EntryKey, usize> =
1041 std::collections::HashMap::new();
1042 for e in &existing_entries {
1043 *remaining.entry(entry_key(e)).or_insert(0) += 1;
1044 }
1045 for e in entries {
1046 match remaining.get_mut(&entry_key(e)) {
1047 // An archived copy is still unconsumed: this incoming entry is
1048 // that re-roll, suppress it.
1049 Some(count) if *count > 0 => *count -= 1,
1050 _ => body.push_str(&e.render()),
1051 }
1052 }
1053 } else {
1054 // Fresh rotation: append every entry. A same-minute, same-fields
1055 // entry that already exists in the archive is a DISTINCT append, not
1056 // a re-roll, and must be preserved.
1057 for e in entries {
1058 body.push_str(&e.render());
1059 }
1060 }
1061
1062 // Nothing new to add (a fully-duplicate re-roll): leave the archive
1063 // byte-for-byte untouched (append-only: don't rewrite identical data).
1064 if body.is_empty() {
1065 return Ok(());
1066 }
1067
1068 let mut full = existing;
1069 if !full.ends_with('\n') {
1070 full.push('\n');
1071 }
1072 full.push_str(&body);
1073 crate::fsx::write_atomic(path, full.as_bytes())?;
1074 } else {
1075 let mut body = String::new();
1076 for e in entries {
1077 body.push_str(&e.render());
1078 }
1079 if let Some(parent) = path.parent() {
1080 fs::create_dir_all(parent)?;
1081 }
1082 let full = compose_active(LOG_FRONTMATTER, &body);
1083 crate::fsx::write_atomic(path, full.as_bytes())?;
1084 }
1085 Ok(())
1086}
1087
1088/// True iff every prior-month entry about to be rolled out (`by_month`) already
1089/// has a matching, unconsumed copy in its month archive — i.e. the whole
1090/// roll-out batch is a multiset-subset of the archives.
1091///
1092/// This is the load-bearing test for "is the `.rotating` marker a genuine
1093/// interrupted-rotation re-roll, or a stale marker over a fresh roll?" A genuine
1094/// interrupted rotation commits its per-month archive appends BEFORE the active
1095/// trim, so on retry the still-untrimmed active file's prior-month entries are
1096/// all present in the archives — exactly the duplicates the dedup must suppress.
1097/// If any entry is missing from its archive, no completed archive write exists to
1098/// re-roll: the marker is stale and these are a fresh roll that must be appended,
1099/// never deduped (deduping a genuinely-distinct same-(minute,kind,object,note)
1100/// entry against an unrelated pre-existing archive copy would, with the
1101/// unconditional active trim, drop it from disk — the bug).
1102///
1103/// Multiset semantics: each archived copy is consumed at most once, so two
1104/// distinct same-minute entries in the batch require two archived copies to count
1105/// as a re-roll. Cheap: only the months actually being rolled are read, and only
1106/// when the marker is present (the cold recovery path).
1107fn batch_is_archived(
1108 store: &Store,
1109 by_month: &BTreeMap<(i32, u32), Vec<LogEntry>>,
1110) -> crate::Result<bool> {
1111 for ((y, m), month_entries) in by_month {
1112 let path = archive_path(store, *y, *m);
1113 if !path.exists() {
1114 // No archive for this month: nothing was rolled here yet, so the
1115 // batch cannot be a completed re-roll.
1116 return Ok(false);
1117 }
1118 let (_header, archived) = parse_active(&fs::read_to_string(&path)?);
1119 let mut available: std::collections::HashMap<EntryKey, usize> =
1120 std::collections::HashMap::new();
1121 for e in &archived {
1122 *available.entry(entry_key(e)).or_insert(0) += 1;
1123 }
1124 for e in month_entries {
1125 match available.get_mut(&entry_key(e)) {
1126 Some(count) if *count > 0 => *count -= 1,
1127 _ => return Ok(false),
1128 }
1129 }
1130 }
1131 Ok(true)
1132}
1133
1134/// A hashable identity for a log entry, used to dedup an idempotent archive
1135/// re-roll (see [`append_to_archive`]). Two entries are "the same" when their
1136/// timestamp, kind, object, and note all match — exactly the fields that
1137/// round-trip through `render`/`parse`, so a re-rolled entry compares equal to
1138/// the one already archived. Owned (rather than borrowed) so keys from the
1139/// existing archive and from the incoming entries share one type regardless of
1140/// where they came from; the cost is paid only on the cold rotation path.
1141type EntryKey = (DateTime<FixedOffset>, String, Option<String>, String);
1142
1143/// Derive the dedup key for `e` (see [`EntryKey`]). Keying on `kind.as_str()`
1144/// (rather than `LogKind`, which is not `Hash`) is exact: `as_str`/`parse`
1145/// round-trips every recognized kind and preserves any `Custom` token.
1146fn entry_key(e: &LogEntry) -> EntryKey {
1147 (
1148 e.timestamp,
1149 e.kind.as_str().to_string(),
1150 e.object.clone(),
1151 e.note.clone(),
1152 )
1153}
1154
1155/// Every `log/<YYYY-MM>.md` archive, sorted **newest month first**.
1156fn list_archives_desc(store: &Store) -> crate::Result<Vec<PathBuf>> {
1157 let dir = archive_dir(store);
1158 if !dir.is_dir() {
1159 return Ok(Vec::new());
1160 }
1161 let mut months: Vec<(String, PathBuf)> = Vec::new();
1162 for entry in fs::read_dir(&dir)? {
1163 let entry = entry?;
1164 let path = entry.path();
1165 if !path.is_file() {
1166 continue;
1167 }
1168 let name = match path.file_name().and_then(|s| s.to_str()) {
1169 Some(n) => n,
1170 None => continue,
1171 };
1172 // Match `YYYY-MM.md`.
1173 if let Some(stem) = name.strip_suffix(".md") {
1174 if is_year_month(stem) {
1175 months.push((stem.to_string(), path.clone()));
1176 }
1177 }
1178 }
1179 // `YYYY-MM` strings sort lexically == chronologically; reverse for newest
1180 // first.
1181 months.sort_by(|a, b| b.0.cmp(&a.0));
1182 Ok(months.into_iter().map(|(_, p)| p).collect())
1183}
1184
1185/// The `(year, month)` an archive file represents, parsed from its
1186/// `log/<YYYY-MM>.md` name. `None` if the name isn't a well-formed month
1187/// archive (in which case the caller scans it rather than risk skipping it).
1188fn archive_year_month(path: &Path) -> Option<(i32, u32)> {
1189 let stem = path
1190 .file_name()
1191 .and_then(|s| s.to_str())
1192 .and_then(|n| n.strip_suffix(".md"))?;
1193 if !is_year_month(stem) {
1194 return None;
1195 }
1196 let year: i32 = stem[..4].parse().ok()?;
1197 let month: u32 = stem[5..7].parse().ok()?;
1198 // The month must be a real calendar month. A hand-created / externally-
1199 // produced `log/2026-00.md` or `log/2026-13.md` parses as two digits but
1200 // names no month; returning `Some((year, 0))` would sort it below every
1201 // legitimate month, so the newest-month-first early-break in `since`/`tail`
1202 // could prune it and silently drop its entries. Out-of-range → `None`, so the
1203 // caller scans the file instead of risk-skipping it (the safe fallback).
1204 if !(1..=12).contains(&month) {
1205 return None;
1206 }
1207 Some((year, month))
1208}
1209
1210/// True if `s` looks like `YYYY-MM` (4 digits, dash, 2 digits).
1211fn is_year_month(s: &str) -> bool {
1212 let bytes = s.as_bytes();
1213 if bytes.len() != 7 {
1214 return false;
1215 }
1216 bytes[..4].iter().all(u8::is_ascii_digit)
1217 && bytes[4] == b'-'
1218 && bytes[5].is_ascii_digit()
1219 && bytes[6].is_ascii_digit()
1220}
1221
1222/// Reverse-read `path` from EOF, parsing entries newest-first and feeding each
1223/// to `take`. `take` returns `true` to stop early (enough collected). The file
1224/// is read backward in blocks; only the tail region needed to satisfy `take`
1225/// is read — the whole file is read only if `take` never returns `true`.
1226fn reverse_collect<F>(path: &Path, mut take: F) -> crate::Result<()>
1227where
1228 F: FnMut(LogEntry) -> bool,
1229{
1230 let mut file = File::open(path)?;
1231 let len = file.metadata()?.len();
1232 if len == 0 {
1233 return Ok(());
1234 }
1235
1236 // Algorithm: grow a tail buffer leftward one block at a time, emitting
1237 // entries strictly newest-first as their left boundary is confirmed, and
1238 // stopping the instant `take` says enough. The whole file is read only if
1239 // `take` never returns `true` (e.g. `tail(n)` with n ≥ entry count).
1240 //
1241 // Invariant: a `## [` line-start anywhere in the buffer is a *complete*
1242 // entry — its header is the entry's first line, and its body lies to the
1243 // right and is therefore already buffered (we read right-to-left). So we
1244 // never split an entry across blocks.
1245 //
1246 // `buf` holds the file's bytes from absolute offset `start` (growing
1247 // leftward toward 0) to EOF. `emitted_abs` records the absolute offsets of
1248 // headers already handed to `take`, so re-visiting a header in a later block
1249 // never double-emits.
1250 let mut buf: Vec<u8> = Vec::new();
1251 let mut start = len;
1252 // O(1) membership: a `Vec` + `.contains()` here would be O(E²) across a large
1253 // single-month file (every header re-checked against all prior emissions).
1254 let mut emitted_abs: std::collections::HashSet<u64> = std::collections::HashSet::new();
1255 // Every header's absolute offset found so far, ascending. Built
1256 // *incrementally*: each block contributes only the markers whose `#` starts
1257 // inside it (all strictly smaller than any already-known offset, so they
1258 // prepend in order). This is the fix for the accidental O(file²) scan — the
1259 // old code re-ran `header_offsets` over the whole accumulated buffer on every
1260 // block (O(file²/block) byte comparisons on the default no-early-stop
1261 // tail/since path); now each byte is scanned for a header exactly once.
1262 let mut headers: Vec<u64> = Vec::new();
1263 let mut stop = false;
1264 // The first backward block has no already-scanned region to its right, so it
1265 // scans exactly `[0, block)`; every later block scans one byte further
1266 // (`block + 1`) to re-classify the prior block's deferred left-edge candidate
1267 // now that its left neighbour is buffered (see the scan call below).
1268 let mut first = true;
1269
1270 while start > 0 && !stop {
1271 let block = std::cmp::min(REVERSE_BLOCK as u64, start);
1272 let new_start = start - block;
1273 file.seek(SeekFrom::Start(new_start))?;
1274 let mut chunk = vec![0u8; block as usize];
1275 file.read_exact(&mut chunk)?;
1276 chunk.extend_from_slice(&buf);
1277 buf = chunk;
1278 start = new_start;
1279
1280 // Scan the freshly-prepended block (buffer indices `[0, block)`) for new
1281 // header markers. A marker straddling the block boundary has its `#` in
1282 // this window and so is still caught (see `header_offsets_range`).
1283 //
1284 // One subtlety the scan must respect: a `## [` whose `#` sits at the
1285 // block's LEFT edge (buffer index 0, absolute offset `start`) cannot have
1286 // its line-start confirmed yet when `start > 0` — the byte at `start - 1`
1287 // is not buffered. Treating index 0 as a line start there fabricates an
1288 // entry from a mid-line `## [` fragment that happens to align with a block
1289 // boundary. So `header_offsets_range` DEFERS the leftmost candidate when
1290 // `base` is not the true file start, and we re-scan one byte further
1291 // right next time: after the first block the buffer carries the previous
1292 // block's left-edge byte at index `block` with its left neighbour now in
1293 // hand, so extending the window to `block + 1` re-classifies that exactly
1294 // once. `first` guards the first block (nothing to re-check on its right).
1295 let base_is_file_start = start == 0;
1296 let scan_hi = if first { block } else { block + 1 } as usize;
1297 let mut new_headers = header_offsets_range(&buf, start, 0, scan_hi, base_is_file_start);
1298 first = false;
1299 if !new_headers.is_empty() {
1300 new_headers.extend_from_slice(&headers);
1301 headers = new_headers;
1302 }
1303
1304 // Process newest (largest offset) → oldest (smallest), emitting any
1305 // header not yet emitted. Hold back only the buffer's *leftmost* header
1306 // while we have not reached file start (`start > 0`): older entries may
1307 // still lie to its left in unread blocks, and newest-first order
1308 // requires we not emit it until we've confirmed it really is the oldest
1309 // (or read enough to bound it on the left). One extra block read at
1310 // most; on the next iteration its left boundary is in-buffer.
1311 for i in (0..headers.len()).rev() {
1312 let abs = headers[i];
1313 if emitted_abs.contains(&abs) {
1314 continue;
1315 }
1316 let is_oldest_in_buf = i == 0;
1317 if is_oldest_in_buf && start > 0 {
1318 continue;
1319 }
1320
1321 let entry_text = entry_text_at(&buf, start, abs, &headers, i);
1322 if let Some(entry) = parse_single_entry(&entry_text) {
1323 emitted_abs.insert(abs);
1324 if take(entry) {
1325 stop = true;
1326 break;
1327 }
1328 } else {
1329 emitted_abs.insert(abs);
1330 }
1331 }
1332 }
1333
1334 // Reached file start (or stopped). If we stopped, done. If we reached
1335 // start, emit any held-back oldest header(s) now (start == 0 means the
1336 // buffer's first header is genuinely the oldest). `headers` already holds
1337 // every offset (the loop scanned down to start == 0), so reuse it.
