syntext 1.3.0

Hybrid code search index for agent workflows
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
use std::path::Path;

use super::*;
use tempfile::TempDir;

#[test]
fn round_trip_empty_segment() {
    let dir = TempDir::new().unwrap();
    let writer = SegmentWriter::new();
    let meta = writer.write_to_dir(dir.path()).unwrap();
    assert_eq!(meta.doc_count, 0);
    assert_eq!(meta.gram_count, 0);
    assert!(dir.path().join(&meta.dict_filename).exists());
    assert!(dir.path().join(&meta.post_filename).exists());

    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path).unwrap();
    assert_eq!(seg.doc_count, 0);
    assert_eq!(seg.gram_count, 0);
}

#[test]
fn round_trip_with_docs_and_grams() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/main.rs"), 0xDEAD, 100);
    writer.add_document(1, Path::new("src/lib.rs"), 0xBEEF, 200);
    writer.add_gram_posting(0xAAAA, 0);
    writer.add_gram_posting(0xAAAA, 1);
    writer.add_gram_posting(0xBBBB, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();
    assert_eq!(meta.doc_count, 2);
    assert_eq!(meta.gram_count, 2);
    assert!(dir.path().join(&meta.dict_filename).exists());
    assert!(dir.path().join(&meta.post_filename).exists());

    let dict_path = dir.path().join(&meta.dict_filename);
    let post_path = dir.path().join(&meta.post_filename);
    let seg = MmapSegment::open_split(&dict_path, &post_path, PostVerify::Full).unwrap();
    assert_eq!(seg.doc_count, 2);

    let d0 = seg.get_doc(0).unwrap();
    assert_eq!(d0.path, Path::new("src/main.rs"));
    assert_eq!(d0.content_hash, 0xDEAD);

    let pl = seg.lookup_gram(0xAAAA).unwrap();
    let ids = pl.to_vec().unwrap();
    assert_eq!(ids, vec![0, 1]);
}

#[test]
fn duplicate_postings_are_deduplicated() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/main.rs"), 0xDEAD, 100);
    writer.add_document(1, Path::new("src/lib.rs"), 0xBEEF, 200);
    writer.add_gram_posting(0xAAAA, 0);
    writer.add_gram_posting(0xAAAA, 0);
    writer.add_gram_posting(0xAAAA, 1);

    let meta = writer.write_to_dir(dir.path()).unwrap();
    assert_eq!(meta.gram_count, 1);
    assert!(dir.path().join(&meta.dict_filename).exists());
    assert!(dir.path().join(&meta.post_filename).exists());

    let dict_path = dir.path().join(&meta.dict_filename);
    let post_path = dir.path().join(&meta.post_filename);
    let seg = MmapSegment::open_split(&dict_path, &post_path, PostVerify::Full).unwrap();
    assert_eq!(seg.gram_cardinality(0xAAAA), Some(2));
}

#[test]
fn corrupt_file_rejected() {
    let dir = TempDir::new().unwrap();
    let bad_path = dir.path().join("bad.dict");
    std::fs::write(&bad_path, b"not a valid segment").unwrap();
    assert!(MmapSegment::open(&bad_path).is_err());
}

#[test]
fn verify_integrity_passes_on_clean_segment() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 1, 10);
    writer.add_gram_posting(0xAA, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path).unwrap();
    assert!(seg.verify_integrity().is_ok());
}

#[test]
fn open_rejects_segment_exceeding_size_limit() {
    // Build a real segment first so we have valid magic/checksum.
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 1, 10);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    // Verify the constant is wired in and a normal-size segment opens fine.
    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path);
    assert!(
        seg.is_ok(),
        "valid segment under size limit must open successfully"
    );
}

