symforge 7.24.0

SymForge — in-memory code intelligence MCP server for Claude Code, Codex, and Gemini CLI
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
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
use std::path::{Path, PathBuf};

use anyhow::Result;

use crate::domain::{FileClassification, LanguageId};

/// A file found during directory traversal that has a recognized language extension.
#[derive(Debug, Clone)]
pub struct DiscoveredFile {
    /// Relative path from the root, using forward slashes (e.g., "src/lib.rs").
    pub relative_path: String,
    /// Absolute path on disk.
    pub absolute_path: PathBuf,
    /// Language inferred from the file extension.
    pub language: LanguageId,
    /// Deterministic semantic-lane classification captured at discovery time.
    pub classification: FileClassification,
}

/// A file found during a full-filesystem walk (all files, not just known-language ones).
///
/// Used by the admission gate to classify every file — including those with unknown or
/// denylisted extensions — before deciding whether to parse them.
#[derive(Debug, Clone)]
pub struct DiscoveredEntry {
    /// Relative path from the root, using forward slashes.
    pub relative_path: String,
    /// Absolute path on disk.
    pub absolute_path: PathBuf,
    /// File size in bytes from the walk metadata (no extra stat syscall).
    pub file_size: u64,
    /// Language inferred from the extension, if recognized.
    pub language: Option<LanguageId>,
    /// Semantic-lane classification (test/vendor/generated/config flags).
    pub classification: FileClassification,
}

/// Environment variable overriding the maximum number of files a single
/// discovery pass will accept before refusing to index the tree.
const MAX_INDEX_FILES_ENV: &str = "SYMFORGE_MAX_INDEX_FILES";
/// Environment variable overriding the maximum cumulative byte size a single
/// discovery pass will accept before refusing to index the tree.
const MAX_INDEX_BYTES_ENV: &str = "SYMFORGE_MAX_INDEX_BYTES";

/// Default file-count ceiling. Generous enough for very large real monorepos
/// (this repo is ~230 files; 50k+ file monorepos are common), while still well
/// below the point where building the in-memory index maps/strings would
/// exhaust memory or trip `String join would overflow memory bounds`.
const DEFAULT_MAX_INDEX_FILES: u64 = 200_000;
/// Default cumulative-bytes ceiling: 16 GiB of accepted file content. A tree
/// whose discoverable files exceed this is almost certainly a generated-file
/// bomb, a mounted volume, or an accidental scratch root, not a project.
const DEFAULT_MAX_INDEX_BYTES: u64 = 16 * 1024 * 1024 * 1024;

/// Resource ceilings applied DURING the filesystem walk, before any in-memory
/// index build commits to the discovered set. Bounding the streaming walk (not
/// the post-collection `Vec`) is what keeps a huge but non-sensitive tree from
/// OOM-ing or panicking the reload: we stop and return a graceful error the
/// moment either ceiling is crossed, instead of collecting megabytes of paths
/// and then letting `LiveIndex::load` blow the memory bound.
#[derive(Debug, Clone, Copy)]
pub struct DiscoveryLimits {
    /// Maximum number of files accepted before refusing the tree.
    pub max_files: u64,
    /// Maximum cumulative bytes of accepted files before refusing the tree.
    pub max_bytes: u64,
}

impl DiscoveryLimits {
    /// Resolve limits from the environment, falling back to the generous
    /// defaults. A non-parseable or empty override is ignored (the default is
    /// used) so a typo can never silently *lower* the ceiling to zero and brick
    /// indexing — only an explicit, well-formed value takes effect.
    pub fn from_env() -> Self {
        let max_files = parse_positive_env(MAX_INDEX_FILES_ENV).unwrap_or(DEFAULT_MAX_INDEX_FILES);
        let max_bytes = parse_positive_env(MAX_INDEX_BYTES_ENV).unwrap_or(DEFAULT_MAX_INDEX_BYTES);
        Self {
            max_files,
            max_bytes,
        }
    }
}

impl Default for DiscoveryLimits {
    fn default() -> Self {
        Self {
            max_files: DEFAULT_MAX_INDEX_FILES,
            max_bytes: DEFAULT_MAX_INDEX_BYTES,
        }
    }
}

/// Parse a strictly-positive `u64` from the named env var, or `None` if the var
/// is unset, empty, non-numeric, or zero. Zero is rejected so an override can
/// never disable indexing entirely; callers fall back to the default instead.
fn parse_positive_env(name: &str) -> Option<u64> {
    std::env::var(name)
        .ok()?
        .trim()
        .parse::<u64>()
        .ok()
        .filter(|&value| value > 0)
}