1338 if !stop && start == 0 {
1339 for i in (0..headers.len()).rev() {
1340 let abs = headers[i];
1341 if emitted_abs.contains(&abs) {
1342 continue;
1343 }
1344 let entry_text = entry_text_at(&buf, start, abs, &headers, i);
1345 if let Some(entry) = parse_single_entry(&entry_text) {
1346 emitted_abs.insert(abs);
1347 if take(entry) {
1348 break;
1349 }
1350 } else {
1351 emitted_abs.insert(abs);
1352 }
1353 }
1354 }
1355
1356 Ok(())
1357}
1358
1359/// Absolute byte offsets of every **valid** entry-header line-start (`## […]`)
1360/// in `buf`, where `buf` begins at absolute offset `base`.
1361///
1362/// Only a `## [` line that [`Log::parse_header`] accepts is an entry boundary,
1363/// mirroring the forward parser ([`parse_entries`]), which folds an unparseable
1364/// `## [` line into the preceding entry's note rather than starting a new entry.
1365/// Without this validity check the reverse reader would split a real entry's
1366/// multi-line note at a continuation line beginning at column 0 with `## [`
1367/// (a shape the SPEC permits — notes are "one or more lines" with no
1368/// restriction), truncating the note and dropping the carved pseudo-entry, so
1369/// `tail`/`since`/`last_validate_at` would return a note diverging from the
1370/// intact on-disk bytes.
1371///
1372/// Whole-buffer convenience wrapper over [`header_offsets_range`]. The runtime
1373/// reverse reader now always scans incrementally (one freshly-prepended window
1374/// per backward block), so this whole-buffer form is retained only as the
1375/// oracle the range-scan tests check the incremental scan against.
1376#[cfg(test)]
1377fn header_offsets(buf: &[u8], base: u64) -> Vec<u64> {
1378 // The whole-buffer oracle treats `base` as the file start iff it is 0, so a
1379 // `## [` at buffer index 0 is a real line-start there.
1380 header_offsets_range(buf, base, 0, buf.len(), base == 0)
1381}
1382
1383/// Like [`header_offsets`] but only reports header *markers whose `#` starts in*
1384/// `buf[scan_lo..scan_hi)`, while still consulting bytes outside that window —
1385/// to the left for the line-start (`buf[i-1] == b'\n'`) check and to the right
1386/// for the header line's content. This is the incremental scan
1387/// [`reverse_collect`] uses: each backward block searches only the freshly-
1388/// prepended region for *new* markers, so total header-scan work is linear in
1389/// the file size, not the O(file²) of re-scanning the whole growing buffer on
1390/// every block.
1391///
1392/// A `## [` marker that *straddles* the boundary (its `#` in the new block, its
1393/// `[` or trailing bytes in the already-scanned region) is still detected here:
1394/// its `#` index is `< scan_hi`, so it falls in this window, and it was never
1395/// reported by an earlier scan (whose window was `[block, …)`, strictly to the
1396/// right of this one) — so each marker is reported exactly once across all
1397/// blocks.
1398///
1399/// **Left-edge line-start safety.** A `## [` whose `#` is at buffer index 0 has
1400/// no buffered left neighbour, so its line-start cannot be confirmed unless
1401/// index 0 really is the file start. `base_is_file_start` says so: when it is
1402/// `false`, an index-0 candidate is DEFERRED (not reported) rather than assumed
1403/// to be at a line start — otherwise a mid-line `## […]` fragment that happens
1404/// to align with a block's left edge would be fabricated into an entry,
1405/// truncating the real entry's note and (after rotation) corrupting the
1406/// append-only archive. The caller re-scans that byte on the next block, once
1407/// its left neighbour is buffered, so a genuine boundary header is still found
1408/// exactly once.
1409fn header_offsets_range(
1410 buf: &[u8],
1411 base: u64,
1412 scan_lo: usize,
1413 scan_hi: usize,
1414 base_is_file_start: bool,
1415) -> Vec<u64> {
1416 const PAT: &[u8] = b"## [";
1417 let mut out = Vec::new();
1418 let n = buf.len();
1419 let hi = scan_hi.min(n);
1420 let mut i = scan_lo;
1421 // A marker's `#` must start strictly before `hi`; the pattern/line content
1422 // may read past `hi` into `buf` (the right neighbour is already buffered).
1423 while i < hi && i + PAT.len() <= n {
1424 if &buf[i..i + PAT.len()] == PAT {
1425 // Index 0 is a line start only when it is the genuine file start;
1426 // otherwise its left neighbour is unbuffered and the candidate is
1427 // deferred to the next block (see the doc comment).
1428 let at_line_start = if i == 0 {
1429 base_is_file_start
1430 } else {
1431 buf[i - 1] == b'\n'
1432 };
1433 if at_line_start && is_valid_header_line(buf, i) {
1434 out.push(base + i as u64);
1435 // skip ahead past this marker
1436 i += PAT.len();
1437 continue;
1438 }
1439 }
1440 i += 1;
1441 }
1442 out
1443}
1444
1445/// Whether the `## [` line starting at byte `i` in `buf` parses as a valid
1446/// entry header. Reads the line up to (but not including) the next `\n` (or
1447/// buffer end) and defers to [`Log::parse_header`] — the same validity gate the
1448/// forward parser applies, keeping the reverse reader's boundary set identical
1449/// to the forward one.
1450fn is_valid_header_line(buf: &[u8], i: usize) -> bool {
1451 let line_end = buf[i..]
1452 .iter()
1453 .position(|&b| b == b'\n')
1454 .map(|p| i + p)
1455 .unwrap_or(buf.len());
1456 let line = String::from_utf8_lossy(&buf[i..line_end]);
1457 Log::parse_header(&line).is_some()
1458}
1459
1460/// Extract the text of the entry whose header is at absolute offset
1461/// `header_abs` (the `headers[idx]` entry), spanning to the next header (or
1462/// buffer end). `buf` begins at absolute offset `base`.
1463fn entry_text_at(buf: &[u8], base: u64, header_abs: u64, headers: &[u64], idx: usize) -> String {
1464 let rel_start = (header_abs - base) as usize;
1465 let rel_end = if idx + 1 < headers.len() {
1466 (headers[idx + 1] - base) as usize
1467 } else {
1468 buf.len()
1469 };
1470 String::from_utf8_lossy(&buf[rel_start..rel_end]).into_owned()
1471}
1472
1473/// Parse a single entry from a text block that begins at its header line.
1474fn parse_single_entry(text: &str) -> Option<LogEntry> {
1475 parse_entries(text).into_iter().next()
1476}
1477
1478#[cfg(test)]
1479mod tests {
1480 use super::*;
1481 use crate::parser::Config;
1482 use std::fs;
1483 use tempfile::TempDir;
1484
1485 /// Build a `Store` rooted at a fresh temp dir with a minimal `DB.md`.
1486 /// Construct the `Store` struct directly so the test stays narrow and never
1487 /// exercises the `Store::open` parser path.
1488 fn temp_store() -> (TempDir, Store) {
1489 let dir = tempfile::tempdir().expect("tempdir");
1490 fs::write(dir.path().join("DB.md"), "---\ntype: db-md\n---\n").expect("write DB.md");
1491 let store = Store {
1492 root: dir.path().to_path_buf(),
1493 config: Config::default(),
1494 };
1495 (dir, store)
1496 }
1497
1498 /// Regression (adversarial review): a hand-created / externally-produced
1499 /// archive with an out-of-range month (`00`, `13`..`99`) must NOT parse as a
1500 /// real month archive — otherwise its `(year, 0)` bucket sorts below every
1501 /// legitimate month and the newest-first early-break in `since`/`tail` can
1502 /// silently prune it. Out-of-range → `None` (the caller scans it instead).
1503 #[test]
1504 fn archive_year_month_rejects_out_of_range_months() {
1505 use std::path::Path;
1506 assert_eq!(
1507 archive_year_month(Path::new("log/2026-05.md")),
1508 Some((2026, 5))
1509 );
1510 assert_eq!(
1511 archive_year_month(Path::new("log/2026-01.md")),
1512 Some((2026, 1))
1513 );
1514 assert_eq!(
1515 archive_year_month(Path::new("log/2026-12.md")),
1516 Some((2026, 12))
1517 );
1518 for bad in ["log/2026-00.md", "log/2026-13.md", "log/2026-99.md"] {
1519 assert_eq!(
1520 archive_year_month(Path::new(bad)),
1521 None,
1522 "{bad} has an out-of-range month and must not parse as an archive"
1523 );
1524 }
1525 }
1526
1527 /// A timestamp at UTC from `YYYY-MM-DD HH:MM` components.
1528 fn ts(y: i32, mo: u32, d: u32, h: u32, mi: u32) -> DateTime<FixedOffset> {
1529 let naive = chrono::NaiveDate::from_ymd_opt(y, mo, d)
1530 .unwrap()
1531 .and_hms_opt(h, mi, 0)
1532 .unwrap();
1533 FixedOffset::east_opt(0)
1534 .unwrap()
1535 .from_local_datetime(&naive)
1536 .single()
1537 .unwrap()
1538 }
1539
1540 #[allow(clippy::too_many_arguments)] // test fixture builder; struct-ifying churns every call site
1541 fn entry(
1542 y: i32,
1543 mo: u32,
1544 d: u32,
1545 h: u32,
1546 mi: u32,
1547 kind: LogKind,
1548 object: Option<&str>,
1549 note: &str,
1550 ) -> LogEntry {
1551 LogEntry {
1552 timestamp: ts(y, mo, d, h, mi),
1553 kind,
1554 object: object.map(|s| s.to_string()),
1555 note: note.to_string(),
1556 }
1557 }
1558
1559 // ── parse_header ────────────────────────────────────────────────────────
1560
1561 #[test]
1562 fn parse_header_with_object() {
1563 let (t, k, o) =
1564 Log::parse_header("## [2026-05-27 10:00] ingest | sources/emails/x.eml").unwrap();
1565 assert_eq!(t, ts(2026, 5, 27, 10, 0));
1566 assert_eq!(k, LogKind::Ingest);
1567 assert_eq!(o.as_deref(), Some("sources/emails/x.eml"));
1568 }
1569
1570 #[test]
1571 fn parse_header_without_object_is_none_object() {
1572 let (t, k, o) = Log::parse_header("## [2026-05-27 10:20] validate").unwrap();
1573 assert_eq!(t, ts(2026, 5, 27, 10, 20));
1574 assert_eq!(k, LogKind::Validate);
1575 assert_eq!(o, None);
1576 }
1577
1578 #[test]
1579 fn parse_header_custom_kind_roundtrips_token() {
1580 let (_, k, o) = Log::parse_header("## [2026-05-27 10:00] proposal | records/x").unwrap();
1581 assert_eq!(k, LogKind::Custom("proposal".to_string()));
1582 assert!(!k.is_recognized());
1583 assert_eq!(o.as_deref(), Some("records/x"));
1584 }
1585
1586 #[test]
1587 fn parse_header_index_rebuild_hyphenated_kind() {
1588 let (_, k, _) = Log::parse_header("## [2026-05-27 10:00] index-rebuild").unwrap();
1589 assert_eq!(k, LogKind::IndexRebuild);
1590 assert_eq!(k.as_str(), "index-rebuild");
1591 }
1592
1593 #[test]
1594 fn parse_header_rejects_non_headers() {
1595 assert!(Log::parse_header("Not a header").is_none());
1596 assert!(Log::parse_header("# Curator log").is_none());
1597 assert!(Log::parse_header("## [garbage] ingest | x").is_none());
1598 assert!(Log::parse_header("## [2026-05-27 10:00]").is_none()); // no kind
1599 // A bracketed but non-timestamp date must be rejected (LOG_BAD_TIMESTAMP territory).
1600 assert!(Log::parse_header("## [2026-13-40 99:99] ingest | x").is_none());
1601 }
1602
1603 // ── kind round-trip ───────────────────────────────────────────────────────
1604
1605 #[test]
1606 fn kind_as_str_parse_roundtrip_for_all_recognized() {
1607 for k in [
1608 LogKind::Ingest,
1609 LogKind::Create,
1610 LogKind::Update,
1611 LogKind::Delete,
1612 LogKind::Rename,
1613 LogKind::Link,
1614 LogKind::Validate,
1615 LogKind::IndexRebuild,
1616 LogKind::Contradiction,
1617 ] {
1618 assert_eq!(LogKind::parse(k.as_str()), k);
1619 assert!(k.is_recognized());
1620 }
1621 }
1622
1623 // ── append: creation + frontmatter ───────────────────────────────────────
1624
1625 #[test]
1626 fn append_creates_log_with_frontmatter_and_entry() {
1627 let (_d, store) = temp_store();
1628 let e = entry(
1629 2026,
1630 5,
1631 27,
1632 10,
1633 0,
1634 LogKind::Ingest,
1635 Some("sources/emails/x.eml"),
1636 "Email received.",
1637 );
1638 Log::append(&store, &e).unwrap();
1639
1640 let content = fs::read_to_string(store.root.join("log.md")).unwrap();
1641 // type: log frontmatter present.