#[test]
fn map_private_copy_unaffected_by_post_open_file_mutation() {
    // With MAP_PRIVATE (map_copy_read_only), the mmap is a copy-on-write
    // snapshot of the file at open time. parse_segment_mmap reads every
    // content page during checksum verification, faulting them all into
    // the process private address space. After that point, on-disk mutations
    // are NOT reflected in the mapping.
    //
    // This is the desired security property: an attacker who gains write
    // access to the index directory after open() cannot affect in-process
    // reads. verify_integrity() checks the private copy against itself and
    // always passes; get_doc() returns the original document.
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 1, 10);
    writer.add_gram_posting(0xAA, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path).unwrap();

    // Atomically replace the on-disk file after open() via rename.
    // On Linux, in-place writes (std::fs::write with O_TRUNC) invalidate
    // MAP_PRIVATE + PROT_READ page cache entries and cause SIGBUS.
    // Rename only changes the directory entry; the mmap holds the original
    // inode open via _file and is unaffected.
    let replacement = dir.path().join("replacement.dict");
    std::fs::write(&replacement, b"SNTX_corrupted_on_disk").unwrap();
    std::fs::rename(&replacement, &dict_path).unwrap();

    // The mmap still reads from the original inode; both must succeed.
    assert!(
        seg.verify_integrity().is_ok(),
        "mmap must survive atomic file replacement via rename"
    );
    assert!(
        seg.get_doc(0).is_some(),
        "mmap must serve doc reads after file replacement"
    );
}

#[test]
fn with_capacity_hint_does_not_panic_when_exceeded() {
    let mut writer = SegmentWriter::with_capacity(1, 2);
    writer.add_document(0, Path::new("a.rs"), 1, 10);
    for i in 0u64..100 {
        writer.add_gram_posting(i, 0);
    }
    let dir = TempDir::new().unwrap();
    assert!(writer.write_to_dir(dir.path()).is_ok());
}

#[test]
fn add_document_rejects_duplicate_doc_ids() {
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 1, 10);
    writer.add_document(1, Path::new("b.rs"), 2, 20);
    // Duplicate id=1 should be caught during serialize.
    writer.add_document(1, Path::new("c.rs"), 3, 30);
    let dir = TempDir::new().unwrap();
    let result = writer.write_to_dir(dir.path());
    assert!(result.is_err(), "duplicate doc_id must be rejected");
}

#[test]
fn format_version_constants_are_distinct() {
    assert_ne!(FORMAT_VERSION_V2, FORMAT_VERSION_V3);
    assert_eq!(FORMAT_VERSION, FORMAT_VERSION_V3);
}

#[test]
fn dict_entry_size_matches_components() {
    // 8 (gram_hash) + 8 (abs_off/post_offset) + 4 (count) = 20 bytes.
    assert_eq!(DICT_ENTRY_SIZE, 20);
}

#[test]
fn v3_writer_produces_two_files() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/lib.rs"), 0xABCD, 100);
    writer.add_gram_posting(0x1234, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    assert!(
        dir.path().join(&meta.dict_filename).exists(),
        "missing .dict"
    );
    assert!(
        dir.path().join(&meta.post_filename).exists(),
        "missing .post"
    );
    // No .seg file for v3
    let any_seg = std::fs::read_dir(dir.path())
        .unwrap()
        .any(|e| e.unwrap().file_name().to_string_lossy().ends_with(".seg"));
    assert!(!any_seg, "v3 writer must not produce a .seg file");
}

#[test]
fn v3_round_trip_lookup_gram() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/main.rs"), 0xDEAD, 100);
    writer.add_document(1, Path::new("src/lib.rs"), 0xBEEF, 200);
    writer.add_gram_posting(0xAAAA, 0);
    writer.add_gram_posting(0xAAAA, 1);
    writer.add_gram_posting(0xBBBB, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let seg = MmapSegment::open_split(
        &dir.path().join(&meta.dict_filename),
        &dir.path().join(&meta.post_filename),
        PostVerify::Full,
    )
    .unwrap();

    assert_eq!(seg.doc_count, 2);
    let d0 = seg.get_doc(0).unwrap();
    assert_eq!(d0.path, Path::new("src/main.rs"));