/// Environment variable Cargo honors to relocate the build directory. When set
/// to a path whose final component is a direct child of the repo root, that
/// child is a build dir and must be skipped regardless of `.gitignore`.
const CARGO_TARGET_DIR_ENV: &str = "CARGO_TARGET_DIR";

/// Returns `true` when `name` is a Cargo build-directory name: exactly `target`
/// or `target-<suffix>` where `<suffix>` is one or more ASCII alphanumerics or
/// underscores (e.g. `target`, `target-wsl`, `target-x86_64`). This matches the
/// regex `^target(-[A-Za-z0-9_]+)?$` without a per-call regex compile.
///
/// Used to hard-skip build dirs at the REPO ROOT independently of each user's
/// `.gitignore`. `/target` is conventionally gitignored, but a `CARGO_TARGET_DIR`
/// variant like `target-wsl` (common on dual Windows/WSL machines) usually is
/// not, so it would otherwise be indexed as source.
fn is_cargo_build_dir_name(name: &str) -> bool {
    let Some(rest) = name.strip_prefix("target") else {
        return false;
    };
    match rest.strip_prefix('-') {
        // Bare `target`.
        None if rest.is_empty() => true,
        // `target` followed by something other than `-<suffix>` (e.g. `targets`,
        // `target_dir`) is NOT a build dir.
        None => false,
        // `target-<suffix>`: suffix must be non-empty and [A-Za-z0-9_]+.
        Some(suffix) => {
            !suffix.is_empty()
                && suffix
                    .bytes()
                    .all(|b| b.is_ascii_alphanumeric() || b == b'_')
        }
    }
}

/// The repo-root child directory name designated by `CARGO_TARGET_DIR`, if that
/// env var is set AND resolves to a direct child of `root`. Returns `None` when
/// the var is unset, empty, or points somewhere that is not a single-segment
/// child of the root (an absolute path outside the tree never produces
/// discoverable entries, so it needs no skip entry here).
///
/// Comparison is done on canonicalized paths so a relative or symlinked
/// `CARGO_TARGET_DIR` still matches the root child the walk actually traverses.
fn cargo_target_dir_root_child(root: &Path) -> Option<String> {
    let raw = std::env::var_os(CARGO_TARGET_DIR_ENV)?;
    if raw.is_empty() {
        return None;
    }
    let target = PathBuf::from(&raw);
    // Resolve against the root for relative values (Cargo interprets a relative
    // CARGO_TARGET_DIR against the working directory; for discovery we only care
    // about the case where it lands directly under the indexed root).
    let target_abs = if target.is_absolute() {
        target
    } else {
        root.join(&target)
    };
    let canon_target = std::fs::canonicalize(&target_abs).unwrap_or(target_abs);
    let parent = canon_target.parent()?;
    if parent != root {
        return None;
    }
    canon_target
        .file_name()
        .and_then(|n| n.to_str())
        .map(|s| s.to_string())
}

/// Returns `true` when `relative_path` (forward-slash normalized, relative to the
/// repo root) lives under a repo-root-level Cargo build directory and must be
/// skipped. Only the FIRST path component is inspected, so a legitimately-named
/// nested source dir such as `src/target/mod.rs` is never over-skipped — only a
/// `target*` (or `CARGO_TARGET_DIR`) directory that is a direct child of the root.
fn is_under_repo_root_build_dir(relative_path: &str, target_dir_child: Option<&str>) -> bool {
    let Some(first) = relative_path.split('/').next() else {
        return false;
    };
    // A path with no separator is a root-level FILE, not a build dir; only treat
    // it as build output when it is the first segment of a deeper path.
    if first == relative_path {
        return false;
    }
    if is_cargo_build_dir_name(first) {
        return true;
    }
    matches!(target_dir_child, Some(child) if child == first)
}