1642 assert!(
1643 content.starts_with("---\ntype: log\n---\n"),
1644 "missing log frontmatter; got:\n{content}"
1645 );
1646 // The entry header is rendered verbatim.
1647 assert!(content.contains("## [2026-05-27 10:00] ingest | sources/emails/x.eml"));
1648 assert!(content.contains("Email received."));
1649 // No archive dir created when nothing rotates.
1650 assert!(!store.root.join("log").exists());
1651 }
1652
1653 // ── append → tail → since round-trip ─────────────────────────────────────
1654
1655 #[test]
1656 fn append_tail_since_roundtrip() {
1657 let (_d, store) = temp_store();
1658 let e1 = entry(2026, 5, 27, 10, 0, LogKind::Ingest, Some("a"), "first");
1659 let e2 = entry(2026, 5, 27, 10, 5, LogKind::Create, Some("b"), "second");
1660 let e3 = entry(2026, 5, 27, 10, 10, LogKind::Update, Some("c"), "third");
1661 Log::append(&store, &e1).unwrap();
1662 Log::append(&store, &e2).unwrap();
1663 Log::append(&store, &e3).unwrap();
1664
1665 // tail(2) returns the two newest, in chronological order.
1666 let tail = Log::tail(&store, 2).unwrap();
1667 assert_eq!(tail.len(), 2);
1668 assert_eq!(tail[0], e2);
1669 assert_eq!(tail[1], e3);
1670
1671 // tail(n) larger than the log returns everything, chronologically.
1672 let all = Log::tail(&store, 99).unwrap();
1673 assert_eq!(all, vec![e1.clone(), e2.clone(), e3.clone()]);
1674
1675 // since(10:05) returns strictly-newer entries (excludes the 10:05 one).
1676 let since = Log::since(&store, ts(2026, 5, 27, 10, 5)).unwrap();
1677 assert_eq!(since, vec![e3.clone()]);
1678
1679 // since before everything returns all.
1680 let since_all = Log::since(&store, ts(2026, 5, 27, 9, 0)).unwrap();
1681 assert_eq!(since_all, vec![e1, e2, e3]);
1682 }
1683
1684 #[test]
1685 fn tail_zero_is_empty() {
1686 let (_d, store) = temp_store();
1687 Log::append(
1688 &store,
1689 &entry(2026, 5, 27, 10, 0, LogKind::Ingest, Some("a"), "x"),
1690 )
1691 .unwrap();
1692 assert!(Log::tail(&store, 0).unwrap().is_empty());
1693 }
1694
1695 #[test]
1696 fn tail_and_since_on_missing_log_are_empty() {
1697 let (_d, store) = temp_store();
1698 assert!(Log::tail(&store, 5).unwrap().is_empty());
1699 assert!(Log::since(&store, ts(2000, 1, 1, 0, 0)).unwrap().is_empty());
1700 assert!(Log::last_validate_at(&store).unwrap().is_none());
1701 }
1702
1703 #[test]
1704 fn since_exact_timestamp_is_exclusive() {
1705 let (_d, store) = temp_store();
1706 let e = entry(2026, 5, 27, 10, 0, LogKind::Validate, None, "PASS");
1707 Log::append(&store, &e).unwrap();
1708 // Equal timestamp must NOT be included (strictly newer).
1709 assert!(Log::since(&store, ts(2026, 5, 27, 10, 0))
1710 .unwrap()
1711 .is_empty());
1712 }
1713
1714 // ── since: out-of-order on disk (append-only correction / merge=union) ────
1715
1716 /// Write a `log.md` at the store root from `entries` in the EXACT given
1717 /// physical order, with the standard `type: log` frontmatter. Unlike
1718 /// [`Log::append`] (which always lands the newest entry at EOF), this lets a
1719 /// test author the non-monotonic on-disk shape the SPEC permits — a
1720 /// backdated corrective entry below the entry it corrects, or a
1721 /// `merge=union` interleave.
1722 fn write_raw_log(store: &Store, entries: &[LogEntry]) {
1723 let mut content = String::from(LOG_FRONTMATTER);
1724 content.push('\n');
1725 for e in entries {
1726 content.push_str(&e.render());
1727 }
1728 fs::write(store.root.join("log.md"), content).expect("write raw log.md");
1729 }
1730
1731 #[test]
1732 fn since_returns_newer_entries_even_when_disk_order_is_non_monotonic() {
1733 // The demonstrated regression: a curator appended a backdated CORRECTIVE
1734 // entry (10:00) below newer entries (10:10, 10:05), so the physical
1735 // on-disk order is 10:10, 10:05, 10:00 — newest-first, not chronological.
1736 // The append-only SPEC explicitly permits this ("append a corrective
1737 // entry below it"; out-of-order is only LOG_OUT_OF_ORDER, a warning).
1738 let (_d, store) = temp_store();
1739 let e_1010 = entry(2026, 5, 27, 10, 10, LogKind::Update, Some("c"), "newest");
1740 let e_1005 = entry(2026, 5, 27, 10, 5, LogKind::Create, Some("b"), "middle");
1741 let e_1000 = entry(
1742 2026,
1743 5,
1744 27,
1745 10,
1746 0,
1747 LogKind::Update,
1748 Some("a"),
1749 "backdated fix",
1750 );
1751 // Physical order on disk: 10:10, 10:05, then the backdated 10:00 LAST.
1752 write_raw_log(&store, &[e_1010, e_1005, e_1000]);
1753
1754 // since 10:02 must return BOTH entries strictly newer than 10:02
1755 // (10:05 and 10:10). The old early-stop hit the physically-last 10:00
1756 // entry (<= 10:02), stopped, and returned EMPTY — silently dropping the
1757 // two newer entries that sit earlier in the file.
1758 let got = Log::since(&store, ts(2026, 5, 27, 10, 2)).unwrap();
1759 let stamps: std::collections::BTreeSet<_> = got.iter().map(|e| e.timestamp).collect();
1760 assert_eq!(
1761 stamps,
1762 [ts(2026, 5, 27, 10, 5), ts(2026, 5, 27, 10, 10)]
1763 .into_iter()
1764 .collect(),
1765 "since(10:02) must include both 10:05 and 10:10 despite the backdated \
1766 10:00 entry sitting physically last, and exclude 10:00; got {got:?}"
1767 );
1768
1769 // A cutoff before everything still returns all three, regardless of the
1770 // scrambled disk order.
1771 let all = Log::since(&store, ts(2026, 5, 27, 9, 0)).unwrap();
1772 let all_stamps: std::collections::BTreeSet<_> = all.iter().map(|e| e.timestamp).collect();
1773 assert_eq!(
1774 all_stamps,
1775 [
1776 ts(2026, 5, 27, 10, 0),
1777 ts(2026, 5, 27, 10, 5),
1778 ts(2026, 5, 27, 10, 10),
1779 ]
1780 .into_iter()
1781 .collect()
1782 );
1783 }
1784
1785 #[test]
1786 fn since_crosses_archive_when_newer_entry_is_out_of_order_inside_it() {
1787 // Out-of-order INSIDE an archive month, with the cutoff landing in that
1788 // month. The April archive is authored newest-physical-first (04-20,
1789 // then a backdated 04-05 last); a naive early-stop on the first
1790 // older-than-cutoff entry would miss the later April entry. The active
1791 // file holds a clean May entry. Cutoff = mid-April.
1792 let (_d, store) = temp_store();
1793
1794 // Active file: one current-month (May) entry.
1795 let may = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may1");
1796 write_raw_log(&store, &[may]);
1797
1798 // April archive authored out of order: 04-20 first, backdated 04-05 last.
1799 let apr_late = entry(
1800 2026,
1801 4,
1802 20,
1803 9,
1804 0,
1805 LogKind::Create,
1806 Some("apr-b"),
1807 "apr-late",
1808 );
1809 let apr_early = entry(
1810 2026,
1811 4,
1812 5,
1813 9,
1814 0,
1815 LogKind::Ingest,
1816 Some("apr-a"),
1817 "apr-early",
1818 );
1819 let dir = store.root.join("log");
1820 fs::create_dir_all(&dir).unwrap();
1821 let mut arch = String::from(LOG_FRONTMATTER);
1822 arch.push('\n');
1823 arch.push_str(&apr_late.render());
1824 arch.push_str(&apr_early.render());
1825 fs::write(dir.join("2026-04.md"), arch).unwrap();
1826
1827 // since mid-April: the later April entry (04-20) AND the May entry must
1828 // come back; the early April entry (04-05) must not.
1829 let got = Log::since(&store, ts(2026, 4, 15, 0, 0)).unwrap();
1830 let stamps: std::collections::BTreeSet<_> = got.iter().map(|e| e.timestamp).collect();
1831 assert_eq!(
1832 stamps,
1833 [ts(2026, 4, 20, 9, 0), ts(2026, 5, 2, 8, 0)]
1834 .into_iter()
1835 .collect(),
1836 "since(mid-April) must include the out-of-order later April entry \
1837 and the May entry, and exclude the earlier April entry; got {got:?}"
1838 );
1839 }
1840
1841 // ── multi-line notes ──────────────────────────────────────────────────────
1842
1843 #[test]
1844 fn multiline_note_is_preserved() {
1845 let (_d, store) = temp_store();
1846 let e = entry(
1847 2026,
1848 5,
1849 27,
1850 10,
1851 0,
1852 LogKind::Create,
1853 Some("records/x"),
1854 "Line one.\nLine two.\nLine three.",
1855 );
1856 Log::append(&store, &e).unwrap();
1857 let got = Log::tail(&store, 1).unwrap();
1858 assert_eq!(got[0].note, "Line one.\nLine two.\nLine three.");
1859 }
1860
1861 #[test]
1862 fn empty_note_roundtrips_as_empty() {
1863 let (_d, store) = temp_store();
1864 let e = entry(2026, 5, 27, 10, 0, LogKind::Validate, None, "");
1865 Log::append(&store, &e).unwrap();
1866 let got = Log::tail(&store, 1).unwrap();
1867 assert_eq!(got[0], e);
1868 assert_eq!(got[0].note, "");
1869 }
1870
1871 // ── last_validate_at ─────────────────────────────────────────────────────
1872
1873 #[test]
1874 fn last_validate_at_finds_most_recent_validate() {
1875 let (_d, store) = temp_store();
1876 Log::append(
1877 &store,
1878 &entry(2026, 5, 27, 10, 0, LogKind::Validate, None, "first pass"),
1879 )
1880 .unwrap();
1881 Log::append(
1882 &store,
1883 &entry(2026, 5, 27, 10, 5, LogKind::Create, Some("a"), "made a"),
1884 )
1885 .unwrap();
1886 Log::append(
1887 &store,
1888 &entry(2026, 5, 27, 10, 10, LogKind::Validate, None, "second pass"),
1889 )
1890 .unwrap();
1891 Log::append(
1892 &store,
1893 &entry(2026, 5, 27, 10, 15, LogKind::Update, Some("a"), "edit a"),
1894 )
1895 .unwrap();
1896
1897 let last = Log::last_validate_at(&store).unwrap();
1898 assert_eq!(last, Some(ts(2026, 5, 27, 10, 10)));
1899 }
1900
1901 #[test]
1902 fn last_validate_at_none_when_no_validate() {
1903 let (_d, store) = temp_store();
1904 Log::append(
1905 &store,
1906 &entry(2026, 5, 27, 10, 0, LogKind::Create, Some("a"), "x"),
1907 )
1908 .unwrap();
1909 assert_eq!(Log::last_validate_at(&store).unwrap(), None);
1910 }
1911
1912 // ── month-boundary rotation ──────────────────────────────────────────────
1913
1914 #[test]
1915 fn rotation_rolls_prior_months_into_archives() {
1916 let (_d, store) = temp_store();
1917 // Two April entries and one May entry, all written while "current" was
1918 // their own month (append-only chronological order).
1919 let a1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr one");
1920 let a2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr two");
1921 Log::append(&store, &a1).unwrap();
1922 Log::append(&store, &a2).unwrap();
1923
1924 // Before rotation: no archive dir, both April entries in active.
1925 assert!(!store.root.join("log").exists());
1926
1927 // Appending a May entry must roll April into log/2026-04.md.
1928 let m1 = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may one");
1929 Log::append(&store, &m1).unwrap();
1930
1931 // Archive exists and holds both April entries with frontmatter.
1932 let arch_path = store.root.join("log").join("2026-04.md");
1933 assert!(arch_path.exists(), "expected April archive to be created");
1934 let arch = fs::read_to_string(&arch_path).unwrap();
1935 assert!(arch.starts_with("---\ntype: log\n---\n"));
1936 assert!(arch.contains("## [2026-04-10 09:00] ingest | apr-a"));
1937 assert!(arch.contains("## [2026-04-20 09:00] create | apr-b"));
1938 assert!(arch.contains("apr one"));
1939 assert!(arch.contains("apr two"));
1940
1941 // Active file now holds ONLY the May entry (no April entries).
1942 let active = fs::read_to_string(store.root.join("log.md")).unwrap();
1943 assert!(active.contains("## [2026-05-02 08:00] update | may-a"));
1944 assert!(
1945 !active.contains("apr-a") && !active.contains("apr-b"),
1946 "April entries must be gone from the active file; got:\n{active}"
1947 );
1948
1949 // The full timeline (archives ++ active) is intact and chronological.