    let pl = seg.lookup_gram(0xAAAA).unwrap();
    assert_eq!(pl.to_vec().unwrap(), vec![0, 1]);

    let pl2 = seg.lookup_gram(0xBBBB).unwrap();
    assert_eq!(pl2.to_vec().unwrap(), vec![0]);

    assert!(seg.lookup_gram(0xCCCC).is_none());
}

#[test]
fn v3_round_trip_get_doc() {
    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 0xAA, 10);
    writer.add_gram_posting(0x11, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let seg = MmapSegment::open_split(
        &dir.path().join(&meta.dict_filename),
        &dir.path().join(&meta.post_filename),
        PostVerify::Full,
    )
    .unwrap();

    let doc = seg.get_doc(0).unwrap();
    assert_eq!(doc.path, Path::new("a.rs"));
    assert_eq!(doc.content_hash, 0xAA);
    assert!(seg.get_doc(1).is_none());
}

#[cfg(unix)]
#[test]
fn round_trip_preserves_non_utf8_path_bytes() {
    use std::ffi::OsString;
    use std::os::unix::ffi::OsStringExt;

    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    let path = std::path::PathBuf::from(OsString::from_vec(b"src/odd\xff.rs".to_vec()));
    writer.add_document(0, &path, 0xDEAD, 100);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path).unwrap();
    let d0 = seg.get_doc(0).unwrap();
    assert_eq!(d0.path, path);
}

#[test]
fn v3_post_file_has_magic_and_checksum() {
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/a.rs"), 0x1234, 100);
    writer.add_gram_posting(0xAAAA, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let post_bytes = std::fs::read(dir.path().join(&meta.post_filename)).unwrap();
    // First 8 bytes must be the magic
    assert_eq!(&post_bytes[..8], b"SNTXPOST", "missing .post magic header");
    // File must be long enough for magic (8) + at least one posting entry + checksum (8)
    assert!(post_bytes.len() >= 17, "post file too short");
    // Last 8 bytes are xxhash64 checksum of the postings data (bytes 8..len-8)
    let postings_data = &post_bytes[8..post_bytes.len() - 8];
    let expected_checksum = xxhash_rust::xxh64::xxh64(postings_data, 0);
    let stored_checksum =
        u64::from_le_bytes(post_bytes[post_bytes.len() - 8..].try_into().unwrap());
    assert_eq!(stored_checksum, expected_checksum, "checksum mismatch");
}

#[test]
fn open_split_rejects_corrupt_post_file() {
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/a.rs"), 0xABCD, 100);
    writer.add_gram_posting(0x1111, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    // Corrupt the .post file by writing wrong magic
    let post_path = dir.path().join(&meta.post_filename);
    let mut post_bytes = std::fs::read(&post_path).unwrap();
    post_bytes[0] = b'X'; // corrupt magic byte
    std::fs::write(&post_path, &post_bytes).unwrap();

    // The magic check is structural: both verification modes must reject it.
    for verify in [PostVerify::Structural, PostVerify::Full] {
        let result = MmapSegment::open_split(
            &dir.path().join(&meta.dict_filename),
            &dir.path().join(&meta.post_filename),
            verify,
        );
        assert!(
            result.is_err(),
            "open_split must reject corrupt .post magic ({verify:?})"
        );
    }
}

#[test]
fn structural_open_tolerates_corrupt_postings_byte_without_panic() {
    // Flip one byte in the middle of the postings data. Structural mode must
    // open the segment, and gram lookups must degrade gracefully (missing
    // candidates or None), never panic. Full mode must reject the file, and
    // verify_postings must detect the corruption post-open.
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/a.rs"), 0xABCD, 100);
    writer.add_gram_posting(0x1111, 0);
    writer.add_gram_posting(0x2222, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let post_path = dir.path().join(&meta.post_filename);
    let mut post_bytes = std::fs::read(&post_path).unwrap();
    let mid = post_bytes.len() / 2;
    post_bytes[mid] ^= 0xFF;
    std::fs::write(&post_path, &post_bytes).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);