/// SF-025: returns `true` when any component of `relative_path` (forward-slash
/// normalized, relative to the repo root) is a "hidden" dotfile/dotdir — a
/// segment beginning with `.` other than the `.`/`..` traversal segments.
///
/// The bulk discovery walk (`discover_all_files`) uses `ignore::WalkBuilder`
/// with its default `.hidden(true)`, which skips any entry whose name — or an
/// ancestor directory's name — starts with `.` (e.g. `.github/workflows/ci.yml`,
/// `.travis.yml`). The single-file (re)index choke point in the watcher did NOT
/// apply that rule, so a freshen-on-read of a tracked hidden file would parse
/// and INSERT it even though a fresh bulk load never discovered it. That made
/// index membership query-history-dependent: a file was invisible to
/// `search_files` until someone happened to `get_file_context` it, and identical
/// health calls disagreed across processes. This predicate lets the single-file
/// path apply the SAME hidden-path exclusion as the walk, keeping admission
/// symmetric and the index deterministic from scan policy alone.
pub fn path_has_hidden_component(relative_path: &str) -> bool {
    relative_path
        .split('/')
        .any(|component| component.starts_with('.') && component != "." && component != "..")
}

/// Build the graceful, explicit over-cap error. Surfaced to the caller (and
/// thus the MCP client) instead of an OOM/panic, and it names the override knob
/// so an operator with a genuinely huge repo can raise the ceiling.
fn tree_too_large_error(files: u64, bytes: u64, limits: &DiscoveryLimits) -> anyhow::Error {
    anyhow::anyhow!(
        "tree too large to index ({files} files / {bytes} bytes exceeds limit of \
         {max_files} files / {max_bytes} bytes); set {MAX_INDEX_FILES_ENV} or \
         {MAX_INDEX_BYTES_ENV} to override",
        max_files = limits.max_files,
        max_bytes = limits.max_bytes,
    )
}

/// Discover all source files under `root` that have a recognized language extension.
///
/// - Respects `.gitignore` files via the `ignore` crate.
/// - Normalizes path separators to `/` in `relative_path`.
/// - Returns files sorted case-insensitively by `relative_path`.
/// - Refuses trees that exceed [`DiscoveryLimits`] with a graceful error rather
///   than collecting an unbounded set and OOM-ing the in-memory index build.
pub fn discover_files(root: &Path) -> Result<Vec<DiscoveredFile>> {
    use ignore::WalkBuilder;

    // Canonicalize root so that strip_prefix succeeds even when the walker
    // resolves symlinks to their canonical targets.
    let root = std::fs::canonicalize(root).unwrap_or_else(|_| root.to_path_buf());

    // Bound the walk by accepted file count. This pass only tracks files with a
    // recognized language, so byte ceilings are enforced by `discover_all_files`
    // (the full-load entry point that has file sizes from the walk metadata).
    let limits = DiscoveryLimits::from_env();
    // Repo-root build-dir child designated by CARGO_TARGET_DIR, resolved once.
    let target_dir_child = cargo_target_dir_root_child(&root);
    let mut files: Vec<DiscoveredFile> = Vec::new();
    for entry_result in WalkBuilder::new(&root).build() {
        let Ok(entry) = entry_result else { continue };
        let path =
            std::fs::canonicalize(entry.path()).unwrap_or_else(|_| entry.path().to_path_buf());

        // Use the already-known file_type from the walker instead of
        // path.is_file() which would issue a redundant stat() syscall.
        if !entry.file_type().map(|ft| ft.is_file()).unwrap_or(false) {
            continue;
        }

        let Some(ext) = path.extension().and_then(|e| e.to_str()) else {
            continue;
        };
        let Some(language) = LanguageId::from_extension(ext) else {
            continue;
        };

        // Compute relative path from root
        let Ok(relative) = path.strip_prefix(&root) else {
            continue;
        };
        // Normalize backslashes to forward slashes
        let relative_path = relative.to_string_lossy().replace('\\', "/");

        // Repo-independent skip for Cargo build dirs at the REPO ROOT level
        // (`target`, `target-wsl`, `CARGO_TARGET_DIR`, …). `/target` is usually
        // gitignored, but variant build dirs often are not, so do not rely on
        // each user's `.gitignore`. Nested source dirs like `src/target/` are
        // unaffected because only the first path component is inspected.
        if is_under_repo_root_build_dir(&relative_path, target_dir_child.as_deref()) {
            continue;
        }

        // Refuse BEFORE growing the set past the ceiling, so a huge tree returns
        // a graceful error rather than collecting an unbounded path vector.
        if files.len() as u64 >= limits.max_files {
            return Err(tree_too_large_error(files.len() as u64 + 1, 0, &limits));
        }

        files.push(DiscoveredFile {
            classification: FileClassification::for_code_path(&relative_path),
            relative_path,
            absolute_path: path,
            language,
        });
    }

    // Cache sort keys once instead of lowercasing paths on every comparator call.
    files.sort_by_cached_key(|file| file.relative_path.to_lowercase());