1950 let all = Log::tail(&store, 99).unwrap();
1951 assert_eq!(all, vec![a1, a2, m1]);
1952 }
1953
1954 #[test]
1955 fn rotation_groups_distinct_prior_months_into_separate_archives() {
1956 let (_d, store) = temp_store();
1957 // March + April entries accumulate, then a May append rolls BOTH prior
1958 // months into their own archive files.
1959 let mar = entry(2026, 3, 5, 9, 0, LogKind::Ingest, Some("mar"), "march");
1960 let apr = entry(2026, 4, 5, 9, 0, LogKind::Create, Some("apr"), "april");
1961 Log::append(&store, &mar).unwrap();
1962 Log::append(&store, &apr).unwrap();
1963 // At this point April is current, March already rolled into its archive.
1964 assert!(store.root.join("log").join("2026-03.md").exists());
1965
1966 let may = entry(2026, 5, 5, 9, 0, LogKind::Update, Some("may"), "may");
1967 Log::append(&store, &may).unwrap();
1968
1969 assert!(store.root.join("log").join("2026-03.md").exists());
1970 assert!(store.root.join("log").join("2026-04.md").exists());
1971
1972 // Each archive holds only its own month.
1973 let mar_arch = fs::read_to_string(store.root.join("log").join("2026-03.md")).unwrap();
1974 let apr_arch = fs::read_to_string(store.root.join("log").join("2026-04.md")).unwrap();
1975 assert!(mar_arch.contains("mar") && !mar_arch.contains("apr"));
1976 assert!(apr_arch.contains("apr") && !apr_arch.contains("mar"));
1977
1978 // Active holds only May.
1979 let active = fs::read_to_string(store.root.join("log.md")).unwrap();
1980 assert!(active.contains("may") && !active.contains("mar") && !active.contains("apr"));
1981
1982 // Timeline intact and ordered across both archives + active.
1983 let all = Log::tail(&store, 99).unwrap();
1984 assert_eq!(all, vec![mar, apr, may]);
1985 }
1986
1987 #[test]
1988 fn tail_crosses_into_archive_when_n_spans_month_boundary() {
1989 let (_d, store) = temp_store();
1990 let a1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr1");
1991 let a2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr2");
1992 let m1 = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may1");
1993 let m2 = entry(2026, 5, 3, 8, 0, LogKind::Update, Some("may-b"), "may2");
1994 for e in [&a1, &a2, &m1, &m2] {
1995 Log::append(&store, e).unwrap();
1996 }
1997 // April is now archived; active holds only May. tail(3) must reach back
1998 // into the archive for the third-newest entry.
1999 let tail3 = Log::tail(&store, 3).unwrap();
2000 assert_eq!(tail3, vec![a2.clone(), m1.clone(), m2.clone()]);
2001
2002 // tail within the active month does NOT need the archive but is still
2003 // correct.
2004 let tail2 = Log::tail(&store, 2).unwrap();
2005 assert_eq!(tail2, vec![m1, m2]);
2006 }
2007
2008 #[test]
2009 fn since_crosses_into_archive_and_early_stops() {
2010 let (_d, store) = temp_store();
2011 let a1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr1");
2012 let a2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr2");
2013 let m1 = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may1");
2014 for e in [&a1, &a2, &m1] {
2015 Log::append(&store, e).unwrap();
2016 }
2017 // since a mid-April time: must include the later April entry (from the
2018 // archive) and the May entry, but not the earlier April one.
2019 let got = Log::since(&store, ts(2026, 4, 15, 0, 0)).unwrap();
2020 assert_eq!(got, vec![a2, m1]);
2021 }
2022
2023 #[test]
2024 fn last_validate_at_crosses_into_archive() {
2025 let (_d, store) = temp_store();
2026 // A validate in April, then non-validate work that rolls April away.
2027 Log::append(
2028 &store,
2029 &entry(2026, 4, 10, 9, 0, LogKind::Validate, None, "apr validate"),
2030 )
2031 .unwrap();
2032 Log::append(
2033 &store,
2034 &entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may work"),
2035 )
2036 .unwrap();
2037 // Active has only the May update; the most-recent validate lives in the
2038 // April archive and must still be found.
2039 let last = Log::last_validate_at(&store).unwrap();
2040 assert_eq!(last, Some(ts(2026, 4, 10, 9, 0)));
2041 }
2042
2043 // ── reverse-read correctness on a large (multi-block) log ────────────────
2044
2045 #[test]
2046 fn reverse_read_correct_on_large_single_month_log() {
2047 let (_d, store) = temp_store();
2048 // Append many same-month entries with chunky multi-line notes so the
2049 // file spans well past one REVERSE_BLOCK (8 KiB). Timestamps are
2050 // strictly increasing (a real append-only log is monotonic): each entry
2051 // is 3 minutes after the previous, all within June, so physical order
2052 // equals chronological order and the last-k-physical ARE the k-newest.
2053 let n = 400usize;
2054 let mut expected: Vec<LogEntry> = Vec::new();
2055 for i in 0..n {
2056 let total_min = (i as u32) * 3;
2057 let day = 1 + total_min / (24 * 60);
2058 let hour = (total_min / 60) % 24;
2059 let min = total_min % 60;
2060 // Unique, multi-line note to bulk up the file and detect mis-parses.
2061 let note = format!(
2062 "entry number {i}\nbody line A for {i}\nbody line B for {i} with padding {}",
2063 "x".repeat(40)
2064 );
2065 let e = entry(
2066 2026,
2067 6,
2068 day,
2069 hour,
2070 min,
2071 LogKind::Update,
2072 Some(&format!("records/item-{i:04}")),
2073 ¬e,
2074 );
2075 Log::append(&store, &e).unwrap();
2076 expected.push(e);
2077 }
2078
2079 // File must actually be multi-block to exercise the backward reader.
2080 let size = fs::metadata(store.root.join("log.md")).unwrap().len();
2081 assert!(
2082 size > (REVERSE_BLOCK as u64) * 2,
2083 "test log not large enough ({size} bytes) to exercise multi-block reverse-read"
2084 );
2085
2086 // tail(5) must equal the 5 newest, exactly.
2087 let tail5 = Log::tail(&store, 5).unwrap();
2088 assert_eq!(tail5, expected[n - 5..].to_vec());
2089
2090 // tail(50) must equal the 50 newest.
2091 let tail50 = Log::tail(&store, 50).unwrap();
2092 assert_eq!(tail50, expected[n - 50..].to_vec());
2093
2094 // tail(all) must reconstruct the whole timeline in order.
2095 let all = Log::tail(&store, n + 10).unwrap();
2096 assert_eq!(all.len(), n);
2097 assert_eq!(all, expected);
2098 }
2099
2100 // ── tail on OUT-OF-ORDER logs (newest-by-timestamp, not last-physical) ────
2101 //
2102 // The append-only contract is non-decreasing time order, but it's only a
2103 // `LOG_OUT_OF_ORDER` warning when violated (corrective entries land below
2104 // the entry they correct; backdated / clock-skewed writes; `merge=union`
2105 // clone merges). `tail N` must return the N newest *by timestamp*, never the
2106 // last N *physical* entries.
2107
2108 /// Write `log.md` verbatim from rendered entries in the given **physical
2109 /// (file) order**, bypassing `Log::append` so the test controls on-disk
2110 /// order exactly (append never reorders within a month, but this is the
2111 /// clearest way to pin a specific physical layout).
2112 fn write_log_physical(store: &Store, entries: &[LogEntry]) {
2113 let mut body = String::new();
2114 for e in entries {
2115 body.push_str(&e.render());
2116 }
2117 let full = compose_active(LOG_FRONTMATTER, &body);
2118 fs::write(store.root.join("log.md"), full).expect("write log.md");
2119 }
2120
2121 #[test]
2122 fn tail_returns_newest_by_timestamp_on_demonstrated_out_of_order_log() {
2123 // The exact case from the review finding: physical order 10:10, 10:05,
2124 // 10:00 (a backdated entry tail). The OLD code returned the last two
2125 // physical entries {10:05, 10:00}; the correct answer is the two newest
2126 // by time {10:05, 10:10}.
2127 let (_d, store) = temp_store();
2128 let e_1010 = entry(2026, 5, 27, 10, 10, LogKind::Update, Some("c"), "ten-ten");
2129 let e_1005 = entry(
2130 2026,
2131 5,
2132 27,
2133 10,
2134 5,
2135 LogKind::Create,
2136 Some("b"),
2137 "ten-oh-five",
2138 );
2139 let e_1000 = entry(2026, 5, 27, 10, 0, LogKind::Ingest, Some("a"), "ten-oh-oh");
2140 // Physical order: newest first, then the two older ones — out of order.
2141 write_log_physical(&store, &[e_1010.clone(), e_1005.clone(), e_1000.clone()]);
2142
2143 let tail2 = Log::tail(&store, 2).unwrap();
2144 assert_eq!(
2145 tail2,
2146 vec![e_1005.clone(), e_1010.clone()],
2147 "tail(2) must be the two NEWEST by timestamp (chronological), \
2148 not the last two physical entries"
2149 );
2150 // The newest entry must be present and the oldest absent.
2151 assert!(tail2.contains(&e_1010), "newest (10:10) must be included");
2152 assert!(!tail2.contains(&e_1000), "oldest (10:00) must be excluded");
2153
2154 // tail(1) is just the single newest.
2155 assert_eq!(Log::tail(&store, 1).unwrap(), vec![e_1010.clone()]);
2156 // tail(all) is the full set in chronological order.
2157 assert_eq!(Log::tail(&store, 99).unwrap(), vec![e_1000, e_1005, e_1010]);
2158 }
2159
2160 #[test]
2161 fn tail_no_early_stop_when_newer_entry_sits_before_an_older_one() {
2162 // Guards the unsound within-file early stop: a newer entry (10:50) sits
2163 // PHYSICALLY BEFORE a much older one (10:00). Reading newest-physical-
2164 // first, the scan meets 10:00 before 10:50; any "stop at the first entry
2165 // below the window minimum" rule would bail and drop 10:50.
2166 //
2167 // Physical (top→bottom): 10:55, 10:10, 10:50, 10:00.
2168 // Reverse-scan order: 10:00, 10:50, 10:10, 10:55.
2169 let (_d, store) = temp_store();
2170 let e55 = entry(2026, 5, 27, 10, 55, LogKind::Update, Some("x55"), "55");
2171 let e10 = entry(2026, 5, 27, 10, 10, LogKind::Update, Some("x10"), "10");
2172 let e50 = entry(2026, 5, 27, 10, 50, LogKind::Update, Some("x50"), "50");
2173 let e00 = entry(2026, 5, 27, 10, 0, LogKind::Update, Some("x00"), "00");
2174 write_log_physical(
2175 &store,
2176 &[e55.clone(), e10.clone(), e50.clone(), e00.clone()],
2177 );
2178
2179 // The two newest by timestamp are 10:55 and 10:50 — NOT the early-stop
2180 // victim 10:10, and NOT the last-physical 10:00.
2181 let tail2 = Log::tail(&store, 2).unwrap();
2182 assert_eq!(tail2, vec![e50.clone(), e55.clone()]);
2183
2184 let tail3 = Log::tail(&store, 3).unwrap();
2185 assert_eq!(tail3, vec![e10.clone(), e50.clone(), e55.clone()]);
2186 }
2187
2188 #[test]
2189 fn tail_orders_equal_timestamps_by_physical_recency() {
2190 // Three entries share 10:00; one is at 09:59. tail(2) must keep both
2191 // 10:00 entries, and among the equal pair the one appended LATER
2192 // (physically last) sorts last ("newest" = most-recently recorded).
2193 let (_d, store) = temp_store();
2194 let early = entry(2026, 5, 27, 9, 59, LogKind::Create, Some("early"), "before");
2195 let tie_a = entry(
2196 2026,
2197 5,
2198 27,
2199 10,
2200 0,
2201 LogKind::Update,
2202 Some("tie-a"),
2203 "first 10:00",
2204 );
2205 let tie_b = entry(
2206 2026,
2207 5,
2208 27,
2209 10,
2210 0,
2211 LogKind::Update,
2212 Some("tie-b"),
2213 "second 10:00",
2214 );
2215 // Physical append order: early, tie_a, tie_b.
2216 write_log_physical(&store, &[early.clone(), tie_a.clone(), tie_b.clone()]);
2217
2218 let tail2 = Log::tail(&store, 2).unwrap();
2219 assert_eq!(
2220 tail2,
2221 vec![tie_a.clone(), tie_b.clone()],
2222 "both 10:00 entries kept, physically-later one (tie_b) last; 09:59 dropped"
2223 );
2224 // tail(1) keeps only the most-recently-recorded of the equal pair.
2225 assert_eq!(Log::tail(&store, 1).unwrap(), vec![tie_b]);
2226 }
2227
2228 #[test]
2229 fn tail_finds_newest_across_a_backdated_entry_spanning_the_month_boundary() {
2230 // A backdated entry can land physically after newer entries even across
2231 // a rotation: append May entries, then a June entry (rolls May to its
2232 // archive), then append a May-dated correction — it goes into the ACTIVE
2233 // file, physically after June. tail must still rank by timestamp, so the
2234 // June entry stays newest and the backdated May entry is not mistaken
2235 // for the tail.