    assert!(
        MmapSegment::open_split(&dict_path, &post_path, PostVerify::Full).is_err(),
        "Full verification must reject a flipped postings byte"
    );

    let seg = MmapSegment::open_split(&dict_path, &post_path, PostVerify::Structural).unwrap();
    // Bounds-checked parsing: lookups may return None or a wrong-but-safe
    // posting list; the call itself must not panic.
    let _ = seg.lookup_gram(0x1111);
    let _ = seg.lookup_gram(0x2222);
    assert!(
        seg.verify_postings().is_err(),
        "verify_postings must detect the flipped byte"
    );
}

#[test]
fn verify_postings_passes_on_clean_segment() {
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/a.rs"), 0x1234, 100);
    writer.add_gram_posting(0xAAAA, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let seg = MmapSegment::open_split(
        &dir.path().join(&meta.dict_filename),
        &dir.path().join(&meta.post_filename),
        PostVerify::Structural,
    )
    .unwrap();
    assert!(seg.verify_postings().is_ok());
}

#[test]
fn segment_meta_records_post_len() {
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("src/a.rs"), 0x1234, 100);
    writer.add_gram_posting(0xAAAA, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let on_disk = std::fs::metadata(dir.path().join(&meta.post_filename))
        .unwrap()
        .len();
    assert_eq!(meta.post_len, on_disk);
}

#[test]
fn get_doc_rejects_abs_off_pointing_into_dict_section() {
    // Security regression test for Fix 1a: craft a segment whose doc table
    // index pointer points into the dictionary section. Without the abs_off
    // range check, get_doc would interpret dict bytes as DocEntry fields
    // (information disclosure). With the check, it must return None.
    use xxhash_rust::xxh64::xxh64;

    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 0xABCD, 10);
    writer.add_gram_posting(0x1111, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let mut bytes = std::fs::read(&dict_path).unwrap();
    let len = bytes.len();
    // Footer layout (from end): [doc_table_off(8)][reserved(8)][dict_off(8)]
    //                           [doc_cnt(4)][gram_cnt(4)][cksum(8)][ver(4)][magic(4)]
    let footer_start = len - FOOTER_SIZE;
    let dict_offset_value = u64::from_le_bytes(
        bytes[footer_start + 16..footer_start + 24]
            .try_into()
            .unwrap(),
    ) as usize;

    // Overwrite doc 0's abs_off pointer to point 4 bytes into the dict section.
    let doc_table_offset = HEADER_SIZE; // always 40 for V3
    let bad_abs_off = (dict_offset_value + 4) as u64;
    bytes[doc_table_offset..doc_table_offset + 8].copy_from_slice(&bad_abs_off.to_le_bytes());

    // Recompute the checksum over the content (everything before the footer).
    let new_cksum = xxh64(&bytes[..footer_start], 0);
    bytes[footer_start + 32..footer_start + 40].copy_from_slice(&new_cksum.to_le_bytes());

    let crafted_path = dir.path().join("crafted.dict");
    std::fs::write(&crafted_path, &bytes).unwrap();

    let seg = MmapSegment::open(&crafted_path).unwrap();
    assert!(
        seg.get_doc(0).is_none(),
        "get_doc must return None when abs_off points into dict section"
    );
}

#[test]
fn mmap_isolation_from_disk_overwrite() {
    // After opening, atomically replacing the file on disk must not corrupt
    // the in-memory mapping. MmapSegment retains an open file handle to the
    // original inode; after an atomic rename the old inode stays alive and
    // the mapping continues to read from it.
    //
    // We use rename rather than in-place write: on Linux, MAP_PRIVATE +
    // PROT_READ pages are still backed by the file's page cache, so
    // truncating the file via std::fs::write invalidates those pages and
    // delivers SIGBUS.  Atomic rename only changes the directory entry;
    // the mmap's inode reference is unaffected.
    let dir = tempfile::TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, std::path::Path::new("a.rs"), 0xABCD, 10);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let seg = MmapSegment::open(&dict_path).unwrap();