    Ok(files)
}

/// Discover ALL files under `root` regardless of extension, for admission-gate classification.
///
/// Unlike `discover_files`, this function:
/// - Yields every file (not just known-language ones), so denylisted/binary files are visible.
/// - Captures file size from walk metadata (avoids a separate stat() call).
/// - Sets `language = None` for files with unrecognized extensions.
/// - Returns files sorted case-insensitively by `relative_path`.
/// - Refuses trees that exceed [`DiscoveryLimits`] (file count AND cumulative
///   bytes) with a graceful error rather than collecting an unbounded set and
///   OOM-ing / panicking the in-memory index build in `LiveIndex::load`.
///
/// This is the discovery entry point used by the full `LiveIndex::load`, so the
/// byte ceiling is enforced here (file sizes are already known from the walk
/// metadata, so no extra `stat` is needed to track cumulative bytes).
pub fn discover_all_files(root: &Path) -> Result<Vec<DiscoveredEntry>> {
    use ignore::WalkBuilder;

    // Canonicalize root so that strip_prefix succeeds even when the walker
    // resolves symlinks to their canonical targets.
    let root = std::fs::canonicalize(root).unwrap_or_else(|_| root.to_path_buf());

    // Bound the streaming walk by accepted file count AND cumulative bytes,
    // refusing the moment either ceiling is crossed — before the unbounded
    // path/byte set is handed to the in-memory index build.
    let limits = DiscoveryLimits::from_env();
    // Repo-root build-dir child designated by CARGO_TARGET_DIR, resolved once.
    let target_dir_child = cargo_target_dir_root_child(&root);
    let mut total_bytes: u64 = 0;
    let mut entries: Vec<DiscoveredEntry> = Vec::new();
    // SF-012(B): the repo-root build-dir heuristic (`is_under_repo_root_build_dir`)
    // matches `target-<alnum>` by design (e.g. `target-wsl`, a `CARGO_TARGET_DIR`
    // variant), but false-positives on legitimately tracked source dirs whose name
    // happens to match — tokio's `target-specs/` (tracked `.md`/`.json`, not
    // gitignored). Build output is NEVER git-tracked, so git-tracked status is a
    // decisive counter-signal: if a path the heuristic would skip is tracked, keep
    // it. The tracked set is computed LAZILY (only on the first build-dir hit) so
    // the common case — no root-level `target-*` dir — pays nothing. `None` means
    // "no git / unreadable index" (fail open: heuristic decides alone, as before).
    let mut tracked_for_build_dirs: Option<Option<std::collections::HashSet<String>>> = None;
    for entry_result in WalkBuilder::new(&root).build() {
        let Ok(entry) = entry_result else { continue };
        let path =
            std::fs::canonicalize(entry.path()).unwrap_or_else(|_| entry.path().to_path_buf());

        if !entry.file_type().map(|ft| ft.is_file()).unwrap_or(false) {
            continue;
        }

        // Get file size from the walk metadata (DirEntry has it on most platforms).
        // Fall back to a stat call only when metadata is unavailable.
        let file_size = entry
            .metadata()
            .ok()
            .map(|m| m.len())
            .unwrap_or_else(|| std::fs::metadata(&path).map(|m| m.len()).unwrap_or(0));

        // Compute relative path from root
        let Ok(relative) = path.strip_prefix(&root) else {
            continue;
        };
        let relative_path = relative.to_string_lossy().replace('\\', "/");

        // Repo-independent skip for Cargo build dirs at the REPO ROOT level
        // (`target`, `target-wsl`, `CARGO_TARGET_DIR`, …), independent of each
        // user's `.gitignore`. Skipping before the size/byte accounting keeps
        // build output from counting against the discovery ceilings. Nested
        // source dirs like `src/target/` are unaffected (first component only).
        if is_under_repo_root_build_dir(&relative_path, target_dir_child.as_deref()) {
            // SF-012(B): rescue genuine source. Only build output reaches the size
            // ceilings, so the heuristic's intent is to drop build artifacts — but
            // a tracked `target-*` source dir (tokio `target-specs/`) is not build
            // output. Consult the git-tracked set (computed once, lazily); a
            // tracked path overrides the heuristic and is admitted normally. When
            // git is unavailable the set is `None` and the heuristic decides alone.
            let tracked = tracked_for_build_dirs
                .get_or_insert_with(|| tracked_path_set_for_build_dir_rescue(&root));
            let rescued = tracked
                .as_ref()
                .is_some_and(|set| set.contains(relative_path.as_str()));
            if !rescued {
                continue;
            }
        }