2236 let (_d, store) = temp_store();
2237 let may1 = entry(2026, 5, 10, 9, 0, LogKind::Ingest, Some("may-1"), "may one");
2238 let may2 = entry(2026, 5, 20, 9, 0, LogKind::Create, Some("may-2"), "may two");
2239 let jun1 = entry(2026, 6, 2, 8, 0, LogKind::Update, Some("jun-1"), "jun one");
2240 Log::append(&store, &may1).unwrap();
2241 Log::append(&store, &may2).unwrap();
2242 Log::append(&store, &jun1).unwrap(); // rotates May -> log/2026-05.md
2243 assert!(store.root.join("log").join("2026-05.md").exists());
2244
2245 // A backdated May correction, appended now: it lands in the active file
2246 // (its month May is not strictly before the active month June), so the
2247 // active file is physically [jun1, may_corr] — out of order.
2248 let may_corr = entry(
2249 2026,
2250 5,
2251 25,
2252 9,
2253 0,
2254 LogKind::Update,
2255 Some("may-2"),
2256 "may correction",
2257 );
2258 Log::append(&store, &may_corr).unwrap();
2259 let active = fs::read_to_string(store.root.join("log.md")).unwrap();
2260 assert!(
2261 active.contains("jun-1") && active.contains("may correction"),
2262 "backdated May entry should be in the active file alongside June; got:\n{active}"
2263 );
2264
2265 // The single newest by timestamp is the June entry, even though the
2266 // backdated May entry is physically last.
2267 assert_eq!(Log::tail(&store, 1).unwrap(), vec![jun1.clone()]);
2268
2269 // tail(2): the two newest by time are may_corr (05-25) and jun1 (06-02).
2270 let tail2 = Log::tail(&store, 2).unwrap();
2271 assert_eq!(tail2, vec![may_corr.clone(), jun1.clone()]);
2272
2273 // tail(3) must reach into the May archive for the third-newest (may2,
2274 // 05-20), proving archive crossing still works on an out-of-order store.
2275 let tail3 = Log::tail(&store, 3).unwrap();
2276 assert_eq!(tail3, vec![may2.clone(), may_corr.clone(), jun1.clone()]);
2277
2278 // tail(all) reconstructs the whole timeline in chronological order.
2279 let all = Log::tail(&store, 99).unwrap();
2280 assert_eq!(all, vec![may1, may2, may_corr, jun1]);
2281 }
2282
2283 #[test]
2284 fn parse_entries_skips_unparseable_header_folding_into_body() {
2285 // A `## [` line that is NOT a valid header should not start a new entry;
2286 // it folds into the preceding entry's note. This guards the
2287 // parse_entries header-validation branch.
2288 let text = "\
2289## [2026-05-27 10:00] create | records/x
2290Body mentions a literal: ## [not a real header here]
2291More body.
2292
2293## [2026-05-27 10:05] update | records/y
2294Second.
2295";
2296 let entries = parse_entries(text);
2297 assert_eq!(entries.len(), 2);
2298 assert_eq!(entries[0].kind, LogKind::Create);
2299 assert!(entries[0].note.contains("## [not a real header here]"));
2300 assert!(entries[0].note.contains("More body."));
2301 assert_eq!(entries[1].kind, LogKind::Update);
2302 assert_eq!(entries[1].note, "Second.");
2303 }
2304
2305 // ── append-only: corrective entries go on the end ─────────────────────────
2306
2307 #[test]
2308 fn append_only_corrective_entry_goes_on_end_without_rewriting() {
2309 let (_d, store) = temp_store();
2310 let original = entry(
2311 2026,
2312 5,
2313 27,
2314 10,
2315 0,
2316 LogKind::Update,
2317 Some("records/northstar"),
2318 "Seat count 120 -> 175.",
2319 );
2320 Log::append(&store, &original).unwrap();
2321 let after_first = fs::read_to_string(store.root.join("log.md")).unwrap();
2322
2323 // A correction is a NEW entry appended on the end; the original text is
2324 // left byte-for-byte intact (append-only contract: no rewrite API).
2325 let correction = entry(
2326 2026,
2327 5,
2328 27,
2329 11,
2330 0,
2331 LogKind::Update,
2332 Some("records/northstar"),
2333 "Correction: seat count is 165, not 175.",
2334 );
2335 Log::append(&store, &correction).unwrap();
2336 let after_second = fs::read_to_string(store.root.join("log.md")).unwrap();
2337
2338 assert!(
2339 after_second.starts_with(&after_first),
2340 "appending must not rewrite earlier bytes"
2341 );
2342 assert!(after_second.contains("Correction: seat count is 165, not 175."));
2343
2344 // Both entries are readable, in order.
2345 let all = Log::tail(&store, 99).unwrap();
2346 assert_eq!(all, vec![original, correction]);
2347 }
2348
2349 // ── concurrent append safety (atomic via temp-file rename) ────────────────
2350
2351 #[test]
2352 fn concurrent_appends_are_atomic_and_total() {
2353 use std::sync::{Arc, Barrier};
2354 use std::thread;
2355
2356 let (_d, store) = temp_store();
2357 // Seed the file so all threads take the read-modify-write path.
2358 Log::append(
2359 &store,
2360 &entry(2026, 7, 1, 0, 0, LogKind::Create, Some("seed"), "seed"),
2361 )
2362 .unwrap();
2363
2364 let threads = 8usize;
2365 let per = 25usize;
2366 let barrier = Arc::new(Barrier::new(threads));
2367 let store = Arc::new(store);
2368
2369 let mut handles = Vec::new();
2370 for tnum in 0..threads {
2371 let b = Arc::clone(&barrier);
2372 let s = Arc::clone(&store);
2373 handles.push(thread::spawn(move || {
2374 b.wait();
2375 for i in 0..per {
2376 let e = entry(
2377 2026,
2378 7,
2379 1,
2380 (tnum % 24) as u32,
2381 (i % 60) as u32,
2382 LogKind::Update,
2383 Some(&format!("t{tnum}-i{i}")),
2384 &format!("thread {tnum} item {i}"),
2385 );
2386 Log::append(&s, &e).unwrap();
2387 }
2388 }));
2389 }
2390 for h in handles {
2391 h.join().unwrap();
2392 }
2393
2394 // The atomic temp-file-rename write means no append truncates or
2395 // corrupts another: the file must remain parseable and every line of
2396 // every entry header must be well-formed. Crucially, no entry should be
2397 // lost to a torn write of the *content already on disk* — though
2398 // interleaved read-modify-write WILL drop some appends (last-writer-
2399 // wins on the snapshot). We therefore assert integrity + that the file
2400 // never went empty / corrupt, not an exact count.
2401 let content = fs::read_to_string(store.root.join("log.md")).unwrap();
2402 assert!(content.starts_with("---\ntype: log\n---\n"));
2403
2404 // Every `## [` line must parse as a valid header (no half-written line).
2405 for line in content.lines() {
2406 if line.starts_with("## [") {
2407 assert!(
2408 Log::parse_header(line).is_some(),
2409 "corrupt/torn header line on disk: {line:?}"
2410 );
2411 }
2412 }
2413
2414 // The seed entry must survive (it was written before the race and
2415 // every snapshot included it).
2416 assert!(content.contains("## [2026-07-01 00:00] create | seed"));
2417
2418 // The reverse reader must still produce a clean, fully-parseable view.
2419 let all = Log::tail(&store, 10_000).unwrap();
2420 assert!(!all.is_empty());
2421 // No duplicate adjacent identical headers from a torn write: every
2422 // returned entry must have a recognized-or-custom kind and a parseable
2423 // timestamp (already guaranteed by parse), and the list must be
2424 // internally consistent (re-render → re-parse identity for each).
2425 for e in &all {
2426 let rendered = e.render();
2427 let reparsed = parse_single_entry(&rendered).unwrap();
2428 assert_eq!(&reparsed, e);
2429 }
2430 }
2431
2432 // ── render/parse identity ────────────────────────────────────────────────
2433
2434 #[test]
2435 fn render_then_parse_is_identity() {
2436 let cases = vec![
2437 entry(
2438 2026,
2439 1,
2440 2,
2441 3,
2442 4,
2443 LogKind::Ingest,
2444 Some("sources/a.eml"),
2445 "n",
2446 ),
2447 entry(
2448 2026,
2449 12,
2450 31,
2451 23,
2452 59,
2453 LogKind::Validate,
2454 None,
2455 "PASS - 0 errors",
2456 ),
2457 entry(
2458 2026,
2459 6,
2460 15,
2461 12,
2462 30,
2463 LogKind::Custom("proposal".to_string()),
2464 Some("records/p"),
2465 "multi\nline\nnote",
2466 ),
2467 entry(2026, 6, 15, 12, 30, LogKind::Contradiction, Some("obj"), ""),
2468 ];
2469 for e in cases {
2470 let rendered = e.render();
2471 let parsed = parse_single_entry(&rendered).unwrap_or_else(|| {
2472 panic!("failed to reparse rendered entry:\n{rendered}");
2473 });
2474 assert_eq!(parsed, e, "round-trip mismatch for {e:?}");
2475 }
2476 }
2477
2478 // ── regression: rotation re-roll must not duplicate archive entries (#3) ──
2479
2480 /// Count occurrences of `needle` in `haystack` (non-overlapping).
2481 fn count_occurrences(haystack: &str, needle: &str) -> usize {
2482 haystack.matches(needle).count()
2483 }
2484
2485 #[test]
2486 fn regression_archive_reroll_is_idempotent_after_interrupted_rotation() {
2487 // Reconstructs the finding's exact failure window: rotation is two
2488 // non-atomic durable writes — (1) roll prior-month entries into the
2489 // archive, then (2) trim the active file. If the process crashes or the
2490 // active rewrite errors AFTER step (1) commits, the prior-month entries
2491 // stay in the untrimmed active file, the agent retries, and the retry
2492 // re-rolls the SAME entries into the archive a second time. The
2493 // mechanism is precisely a second `append_to_archive` of identical
2494 // entries onto an archive that already holds them.
2495 let (_d, store) = temp_store();
2496 let dir = archive_dir(&store);
2497 let arch = archive_path(&store, 2026, 4);
2498
2499 let apr1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr one");
2500 let apr2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr two");
2501 let month = [apr1.clone(), apr2.clone()];
2502
2503 // First roll: a FRESH rotation (no in-progress marker) appends both.
2504 fs::create_dir_all(&dir).unwrap();
2505 append_to_archive(&arch, &month, false).unwrap();
2506
2507 // The retries are crash-RECOVERIES (the in-progress-rotation marker is
2508 // present), so they dedup the re-rolled identical entries to a no-op.
2509 // Pre-fix this blindly concatenated, doubling every entry; do it twice to
2510 // prove the amplification a real retry loop would cause is suppressed.
2511 append_to_archive(&arch, &month, true).unwrap();
2512 append_to_archive(&arch, &month, true).unwrap();
2513
2514 let archived = fs::read_to_string(&arch).unwrap();
2515 // Each entry header must appear EXACTLY once despite the re-rolls.
2516 assert_eq!(
2517 count_occurrences(&archived, "## [2026-04-10 09:00] ingest | apr-a"),
2518 1,
2519 "re-rolled archive duplicated the first April entry; got:\n{archived}"
2520 );
2521 assert_eq!(
2522 count_occurrences(&archived, "## [2026-04-20 09:00] create | apr-b"),
2523 1,
2524 "re-rolled archive duplicated the second April entry; got:\n{archived}"
2525 );
2526
2527 // And the reader surface (`since`) must return each entry once, not the
2528 // duplicated set the pre-fix archive would have yielded.
2529 let got = Log::since(&store, ts(2026, 4, 1, 0, 0)).unwrap();
2530 assert_eq!(
2531 got,
2532 vec![apr1, apr2],
2533 "since over the re-rolled archive must return each April entry once"
2534 );
2535 }
2536
2537 #[test]
2538 fn regression_rotation_reroll_after_active_untrimmed_does_not_duplicate() {
2539 // End-to-end variant driving the real `Log::append` rotation path. We
2540 // rotate April into its archive via a May append, then SIMULATE the
2541 // partial failure by restoring the pre-trim active file (April + May)
2542 // and re-running `append` — exactly the state a crash-between-the-two-
2543 // writes / failed-active-rewrite + agent-retry produces. The archive
2544 // must still hold each April entry once.
2545 let (_d, store) = temp_store();
2546 let apr1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr one");
2547 let apr2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr two");
2548 Log::append(&store, &apr1).unwrap();
2549 Log::append(&store, &apr2).unwrap();
2550
2551 // Snapshot the active file holding both April entries (this is what is
2552 // still on disk if the post-rotation active rewrite never lands).
2553 let active_path = active_log_path(&store);
2554 let pre_rotation_active = fs::read_to_string(&active_path).unwrap();
2555
2556 // A May append rotates April out and trims the active file.
2557 let may = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may-a"), "may one");
2558 Log::append(&store, &may).unwrap();
2559 let arch = archive_path(&store, 2026, 4);
2560 assert!(arch.exists(), "April should have rotated to its archive");
2561
2562 // Simulate the crash/error: the active rewrite never persisted, so the
2563 // active file still contains the (now also archived) April entries.