    // Atomically replace the file at the same path with corrupted content.
    // The segment's inode (held open via _file) is unlinked from the
    // directory but remains alive.
    let replacement = dir.path().join("replacement.dict");
    std::fs::write(&replacement, b"CORRUPTED").unwrap();
    std::fs::rename(&replacement, &dict_path).unwrap();

    // The mmap still references the original inode; the document must be readable.
    let doc = seg.get_doc(0);
    assert!(
        doc.is_some(),
        "mmap must survive atomic file replacement via rename"
    );
}

#[test]
fn v2_posting_offset_below_postings_start_returns_none() {
    // B03 regression: read_posting_list_mmap must reject abs_off values in
    // [HEADER_SIZE, postings_start) — the old check was abs_off < HEADER_SIZE,
    // which accepted values pointing into the doc table section. The fix
    // uses abs_off < postings_start (= doc_table_offset + doc_count * 8).
    //
    // Strategy: write a V3 segment, open the .dict file via open() (V2Mmap
    // path), craft a dict entry whose abs_off points into the doc table
    // region [doc_table_offset, postings_start), recompute the checksum,
    // and verify lookup_gram returns None.
    use xxhash_rust::xxh64::xxh64;

    let dir = TempDir::new().unwrap();
    let mut writer = SegmentWriter::new();
    writer.add_document(0, Path::new("a.rs"), 0xABCD, 10);
    writer.add_gram_posting(0x1111_2222_3333_4444u64, 0);
    let meta = writer.write_to_dir(dir.path()).unwrap();

    let dict_path = dir.path().join(&meta.dict_filename);
    let mut bytes = std::fs::read(&dict_path).unwrap();
    let len = bytes.len();
    let footer_start = len - FOOTER_SIZE;

    // Read layout fields from footer.
    let doc_table_offset =
        u64::from_le_bytes(bytes[footer_start..footer_start + 8].try_into().unwrap()) as usize;
    let doc_count = u32::from_le_bytes(
        bytes[footer_start + 24..footer_start + 28]
            .try_into()
            .unwrap(),
    );
    let dict_offset = u64::from_le_bytes(
        bytes[footer_start + 16..footer_start + 24]
            .try_into()
            .unwrap(),
    ) as usize;
    let postings_start = doc_table_offset + doc_count as usize * 8;

    assert!(
        postings_start > HEADER_SIZE,
        "postings_start({postings_start}) must exceed HEADER_SIZE({HEADER_SIZE}) \
            for this test to distinguish old vs new check"
    );

    // Overwrite the first gram entry's abs_off to doc_table_offset, which
    // falls in [HEADER_SIZE, postings_start). Old check accepted this; new
    // check must reject it. Dict entry: gram_hash(8) + abs_off(8) + count(4).
    let abs_off_field_start = dict_offset + 8; // skip gram_hash bytes
    let crafted_abs_off = doc_table_offset as u64;
    bytes[abs_off_field_start..abs_off_field_start + 8]
        .copy_from_slice(&crafted_abs_off.to_le_bytes());

    // Recompute checksum over content (everything before footer).
    let new_cksum = xxh64(&bytes[..footer_start], 0);
    bytes[footer_start + 32..footer_start + 40].copy_from_slice(&new_cksum.to_le_bytes());

    let crafted_path = dir.path().join("crafted_b03.dict");
    std::fs::write(&crafted_path, &bytes).unwrap();

    // open() creates V2Mmap backing; read_posting_list_mmap is called by
    // lookup_gram. abs_off(doc_table_offset) < postings_start → must be None.
    let seg = MmapSegment::open(&crafted_path).unwrap();
    let result = seg.lookup_gram(0x1111_2222_3333_4444u64);
    assert!(
        result.is_none(),
        "lookup_gram must return None when abs_off({crafted_abs_off}) < \
            postings_start({postings_start}): {result:?}"
    );
}