        // Attempt language detection; None for unknown/denylisted extensions.
        let language = path
            .extension()
            .and_then(|e| e.to_str())
            .and_then(LanguageId::from_extension);

        let classification = FileClassification::for_code_path(&relative_path);

        // Refuse BEFORE pushing past either ceiling. `saturating_add` keeps the
        // byte counter from wrapping on a pathological tree; once it crosses the
        // limit we return the graceful error instead of continuing to allocate.
        let projected_files = entries.len() as u64 + 1;
        let projected_bytes = total_bytes.saturating_add(file_size);
        if projected_files > limits.max_files || projected_bytes > limits.max_bytes {
            return Err(tree_too_large_error(
                projected_files,
                projected_bytes,
                &limits,
            ));
        }
        total_bytes = projected_bytes;

        entries.push(DiscoveredEntry {
            relative_path,
            absolute_path: path,
            file_size,
            language,
            classification,
        });
    }

    // Cache sort keys once instead of lowercasing paths on every comparator call.
    entries.sort_by_cached_key(|entry| entry.relative_path.to_lowercase());

    Ok(entries)
}

/// Load all `.gitignore` patterns from a repository root and nested directories.
///
/// Uses `ignore::gitignore::GitignoreBuilder` to build a composite gitignore matcher.
/// Walks nested `.gitignore` files up to `max_depth` levels (default 6).
/// Returns `None` if no `.gitignore` files are found or if loading fails.
pub fn load_gitignore(root: &Path) -> Option<ignore::gitignore::Gitignore> {
    use ignore::gitignore::GitignoreBuilder;
    use std::collections::VecDeque;

    let root_gitignore = root.join(".gitignore");
    if !root_gitignore.exists() {
        return None;
    }

    let mut builder = GitignoreBuilder::new(root);

    // BFS to find nested .gitignore files (max depth 6)
    let max_depth: usize = 6;
    let mut queue: VecDeque<(PathBuf, usize)> = VecDeque::new();
    queue.push_back((root.to_path_buf(), 0));

    while let Some((dir, depth)) = queue.pop_front() {
        let gitignore_path = dir.join(".gitignore");
        if gitignore_path.is_file()
            && let Some(err) = builder.add(&gitignore_path)
        {
            tracing::debug!("failed to load {:?}: {}", gitignore_path, err);
        }

        if depth < max_depth
            && let Ok(entries) = std::fs::read_dir(&dir)
        {
            for entry in entries.flatten() {
                let path = entry.path();
                if path.is_dir() {
                    // Skip common directories that won't have relevant .gitignore files
                    let name = entry.file_name();
                    let name_str = name.to_string_lossy();
                    if name_str.starts_with('.') && name_str != ".github" {
                        continue;
                    }
                    queue.push_back((path, depth + 1));
                }
            }
        }
    }

    match builder.build() {
        Ok(gi) => {
            // Only return Some if there are actual patterns
            if gi.is_empty() { None } else { Some(gi) }
        }
        Err(e) => {
            tracing::debug!("failed to build gitignore matcher: {}", e);
            None
        }
    }
}

/// Walk upward from the current working directory, looking for a `.git` directory.
/// Returns `None` if no git root is found and the cwd is a forbidden directory.
pub fn find_project_root() -> Option<PathBuf> {
    let cwd = std::env::current_dir().unwrap_or_else(|_| PathBuf::from("."));

    // Try to find a git root first (scoped by repo boundary), BUT run the same
    // sensitive/forbidden guard on the discovered `.git` root that the cwd
    // fallback below uses. A `.git` planted at a sensitive ancestor (e.g.
    // `git init` in `C:\Users\<name>` or a malicious `/etc/.git`) must NOT be
    // selected and indexed unguarded: if the `.git`-bearing ancestor is
    // forbidden we skip it and keep walking up, exactly as the rest of the
    // guard does, so a deeper legitimate `.git` is still found and a genuine
    // project `.git` continues to be selected.
    let mut current = cwd.clone();
    loop {
        if current.join(".git").exists() && !is_forbidden_root(&current) {
            return Some(current);
        }
        match current.parent() {
            Some(parent) => current = parent.to_path_buf(),
            None => break,
        }
    }

    // No git root found — use cwd if it's not a forbidden directory.
    if is_forbidden_root(&cwd) {
        tracing::warn!(
            path = %cwd.display(),
            "refusing to auto-index: directory is too broad (home dir, drive root, or system path)"
        );
        None
    } else {
        Some(cwd)
    }
}