2564 fs::write(&active_path, &pre_rotation_active).unwrap();
2565 // A real crash leaves the in-progress-rotation marker behind too — it is
2566 // deleted only AFTER the active trim commits. Restore it so the retry is
2567 // recognized as a crash-recovery re-roll (deduped), not a fresh rotation
2568 // (which would correctly append a genuinely-distinct repeat).
2569 fs::write(rotation_marker_path(&store), b"").unwrap();
2570
2571 // The agent retries the append. Re-partitioning sees April as prior
2572 // months again and re-rolls them — which must NOT duplicate the archive.
2573 let may2 = entry(2026, 5, 3, 8, 0, LogKind::Update, Some("may-b"), "may two");
2574 Log::append(&store, &may2).unwrap();
2575
2576 let archived = fs::read_to_string(&arch).unwrap();
2577 assert_eq!(
2578 count_occurrences(&archived, "## [2026-04-10 09:00] ingest | apr-a"),
2579 1,
2580 "retried rotation duplicated an April entry in the archive; got:\n{archived}"
2581 );
2582 assert_eq!(
2583 count_occurrences(&archived, "## [2026-04-20 09:00] create | apr-b"),
2584 1,
2585 "retried rotation duplicated an April entry in the archive; got:\n{archived}"
2586 );
2587 }
2588
2589 /// THE BUG (write side, data-loss). A STALE `.rotating` marker (e.g.
2590 /// committed/synced into a `merge=union` clone after a crash stranded it
2591 /// between the archive write and the active trim) must NOT make a fresh
2592 /// rotation treat a genuinely-distinct same-(minute,kind,object,note) entry
2593 /// as a crash re-roll and silently drop it.
2594 ///
2595 /// On-disk shape (the exact CLI repro): the January archive already holds one
2596 /// `dup` entry (authored independently); the active log holds a February entry
2597 /// AND a SECOND, distinct, byte-identical-at-minute-precision January `dup`
2598 /// (a backdated append that merged in). The stale marker is present. Rotating
2599 /// (a March append) rolls both prior months out. The marker is NOT proof of a
2600 /// re-roll here: February is absent from the archives, so no completed prior
2601 /// rotation of this batch exists. Pre-fix, the global marker flag drove the
2602 /// archive dedup against the ENTIRE existing archive, suppressed the active
2603 /// `dup` against the unrelated pre-existing archive `dup`, wrote nothing to the
2604 /// archive, and still trimmed the active file — so the entry vanished from
2605 /// disk. The archive must end with BOTH `dup` entries.
2606 #[test]
2607 fn regression_stale_marker_does_not_drop_distinct_same_minute_on_fresh_roll() {
2608 let (_d, store) = temp_store();
2609 let dir = archive_dir(&store);
2610 fs::create_dir_all(&dir).unwrap();
2611
2612 // Pre-existing, independently-authored January archive entry.
2613 let jan = entry(2026, 1, 15, 9, 0, LogKind::Create, Some("dup"), "body");
2614 let mut arch = String::from(LOG_FRONTMATTER);
2615 arch.push('\n');
2616 arch.push_str(&jan.render());
2617 fs::write(archive_path(&store, 2026, 1), arch).unwrap();
2618
2619 // Active file: a February entry plus a SECOND, distinct, byte-identical
2620 // January `dup` (backdated, physically alongside February).
2621 let feb = entry(2026, 2, 5, 9, 0, LogKind::Create, Some("feb"), "feb");
2622 write_raw_log(&store, &[feb, jan.clone()]);
2623
2624 // A stale rotation marker lingers (crash stranded it; not gitignored, so
2625 // it can ride into a clone).
2626 fs::write(rotation_marker_path(&store), b"").unwrap();
2627
2628 // A March append rotates January AND February out as a FRESH roll.
2629 let mar = entry(2026, 3, 1, 0, 0, LogKind::Create, Some("mar"), "mar");
2630 Log::append(&store, &mar).unwrap();
2631
2632 // The genuinely-distinct January `dup` must survive: BOTH copies in the
2633 // archive, the entry never lost.
2634 let jan_arch = fs::read_to_string(archive_path(&store, 2026, 1)).unwrap();
2635 assert_eq!(
2636 count_occurrences(&jan_arch, "## [2026-01-15 09:00] create | dup"),
2637 2,
2638 "stale marker dropped a distinct same-minute January entry; got:\n{jan_arch}"
2639 );
2640 // February rolled to its own (newly created) archive exactly once.
2641 let feb_arch = fs::read_to_string(archive_path(&store, 2026, 2)).unwrap();
2642 assert_eq!(
2643 count_occurrences(&feb_arch, "## [2026-02-05 09:00] create | feb"),
2644 1,
2645 "February did not roll cleanly; got:\n{feb_arch}"
2646 );
2647 // The marker is cleared after the committed rotation.
2648 assert!(
2649 !rotation_marker_path(&store).exists(),
2650 "rotation marker must be cleared after a committed rotation"
2651 );
2652 // The reader agrees: both January `dup`s are visible (no marker now).
2653 let dups = Log::since(&store, ts(2026, 1, 1, 0, 0))
2654 .unwrap()
2655 .into_iter()
2656 .filter(|e| e.object.as_deref() == Some("dup"))
2657 .count();
2658 assert_eq!(dups, 2, "since must report both distinct January dups");
2659 }
2660
2661 /// PRESERVED INVARIANT (write side). A GENUINE interrupted rotation — the
2662 /// whole prior-month roll-out batch is already present in the archives (the
2663 /// archive write committed before the crash), the same entries still sit in
2664 /// the untrimmed active file, and the marker is present — must STILL dedup the
2665 /// re-roll exactly once. This is the case the scoped recovery dedup must keep
2666 /// suppressing; the fix narrows recovery mode to "batch already archived",
2667 /// which this scenario satisfies.
2668 #[test]
2669 fn regression_true_crash_retry_still_dedups_when_whole_batch_already_archived() {
2670 let (_d, store) = temp_store();
2671 let dir = archive_dir(&store);
2672 fs::create_dir_all(&dir).unwrap();
2673
2674 let apr1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr one");
2675 let apr2 = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr two");
2676
2677 // The interrupted rotation already committed both April entries to the
2678 // archive.
2679 let mut arch = String::from(LOG_FRONTMATTER);
2680 arch.push('\n');
2681 arch.push_str(&apr1.render());
2682 arch.push_str(&apr2.render());
2683 fs::write(archive_path(&store, 2026, 4), arch).unwrap();
2684
2685 // The active file still holds the SAME April entries (trim never landed).
2686 write_raw_log(&store, &[apr1.clone(), apr2.clone()]);
2687
2688 // The crash left the in-progress-rotation marker behind.
2689 fs::write(rotation_marker_path(&store), b"").unwrap();
2690
2691 // The agent retries with a May append: April is re-rolled. Because the
2692 // whole April batch is already in the archive, this is a true re-roll and
2693 // must NOT duplicate.
2694 let may = entry(2026, 5, 2, 8, 0, LogKind::Update, Some("may"), "may one");
2695 Log::append(&store, &may).unwrap();
2696
2697 let archived = fs::read_to_string(archive_path(&store, 2026, 4)).unwrap();
2698 assert_eq!(
2699 count_occurrences(&archived, "## [2026-04-10 09:00] ingest | apr-a"),
2700 1,
2701 "true crash-retry duplicated the first April entry; got:\n{archived}"
2702 );
2703 assert_eq!(
2704 count_occurrences(&archived, "## [2026-04-20 09:00] create | apr-b"),
2705 1,
2706 "true crash-retry duplicated the second April entry; got:\n{archived}"
2707 );
2708 }
2709
2710 /// Adversarial review (#7) — two GENUINELY-DISTINCT appends that render
2711 /// byte-identically at minute precision (same minute/kind/object/note) must
2712 /// BOTH survive rotation. The backdated-duplicate case: apr1 rotates in May;
2713 /// the backdated apr2 lands in the active file later and rotates in June as a
2714 /// FRESH roll (no in-progress marker), so it must be appended even though the
2715 /// April archive already holds the byte-identical apr1. Pre-fix the
2716 /// set-membership dedup dropped apr2 — silent, unrecoverable audit-log loss.
2717 #[test]
2718 fn regression_distinct_same_minute_entries_both_survive_rotation() {
2719 let (_d, store) = temp_store();
2720 let apr1 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("x"), "dup");
2721 let apr2 = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("x"), "dup");
2722
2723 Log::append(&store, &apr1).unwrap();
2724 // A May append rotates apr1 into the April archive and COMPLETES (no
2725 // marker left behind).
2726 Log::append(
2727 &store,
2728 &entry(2026, 5, 2, 8, 0, LogKind::Ingest, Some("may"), "m"),
2729 )
2730 .unwrap();
2731 // The backdated apr2 lands in the active file beside the May entry.
2732 Log::append(&store, &apr2).unwrap();
2733 // A June append rotates the May entry AND apr2 out. apr2 is a fresh roll.
2734 Log::append(
2735 &store,
2736 &entry(2026, 6, 1, 8, 0, LogKind::Ingest, Some("jun"), "j"),
2737 )
2738 .unwrap();
2739
2740 let archived = fs::read_to_string(archive_path(&store, 2026, 4)).unwrap();
2741 assert_eq!(
2742 count_occurrences(&archived, "## [2026-04-10 09:00] ingest | x"),
2743 2,
2744 "two distinct same-minute April appends must BOTH survive rotation; got:\n{archived}"
2745 );
2746 // The reader must return both too (read-dedup must not collapse distinct
2747 // same-minute archive entries).
2748 let got = Log::since(&store, ts(2026, 4, 1, 0, 0)).unwrap();
2749 let dups = got
2750 .iter()
2751 .filter(|e| e.object.as_deref() == Some("x"))
2752 .count();
2753 assert_eq!(
2754 dups, 2,
2755 "since must return both distinct same-minute entries; got {got:#?}"
2756 );
2757 }
2758
2759 /// Adversarial review (#12) — `tail`/`since` must return two byte-identical
2760 /// same-minute entries that both live in the ACTIVE log (no archive). Pre-fix
2761 /// a global content-keyed `seen` set suppressed the second on read, so the
2762 /// reader under-reported what was on disk (`grep` saw 2, `tail` saw 1).
2763 #[test]
2764 fn regression_tail_since_return_distinct_same_minute_active_entries() {
2765 let (_d, store) = temp_store();
2766 Log::append(
2767 &store,
2768 &entry(2026, 6, 10, 9, 0, LogKind::Ingest, Some("x"), "dup"),
2769 )
2770 .unwrap();
2771 Log::append(
2772 &store,
2773 &entry(2026, 6, 10, 9, 0, LogKind::Ingest, Some("x"), "dup"),
2774 )
2775 .unwrap();
2776
2777 let tail = Log::tail(&store, 20).unwrap();
2778 assert_eq!(
2779 tail.len(),
2780 2,
2781 "tail must return both same-minute active entries; got {tail:#?}"
2782 );
2783 let since = Log::since(&store, ts(2026, 6, 1, 0, 0)).unwrap();
2784 assert_eq!(
2785 since.len(),
2786 2,
2787 "since must return both same-minute active entries; got {since:#?}"
2788 );
2789 }
2790
2791 // ── regression: read-side active↔archive dedup must be marker-gated ────────
2792
2793 /// THE BUG (HIGH, data-loss). Two GENUINELY-DISTINCT same-(minute,kind,
2794 /// object,note) entries, one in the active `log.md` and one in its month
2795 /// archive, with NO rotation marker present (normal operation). The write
2796 /// side deliberately preserved both on disk (a backdated append after a
2797 /// completed rotation that merely collides on those minute-precision fields
2798 /// is a distinct real event). Pre-fix the read side deduped the
2799 /// active↔archive overlap UNCONDITIONALLY, so `tail`/`since` silently
2800 /// dropped the archive copy — reporting 1 where disk holds 2. The fix gates
2801 /// that dedup on the `.rotating` marker (mirroring the write side); with no
2802 /// marker, both must come back.
2803 #[test]
2804 fn regression_tail_since_keep_distinct_split_entries_without_rotation_marker() {
2805 let (_d, store) = temp_store();
2806
2807 // Author the SAME (minute,kind,object,note) entry once in the May
2808 // archive and once in the active file. This is exactly the on-disk shape
2809 // the repro produces: an entry rotates into log/2026-05.md, then a second
2810 // backdated append of identical fields lands in the active log.md after a
2811 // later-month rotation completed (so NO marker lingers).
2812 let dir = archive_dir(&store);
2813 fs::create_dir_all(&dir).unwrap();
2814 let dup = entry(
2815 2026,
2816 5,
2817 10,
2818 8,
2819 0,
2820 LogKind::Ingest,
2821 Some("records/x.md"),
2822 "same note text",
2823 );
2824 let mut arch = String::from(LOG_FRONTMATTER);
2825 arch.push('\n');
2826 arch.push_str(&dup.render());
2827 fs::write(archive_path(&store, 2026, 5), arch).unwrap();
2828
2829 // Active file: a current-month (June) entry plus the SECOND distinct copy
2830 // of the May-dated event (backdated, so physically alongside June).
2831 let jun = entry(
2832 2026,
2833 6,
2834 1,
2835 9,
2836 0,
2837 LogKind::Create,
2838 Some("records/june.md"),
2839 "june",
2840 );
2841 write_raw_log(&store, &[jun, dup.clone()]);
2842
2843 // No rotation marker => normal operation => trust the disk: BOTH distinct
2844 // copies of the May event must be reported by since and tail.