/// Returns `true` if `path` is a directory that should never be auto-indexed
/// because it would be too large or contain unrelated files.
fn is_forbidden_root(path: &Path) -> bool {
    // Canonicalize for reliable comparison (resolves symlinks, normalizes separators).
    let path = path.canonicalize().unwrap_or_else(|_| path.to_path_buf());

    // 0. Unified trust-boundary guard. `paths::is_sensitive_path` is the SINGLE
    //    canonical guard shared with the attacker-facing index tools
    //    (`tools::index_folder`, `daemon::index_folder_for_session`,
    //    `daemon::open_project_session`). Delegating here makes the trusted
    //    launcher AT LEAST as strict as the tool surface, so the two can never
    //    drift apart again — the drift that caused the original daemon bypass.
    //    The launcher-specific rules below (running-user `$HOME`, WSL probe)
    //    remain as additional, narrower checks on top of this shared floor.
    if crate::paths::is_sensitive_path(&path) {
        return true;
    }

    // 1. Drive roots: C:\, D:\, /, etc.
    if path.parent().is_none() {
        return true;
    }

    // 2. Windows drive roots that have a parent but are still just "C:\"
    #[cfg(target_os = "windows")]
    {
        let path_str = path.to_string_lossy();
        if path_str.len() <= 7 && path_str.ends_with('\\') {
            return true;
        }
    }

    // 3. User home directories.
    if let Some(home) = home_dir() {
        let home = home.canonicalize().unwrap_or(home);
        if path == home {
            return true;
        }
    }

    // 4a. System directory names — always forbidden anywhere.
    //     These are unambiguous: a directory literally named `system32`
    //     or `node_modules` is virtually never a project root.
    if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
        let lower = name.to_lowercase();
        const SYSTEM_NAMES: &[&str] = &[
            "windows",
            "system32",
            "program files",
            "program files (x86)",
            "programdata",
            "node_modules",
            ".npm",
            ".cargo",
        ];
        if SYSTEM_NAMES.contains(&lower.as_str()) {
            return true;
        }
    }

    // 4b. Top-level container names — forbidden only when sitting directly
    //     under a filesystem root or drive root. A legitimate project named
    //     `tmp` or `var` deeper in the tree is allowed.
    if let Some(name) = path.file_name().and_then(|n| n.to_str()) {
        let lower = name.to_lowercase();
        const CONTAINER_NAMES: &[&str] = &["users", "home", "tmp", "temp", "var", "appdata"];
        if CONTAINER_NAMES.contains(&lower.as_str())
            && path
                .parent()
                .map(|p| {
                    // Parent is a drive root or filesystem root → forbid.
                    p.parent().is_none() || {
                        #[cfg(target_os = "windows")]
                        {
                            let pstr = p.to_string_lossy();
                            pstr.len() <= 7 && pstr.ends_with('\\')
                        }
                        #[cfg(not(target_os = "windows"))]
                        {
                            false
                        }
                    }
                })
                .unwrap_or(false)
        {
            return true;
        }
    }

    // 4c. WSL DrvFs Windows-profile / drive-root guard (Linux only).
    //     Under WSL, Windows drives mount at /mnt/<drive> (default automount root)
    //     and the Windows user profile surfaces at /mnt/<drive>/Users/<name>. None
    //     of the rules above catch this: $HOME is the Linux home (/home/<user>), so
    //     the home-based guards never match, and the leaf-name guards never inspect
    //     the intermediate `Users` component. Auto-indexing any of these roots walks
    //     a huge tree over the slow DrvFs/9p mount and hangs the daemon.
    //
    //     We forbid the broad container roots only — NOT deep project dirs — so a
    //     non-git project kept at /mnt/c/Users/<name>/dev/proj stays auto-indexable:
    //       /mnt/<drive>                 (bare Windows drive root)
    //       /mnt/<drive>/Users           (the profile container)
    //       /mnt/<drive>/Users/<name>    (a bare profile root)
    //     A genuine git repo anywhere under these is still indexable because the
    //     `.git` fast-path in `find_project_root` returns before this gate runs.
    //
    //     Gated on an actual WSL probe so a real Linux host that merely mounts a
    //     volume at /mnt/<letter>/Users is not falsely forbidden. The `Users`
    //     segment is matched case-insensitively because DrvFs is case-insensitive
    //     but path canonicalization is case-preserving — `cd /mnt/c/users/...`
    //     reaches the identical Windows tree and must be caught too.
    #[cfg(not(target_os = "windows"))]
    {
        if is_running_under_wsl() && is_wsl_windows_container_path(&path) {
            return true;
        }
    }

    // 5. Parent-of-home: e.g. C:\Users or /home
    if let Some(home) = home_dir() {
        let home = home.canonicalize().unwrap_or(home);
        if let Some(parent) = home.parent() {
            let parent = parent
                .canonicalize()
                .unwrap_or_else(|_| parent.to_path_buf());
            if path == parent {
                return true;
            }
        }
    }

    false
}

/// Cross-platform home directory lookup.
fn home_dir() -> Option<PathBuf> {
    // std::env::home_dir is deprecated but dirs::home_dir may not be available.
    // Use environment variables directly for reliability.
    #[cfg(target_os = "windows")]
    {
        std::env::var("USERPROFILE").ok().map(PathBuf::from)
    }
    #[cfg(not(target_os = "windows"))]
    {
        std::env::var("HOME").ok().map(PathBuf::from)
    }
}

/// Returns `true` when running inside the Windows Subsystem for Linux.
///
/// Detected by sniffing `/proc/version` for the `microsoft` / `WSL` marker the
/// WSL kernel writes there. The result is computed once and cached, so the file
/// is read at most one time per process. Always `false` on non-Linux targets.
#[cfg(not(target_os = "windows"))]
fn is_running_under_wsl() -> bool {
    use std::sync::OnceLock;
    static IS_WSL: OnceLock<bool> = OnceLock::new();
    *IS_WSL.get_or_init(|| {
        std::fs::read_to_string("/proc/version")
            .map(|v| {
                let v = v.to_ascii_lowercase();
                v.contains("microsoft") || v.contains("wsl")
            })
            .unwrap_or(false)
    })
}

/// Pure path-shape test for the broad WSL DrvFs container roots that must never
/// be auto-indexed: the bare Windows drive mount and the Windows user-profile
/// container/root surfaced under WSL's default `/mnt/` automount.
///
/// Returns `true` for exactly:
///
/// - `/mnt/<drive>` (bare drive root)
/// - `/mnt/<drive>/Users` (profile container)
/// - `/mnt/<drive>/Users/<name>` (bare profile root)
///
/// where `<drive>` is a single ASCII letter and `Users` matches case-insensitively
/// (DrvFs is case-insensitive but canonicalization is case-preserving). Anything
/// deeper (`/mnt/<drive>/Users/<name>/...`) and any non-`Users` mount path
/// (`/mnt/<drive>/code/proj`) returns `false` and stays indexable.
///
/// Path-shape only — the caller is responsible for confirming the host is WSL.
/// Kept separate from the WSL probe so it is host-independent and unit-testable.
#[cfg(not(target_os = "windows"))]
fn is_wsl_windows_container_path(path: &Path) -> bool {
    // Lexically normalize: drop `.`, pop on `..`. A path that escapes above the
    // root via `..` is treated as non-matching rather than silently collapsing,
    // so the gate never misfires on `..`-bearing input a future caller passes.
    let mut comps: Vec<&str> = Vec::new();
    for component in path.components() {
        match component {
            std::path::Component::Normal(os) => {
                if let Some(s) = os.to_str() {
                    comps.push(s);
                }
            }
            // `..` that pops past the root means the path escapes above `/mnt`;
            // treat as non-matching rather than silently collapsing.
            std::path::Component::ParentDir if comps.pop().is_none() => return false,
            std::path::Component::ParentDir => {}
            // RootDir / CurDir / Prefix carry no addressable segment.
            _ => {}
        }
    }

    if comps.first() != Some(&"mnt") {
        return false;
    }

    let is_drive_letter =
        |s: &str| s.len() == 1 && s.chars().next().is_some_and(|c| c.is_ascii_alphabetic());

    let is_users = |s: &str| s.eq_ignore_ascii_case("Users");

    match comps.as_slice() {
        // /mnt/<drive> — bare Windows drive root.
        [_mnt, drive] => is_drive_letter(drive),
        // /mnt/<drive>/Users — the profile container.
        [_mnt, drive, users] => is_drive_letter(drive) && is_users(users),
        // /mnt/<drive>/Users/<name> — a bare profile root (exactly 4 segments).
        [_mnt, drive, users, _name] => is_drive_letter(drive) && is_users(users),
        // Bare /mnt, or deeper than a bare profile root
        // (/mnt/<drive>/Users/<name>/...), stays indexable.
        _ => false,
    }
}