2845 assert!(
2846 !rotation_marker_path(&store).exists(),
2847 "precondition: no rotation marker (normal operation)"
2848 );
2849
2850 let since = Log::since(&store, ts(2026, 5, 1, 0, 0)).unwrap();
2851 let since_dups = since.iter().filter(|e| **e == dup).count();
2852 assert_eq!(
2853 since_dups, 2,
2854 "since must return BOTH distinct same-minute entries split across \
2855 active+archive when no rotation marker is present; got {since:#?}"
2856 );
2857
2858 let tail = Log::tail(&store, 10).unwrap();
2859 let tail_dups = tail.iter().filter(|e| **e == dup).count();
2860 assert_eq!(
2861 tail_dups, 2,
2862 "tail must return BOTH distinct same-minute entries split across \
2863 active+archive when no rotation marker is present; got {tail:#?}"
2864 );
2865 }
2866
2867 /// PRESERVED INVARIANT. When a rotation IS in flight (the `.rotating` marker
2868 /// is present), an interrupted rotation can leave the SAME physical entry in
2869 /// both the untrimmed active file and its archive. That crash-induced
2870 /// duplicate must still be deduped on read so it is not double-reported —
2871 /// the gating must not throw away the legitimate crash-recovery masking.
2872 #[test]
2873 fn regression_tail_since_dedup_crash_overlap_when_rotation_marker_present() {
2874 let (_d, store) = temp_store();
2875
2876 // Simulate the mid-rotation crash state: the SAME physical May entry is
2877 // in BOTH the archive (write committed) and the active file (trim never
2878 // landed), and the in-flight marker is still on disk.
2879 let dir = archive_dir(&store);
2880 fs::create_dir_all(&dir).unwrap();
2881 let rolled = entry(
2882 2026,
2883 5,
2884 10,
2885 8,
2886 0,
2887 LogKind::Ingest,
2888 Some("records/x.md"),
2889 "same note text",
2890 );
2891 let mut arch = String::from(LOG_FRONTMATTER);
2892 arch.push('\n');
2893 arch.push_str(&rolled.render());
2894 fs::write(archive_path(&store, 2026, 5), arch).unwrap();
2895
2896 // Active file still holds the un-trimmed May entry plus a current-month
2897 // (June) entry — the pre-trim shape a crash leaves behind.
2898 let jun = entry(
2899 2026,
2900 6,
2901 1,
2902 9,
2903 0,
2904 LogKind::Create,
2905 Some("records/june.md"),
2906 "june",
2907 );
2908 write_raw_log(&store, &[jun, rolled.clone()]);
2909
2910 // The crash leaves the in-progress-rotation marker behind.
2911 fs::write(rotation_marker_path(&store), b"").unwrap();
2912 assert!(
2913 rotation_marker_path(&store).exists(),
2914 "precondition: rotation marker present (crash mid-rotation)"
2915 );
2916
2917 // Recovering => the active↔archive overlap is the crash duplicate, masked
2918 // on read: the May entry must be reported ONCE, not twice.
2919 let since = Log::since(&store, ts(2026, 5, 1, 0, 0)).unwrap();
2920 let since_dups = since.iter().filter(|e| **e == rolled).count();
2921 assert_eq!(
2922 since_dups, 1,
2923 "since must dedup the crash-induced active↔archive overlap when the \
2924 rotation marker is present; got {since:#?}"
2925 );
2926
2927 let tail = Log::tail(&store, 10).unwrap();
2928 let tail_dups = tail.iter().filter(|e| **e == rolled).count();
2929 assert_eq!(
2930 tail_dups, 1,
2931 "tail must dedup the crash-induced active↔archive overlap when the \
2932 rotation marker is present; got {tail:#?}"
2933 );
2934 }
2935
2936 /// Adversarial review (#15) — rotation must NOT erase lines before the first
2937 /// VALID entry header. An active log whose entries region opens with a
2938 /// `## [`-shaped line that `parse_header` rejects (a merge orphan / malformed
2939 /// export) before the first real entry: pre-fix `find_first_header` landed on
2940 /// it, `parse_entries` dropped it (no open entry yet), and the rotation
2941 /// re-emitted without it — silently erasing append-only content. The fix
2942 /// folds everything before the first valid header into the preserved header
2943 /// block, which rotation re-emits verbatim.
2944 #[test]
2945 fn regression_rotation_preserves_lines_before_first_valid_header() {
2946 let (_d, store) = temp_store();
2947 let active = active_log_path(&store);
2948 let content = "---\ntype: log\n---\n\n## [orphan from a merge] stray text\n## [2026-04-10 09:00] ingest | x\nbody line\n";
2949 fs::write(&active, content).unwrap();
2950
2951 // A June append rotates the April entry out and rewrites the active file.
2952 Log::append(
2953 &store,
2954 &entry(2026, 6, 1, 8, 0, LogKind::Ingest, Some("jun"), "j"),
2955 )
2956 .unwrap();
2957
2958 let active_after = fs::read_to_string(&active).unwrap();
2959 let arch_after = fs::read_to_string(archive_path(&store, 2026, 4)).unwrap_or_default();
2960 assert!(
2961 active_after.contains("orphan from a merge") || arch_after.contains("orphan from a merge"),
2962 "the pre-first-valid-header line was erased by rotation;\nactive:\n{active_after}\narchive:\n{arch_after}"
2963 );
2964 // Sanity: the real April entry still rotated into its archive.
2965 assert!(
2966 arch_after.contains("## [2026-04-10 09:00] ingest | x"),
2967 "the valid April entry must still rotate to its archive; got:\n{arch_after}"
2968 );
2969 }
2970
2971 // ── regression: reverse reader keeps a `## [` continuation note line (#10) ─
2972
2973 #[test]
2974 fn regression_reverse_reader_preserves_note_line_starting_with_bracket_header() {
2975 // SPEC permits a note of "one or more lines" with no restriction on a
2976 // continuation line starting at column 0 with `## [`. The forward parser
2977 // folds such an unparseable `## [` line into the note; the reverse
2978 // reader (tail/since/last_validate_at) must agree, not split on it.
2979 let (_d, store) = temp_store();
2980 let multi = "First line.\n## [draft outline] more\nThird line.";
2981 let e = entry(
2982 2026,
2983 5,
2984 27,
2985 10,
2986 0,
2987 LogKind::Update,
2988 Some("records/x"),
2989 multi,
2990 );
2991 // Author the log verbatim (render writes the note as-is); this is the
2992 // on-disk shape a hand-written / appended multi-line note produces.
2993 write_raw_log(&store, std::slice::from_ref(&e));
2994
2995 // Pre-fix: header_offsets treated `## [draft outline] more` as a second
2996 // entry boundary, truncating the note to "First line." and dropping the
2997 // carved (non-header) fragment. Post-fix: the full note survives.
2998 let got = Log::tail(&store, 1).unwrap();
2999 assert_eq!(got.len(), 1, "the single entry must be returned");
3000 assert_eq!(
3001 got[0].note, multi,
3002 "reverse reader truncated the note at the `## [` continuation line; \
3003 got {:?}",
3004 got[0].note
3005 );
3006 assert_eq!(got[0], e, "the whole entry must round-trip through tail");
3007
3008 // `since` (the other reverse-reading surface) must agree.
3009 let since = Log::since(&store, ts(2026, 5, 27, 9, 0)).unwrap();
3010 assert_eq!(since, vec![e]);
3011 }
3012
3013 // ── regression: `since` archive pruning uses the UTC month, not local (#11) ─
3014
3015 /// A `DateTime<FixedOffset>` at the given fixed offset (hours east of UTC).
3016 fn ts_offset(
3017 y: i32,
3018 mo: u32,
3019 d: u32,
3020 h: u32,
3021 mi: u32,
3022 offset_hours: i32,
3023 ) -> DateTime<FixedOffset> {
3024 let naive = chrono::NaiveDate::from_ymd_opt(y, mo, d)
3025 .unwrap()
3026 .and_hms_opt(h, mi, 0)
3027 .unwrap();
3028 FixedOffset::east_opt(offset_hours * 3600)
3029 .unwrap()
3030 .from_local_datetime(&naive)
3031 .single()
3032 .unwrap()
3033 }
3034
3035 #[test]
3036 fn regression_since_prunes_archives_on_utc_month_not_local_offset_month() {
3037 // Archive months are bucketed on the UTC calendar. A `since` cutoff with
3038 // a non-UTC offset near a month boundary must not prune an archive whose
3039 // UTC month equals the cutoff's UTC month just because the cutoff's
3040 // LOCAL month is later.
3041 let (_d, store) = temp_store();
3042
3043 // April archive: an entry late on 2026-04-30 at 18:00 UTC.
3044 let apr = entry(
3045 2026,
3046 4,
3047 30,
3048 18,
3049 0,
3050 LogKind::Update,
3051 Some("apr-late"),
3052 "april late",
3053 );
3054 let dir = archive_dir(&store);
3055 fs::create_dir_all(&dir).unwrap();
3056 let mut arch = String::from(LOG_FRONTMATTER);
3057 arch.push('\n');
3058 arch.push_str(&apr.render());
3059 fs::write(archive_path(&store, 2026, 4), arch).unwrap();
3060
3061 // Active file: a clean May entry, so an archive scan is actually needed.
3062 let may = entry(2026, 5, 5, 8, 0, LogKind::Update, Some("may-a"), "may one");
3063 write_raw_log(&store, std::slice::from_ref(&may));
3064
3065 // Cutoff 2026-05-01T00:30:00+07:00 == 2026-04-30T17:30:00Z. The April
3066 // 18:00 UTC entry is strictly newer than this instant.
3067 let cutoff = ts_offset(2026, 5, 1, 0, 30, 7);
3068 // Sanity: the cutoff's UTC month is April, its local month is May.
3069 assert_eq!((cutoff.year(), cutoff.month()), (2026, 5));
3070 assert_eq!(
3071 (
3072 cutoff.with_timezone(&Utc).year(),
3073 cutoff.with_timezone(&Utc).month()
3074 ),
3075 (2026, 4)
3076 );
3077
3078 // Pre-fix: cutoff_ym = (2026, 5) from local fields, so the (2026, 4)
3079 // archive was pruned and the genuinely-newer 18:00 UTC entry was dropped
3080 // — `since` returned only the May entry. Post-fix: cutoff_ym is UTC
3081 // (2026, 4), the April archive is scanned, and both come back.
3082 let got = Log::since(&store, cutoff).unwrap();
3083 let stamps: std::collections::BTreeSet<_> = got.iter().map(|e| e.timestamp).collect();
3084 assert_eq!(
3085 stamps,
3086 [ts(2026, 4, 30, 18, 0), ts(2026, 5, 5, 8, 0)]
3087 .into_iter()
3088 .collect(),
3089 "since(non-UTC cutoff near a month boundary) must include the April \
3090 archive entry newer than the cutoff instant; got {got:?}"
3091 );
3092 }
3093
3094 // ── regression: header-shaped note line corrupts the append-only log (#critical)
3095
3096 #[test]
3097 fn note_line_shaped_like_a_header_is_escaped_and_round_trips() {
3098 // A `contradiction` note quoting an earlier entry header is the
3099 // demonstrated corruption: the verbatim `## [2020-01-01 00:00] delete |
3100 // …` line was parsed as a REAL entry on readback (fabricated entry, real
3101 // note truncated). With write-path escaping it stays note body.
3102 let (_d, store) = temp_store();
3103 let note = "quoting earlier entry:\n## [2020-01-01 00:00] delete | records/contacts/jane.md\nend of quote";
3104 let e = entry(
3105 2026,
3106 6,
3107 11,
3108 4,
3109 41,
3110 LogKind::Contradiction,
3111 Some("records/contacts/jane.md"),
3112 note,
3113 );
3114 Log::append(&store, &e).unwrap();
3115
3116 // On disk: the header-shaped note line must NOT sit at column 0 as a
3117 // `## [` header — `grep "^## \["` must see exactly the one real header.
3118 let raw = fs::read_to_string(store.root.join("log.md")).unwrap();
3119 let header_lines = raw.lines().filter(|l| l.starts_with("## [")).count();
3120 assert_eq!(
3121 header_lines, 1,
3122 "exactly one real entry header may sit at column 0; got:\n{raw}"
3123 );
3124
3125 // Readback returns ONE entry, with the full note intact (no fabricated
3126 // 2020 entry, no truncation).
3127 let got = Log::tail(&store, 10).unwrap();
3128 assert_eq!(got.len(), 1, "exactly one entry; got {got:?}");
3129 assert_eq!(got[0].note, note, "note must round-trip verbatim");
3130 assert_eq!(got[0], e);
3131 let since = Log::since(&store, ts(2026, 1, 1, 0, 0)).unwrap();
3132 assert_eq!(since, vec![e.clone()]);
3133 }
3134
3135 #[test]
3136 fn header_shaped_note_survives_a_later_rotation_uncorrupted() {
3137 // Physical corruption: pre-fix, the fabricated past-dated pseudo-entry
3138 // (year 2020 < current) was rolled into an archive on the NEXT append,
3139 // splitting the real note. With escaping the line is note text, so a
3140 // later append never sees a phantom prior-month entry to roll out.
3141 let (_d, store) = temp_store();
3142 let note = "see\n## [2020-01-01 00:00] delete | records/x.md\nbelow";
3143 let first = entry(
3144 2026,
3145 6,
3146 11,
3147 4,
3148 41,
3149 LogKind::Contradiction,
3150 Some("records/x.md"),
3151 note,
3152 );
3153 Log::append(&store, &first).unwrap();
3154
3155 // Append another current-month entry — the path that re-parses + may
3156 // rotate. No 2020 archive must be created and the first note stays whole.
3157 let second = entry(
3158 2026,
3159 6,
3160 11,
3161 5,
3162 0,
3163 LogKind::Update,
3164 Some("records/y.md"),
3165 "y",
3166 );
3167 Log::append(&store, &second).unwrap();
3168
3169 assert!(
3170 !store.root.join("log").join("2020-01.md").exists(),
3171 "a header-shaped note line must not fabricate a 2020 archive"
3172 );
3173 let got = Log::tail(&store, 10).unwrap();
3174 assert_eq!(got.len(), 2, "two real entries only; got {got:?}");
3175 let first_back = got
3176 .iter()
3177 .find(|e| e.object.as_deref() == Some("records/x.md"));
3178 assert_eq!(
3179 first_back.map(|e| e.note.as_str()),
3180 Some(note),
3181 "the header-shaped note must survive the rotation pass intact"
3182 );
3183 }
3184
3185 #[test]
3186 fn escape_unescape_note_line_round_trips_including_literal_backslash() {
3187 // The escape must be lossless for arbitrary note lines, including a line
3188 // the author genuinely wrote starting with `\` before a header shape.
3189 let valid_header = "## [2020-01-01 00:00] delete | x";
3190 // A real header shape: escaped on write, restored on read.
3191 assert_eq!(
3192 &*escape_note_line(valid_header),
3193 &format!("\\{valid_header}")
3194 );
3195 let escaped = escape_note_line(valid_header).into_owned();
3196 assert_eq!(&*unescape_note_line(&escaped), valid_header);
3197 // An already-`\`-prefixed header-shape line escapes to two backslashes
3198 // and restores to one (never collapses to a bare header).
3199 let pre = format!("\\{valid_header}");
3200 assert_eq!(&*escape_note_line(&pre), &format!("\\{pre}"));
3201 let pre_escaped = escape_note_line(&pre).into_owned();
3202 assert_eq!(&*unescape_note_line(&pre_escaped), &pre);
3203 // Ordinary text (including a `\` that does NOT lead into a header) is
3204 // untouched both ways.
3205 for plain in ["plain note", "## [not a header]", "\\not a header", ""] {
3206 assert_eq!(&*escape_note_line(plain), plain);
3207 assert_eq!(&*unescape_note_line(plain), plain);
3208 }
3209 }
3210
3211 // ── regression: reverse reader scans each block once (no O(file²)) (#perf) ──
3212
3213 #[test]
3214 fn reverse_read_correct_with_header_straddling_a_block_boundary() {
3215 // The incremental per-block header scan must still catch a `## [` marker
3216 // whose `#` falls in one block but whose bytes extend into the already-
3217 // scanned region. Build a log whose total size crosses several blocks and
3218 // verify a full read reconstructs every entry — the straddle case is hit
3219 // by construction across the many block boundaries.
3220 let (_d, store) = temp_store();
3221 let n = 600usize;
3222 let mut expected: Vec<LogEntry> = Vec::new();
3223 for i in 0..n {
3224 let total_min = (i as u32) * 2;
3225 let day = 1 + total_min / (24 * 60);
3226 let hour = (total_min / 60) % 24;
3227 let min = total_min % 60;
3228 // Vary note length so headers land at many offsets relative to the
3229 // fixed 8 KiB block grid, exercising boundary straddles.
3230 let note = format!("note {i} {}", "y".repeat(i % 97));
3231 let e = entry(
3232 2026,
3233 6,
3234 day,
3235 hour,
3236 min,
3237 LogKind::Update,
3238 Some(&format!("records/item-{i:05}")),
3239 ¬e,
3240 );
3241 Log::append(&store, &e).unwrap();
3242 expected.push(e);
3243 }
3244 let size = fs::metadata(store.root.join("log.md")).unwrap().len();
3245 assert!(
3246 size > (REVERSE_BLOCK as u64) * 3,
3247 "test log not large enough ({size} bytes) to cross several blocks"
3248 );
3249 let all = Log::tail(&store, n + 10).unwrap();
3250 assert_eq!(all, expected, "every entry must reconstruct across blocks");
3251 // A small tail must also be exact (the n-newest by timestamp).
3252 assert_eq!(Log::tail(&store, 7).unwrap(), expected[n - 7..].to_vec());
3253 }
3254
3255 #[test]
3256 fn header_offsets_range_finds_boundary_straddling_marker_once() {
3257 // Two headers; `header_offsets` (whole-buffer) finds both. The range
3258 // scan with a window that splits the buffer between them must report the
3259 // one in its window exactly once, consulting the left neighbour for the
3260 // line-start check.
3261 let buf =
3262 b"## [2026-06-01 00:00] update | a\nnote a\n## [2026-06-01 00:01] update | b\nnote b\n";
3263 let full = header_offsets(buf, 0);
3264 assert_eq!(full.len(), 2, "both headers found over the whole buffer");
3265 let second = full[1] as usize;
3266 // A window covering only the SECOND header's `#` reports just it. Its `#`
3267 // is not at index 0, so `base_is_file_start` is irrelevant here.
3268 let only_second = header_offsets_range(buf, 0, second, second + 1, false);
3269 assert_eq!(only_second, vec![full[1]]);
3270 // A window covering only the FIRST reports just it (right content read
3271 // past the window into the buffer). `base == 0` is the true file start,
3272 // so the index-0 candidate is a real line start.
3273 let only_first = header_offsets_range(buf, 0, 0, 1, true);
3274 assert_eq!(only_first, vec![full[0]]);
3275 // Disjoint windows partition the markers with no double-count.
3276 let mut combined = header_offsets_range(buf, 0, 0, second, true);
3277 combined.extend(header_offsets_range(buf, 0, second, buf.len(), false));
3278 assert_eq!(combined, full);
3279 }
3280
3281 /// CRITICAL regression: a MID-LINE `## [<valid header>]` fragment inside a
3282 /// real entry's note that happens to align with a reverse-read block boundary
3283 /// must NOT be fabricated into an entry. The incremental backward scan reads
3284 /// each block's left edge before its left neighbour is buffered; treating
3285 /// buffer index 0 as a line start there would carve a phantom entry from the
3286 /// fragment and truncate the real entry's note. The fix defers the left-edge
3287 /// candidate until its neighbour is read, so the fragment is correctly seen
3288 /// as note body (its `#` is not at a line start).
3289 #[test]
3290 fn reverse_read_does_not_fabricate_entry_from_midline_header_at_block_boundary() {
3291 let (_d, store) = temp_store();
3292
3293 // A single real entry. Its note carries a mid-line `## [` fragment that
3294 // is a *valid* header shape but is NOT at column 0 (so the writer's
3295 // column-0 escape correctly leaves it verbatim — it is the trigger).
3296 let fragment = "see ## [2020-01-01 00:00] delete | records/x.md";
3297 let hash_in_fragment = fragment.find("##").expect("fragment has `##`");
3298
3299 // Build the raw active log by hand so the fragment's `#` lands at the
3300 // FIRST backward block's left edge: the reverse reader anchors its blocks
3301 // at EOF (`new_start = len - REVERSE_BLOCK` on the first block), so the
3302 // `#` must sit exactly `REVERSE_BLOCK` bytes before EOF. We append note
3303 // padding AFTER the fragment to push EOF out to that distance.
3304 //
3305 // Layout (one entry):
3306 // <frontmatter>\n## [<header>] | records/real.md\nlead\n<fragment><tail>\n\n
3307 let header_line = "## [2026-06-14 10:00] update | records/real.md\n";
3308 let mut head = String::from(LOG_FRONTMATTER);
3309 head.push('\n');
3310 head.push_str(header_line);
3311 head.push_str("lead\n");
3312 head.push_str(fragment); // fragment opens the second note line
3313
3314 // Absolute offset of the fragment's `#`.
3315 let hash_off = head.len() - fragment.len() + hash_in_fragment;
3316 // We append `<tail>\n\n`. Bytes after `#` = (head.len() - hash_off) +
3317 // tail_len + 2. Need that == REVERSE_BLOCK so `#` is at `len -
3318 // REVERSE_BLOCK` (the first block's left edge).
3319 let after_hash_in_head = head.len() - hash_off;
3320 let tail_len = REVERSE_BLOCK
3321 .checked_sub(after_hash_in_head + 2)
3322 .expect("REVERSE_BLOCK comfortably exceeds the post-`#` head bytes");
3323 let mut body = head;
3324 body.push_str(&"z".repeat(tail_len)); // valid note bytes on the fragment line
3325 body.push('\n');
3326 body.push('\n');
3327 fs::write(store.root.join("log.md"), &body).unwrap();
3328
3329 // The file must be large enough to cross at least one block boundary.
3330 assert!(
3331 body.len() as u64 > REVERSE_BLOCK as u64,
3332 "test log must span >1 block (len {})",
3333 body.len()
3334 );
3335 // And the fragment's `#` sits exactly at the first block's left edge.
3336 let real_hash_off = body.find("see ##").unwrap() + hash_in_fragment;
3337 assert_eq!(
3338 real_hash_off,
3339 body.len() - REVERSE_BLOCK,
3340 "fragment `#` must land on the first backward block's left edge to exercise the bug"
3341 );
3342
3343 // Reverse read must return EXACTLY ONE entry — the real one — and never a
3344 // fabricated `2020-01-01 delete records/x.md` carved from the fragment.
3345 let got = Log::tail(&store, 10).unwrap();
3346 assert_eq!(
3347 got.len(),
3348 1,
3349 "exactly the one real entry; got {} (a fabricated entry means the boundary `#` was mis-read as a header): {got:#?}",
3350 got.len()
3351 );
3352 let only = &got[0];
3353 assert_eq!(only.object.as_deref(), Some("records/real.md"));
3354 assert_eq!(only.timestamp, ts(2026, 6, 14, 10, 0));
3355 // The note is intact end-to-end (not truncated at the fragment): both the
3356 // lead and the verbatim fragment survive.
3357 assert!(
3358 only.note.contains("lead"),
3359 "note keeps its lead; got {:?}",
3360 only.note
3361 );
3362 assert!(
3363 only.note.contains(fragment),
3364 "note keeps the verbatim mid-line fragment (not truncated); got {:?}",
3365 only.note
3366 );
3367 }
3368
3369 // ── regression: tail/since dedup across active+archive on interrupted rotation
3370
3371 #[test]
3372 fn tail_and_since_dedup_entries_present_in_both_active_and_archive() {
3373 // Reconstructs the finding's crash window: the archive write committed
3374 // but the active rewrite never trimmed, so the same April entries live in
3375 // BOTH the untrimmed active file and `log/2026-04.md`. Readers must
3376 // return each entry ONCE, not twice.
3377 //
3378 // A real crash in that window necessarily leaves the `.rotating` marker
3379 // behind — it is written BEFORE the archive append (Log::append step 1)
3380 // and removed only AFTER the active trim commits, so any state where the
3381 // archive holds the entries but the active was never trimmed has the
3382 // marker present. The read-side overlap dedup is gated on that marker
3383 // (mirroring the write side); without it, an active↔archive collision is
3384 // treated as two genuinely-distinct entries, not a crash duplicate. So
3385 // the test must set the marker to model the crash it claims to.
3386 let (_d, store) = temp_store();
3387 let apr_a = entry(2026, 4, 10, 9, 0, LogKind::Ingest, Some("apr-a"), "apr one");
3388 let apr_b = entry(2026, 4, 20, 9, 0, LogKind::Create, Some("apr-b"), "apr two");
3389
3390 // Active file still holds both April entries (the un-trimmed state).
3391 write_raw_log(&store, &[apr_a.clone(), apr_b.clone()]);
3392 // The committed step-1 archive holds the same two entries.
3393 let dir = archive_dir(&store);
3394 fs::create_dir_all(&dir).unwrap();
3395 let mut arch = String::from(LOG_FRONTMATTER);
3396 arch.push('\n');
3397 arch.push_str(&apr_a.render());
3398 arch.push_str(&apr_b.render());
3399 fs::write(archive_path(&store, 2026, 4), arch).unwrap();
3400 // The crash leaves the in-progress-rotation marker on disk; this is what
3401 // authorizes the read-side overlap dedup.
3402 fs::write(rotation_marker_path(&store), b"").unwrap();
3403
3404 // `since` must return each April entry exactly once.
3405 let since = Log::since(&store, ts(2026, 4, 1, 0, 0)).unwrap();
3406 assert_eq!(
3407 since,
3408 vec![apr_a.clone(), apr_b.clone()],
3409 "since must dedup the doubly-present entries; got {since:?}"
3410 );
3411
3412 // `tail` must too — no duplicate window slots.
3413 let tail = Log::tail(&store, 10).unwrap();
3414 assert_eq!(
3415 tail,
3416 vec![apr_a, apr_b],
3417 "tail must dedup the doubly-present entries; got {tail:?}"
3418 );
3419 }
3420}