/// Check if content appears to be binary.
/// Examines up to BINARY_SNIFF_BYTES of the content using three heuristics:
/// 1. NUL byte present -> binary
/// 2. UTF-8 decode failure -> binary
/// 3. >30% suspicious control bytes (excluding \t, \n, \r) -> binary
pub fn is_binary_content(content: &[u8]) -> bool {
    if content.is_empty() {
        return false;
    }
    let check_len = content.len().min(crate::domain::index::BINARY_SNIFF_BYTES);
    let window = &content[..check_len];

    // Heuristic 1: NUL byte
    if window.contains(&0) {
        return true;
    }

    // Heuristic 2: Invalid UTF-8
    if std::str::from_utf8(window).is_err() {
        return true;
    }

    // Heuristic 3: High control byte ratio
    // Control bytes: 0x01-0x08, 0x0E-0x1F, 0x7F
    // Excludes common text controls: \t (0x09), \n (0x0A), \r (0x0D)
    let suspicious_controls = window
        .iter()
        .filter(|&&b| matches!(b, 0x01..=0x08 | 0x0E..=0x1F | 0x7F))
        .count();
    let ratio = suspicious_controls as f64 / window.len() as f64;
    if ratio > 0.30 {
        return true;
    }

    false
}

use crate::domain::index::{
    AdmissionDecision, AdmissionTier, HARD_SKIP_BYTES, METADATA_ONLY_BYTES, SkipReason,
};

/// Classify a file's admission tier. Returns AdmissionDecision with both tier and reason.
///
/// Precedence (first match wins):
/// 1. Hard-skip size ceiling (>100MB) → Tier 3
/// 2. Dependency lockfile (exact basename) → Tier 2
/// 3. Extension denylist → Tier 2
/// 4. Metadata-only size threshold (>1MB) → Tier 2
/// 5. Binary sniff (null bytes in first 8KB) → Tier 2
/// 6. All else → Tier 1
pub fn classify_admission(
    path: &std::path::Path,
    file_size: u64,
    content_sample: Option<&[u8]>,
) -> AdmissionDecision {
    use crate::domain::index::{is_denylisted_extension, is_dependency_lockfile};

    if file_size > HARD_SKIP_BYTES {
        return AdmissionDecision::skip(AdmissionTier::HardSkip, SkipReason::SizeCeiling);
    }
    // Dependency lockfiles are machine-generated manifests: their resolved
    // dependency trees parse into thousands of meaningless key/value symbols that
    // pollute symbol counts and `conventions` complexity stats. Demote to Tier-2
    // metadata-only (path stays searchable; no symbol extraction). Checked before
    // the size threshold so a >1MB lockfile still reports `lockfile`, not `>1MB`.
    if is_dependency_lockfile(path) {
        return AdmissionDecision::skip(
            AdmissionTier::MetadataOnly,
            SkipReason::DependencyLockfile,
        );
    }
    if let Some(ext) = path.extension().and_then(|e| e.to_str())
        && is_denylisted_extension(ext)
    {
        return AdmissionDecision::skip(
            AdmissionTier::MetadataOnly,
            SkipReason::DenylistedExtension,
        );
    }
    if file_size > METADATA_ONLY_BYTES {
        return AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::SizeThreshold);
    }
    if let Some(content) = content_sample
        && is_binary_content(content)
    {
        return AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::BinaryContent);
    }
    AdmissionDecision::normal()
}

/// SF-004 / SF-012: reconcile a `classify_admission` result for a file whose
/// extension maps to no supported tree-sitter grammar.
///
/// `classify_admission` only inspects size / denylist / binary content — it has
/// no concept of language recognition, so a small, non-binary, non-denylisted
/// file with an unknown extension (`.tcl`, `.sh`, `.m`, `.eex`, extensionless
/// `LICENSE`/`Makefile`, …) comes back `AdmissionTier::Normal`. But the parser
/// cannot extract symbols from it, so storing a `Normal` decision is
/// self-contradictory: such records were silently dropped by `tier_counts`
/// (the `Normal => {}` arm) and minted a false "File not found" in
/// `get_file_context`.
///
/// This helper is the single place that maps that "Normal but unparseable"
/// state onto an honest `Tier-2 metadata-only / UnsupportedLanguage` decision.
/// A non-`Normal` decision (real size/denylist/binary skip) is returned
/// unchanged, so this never overrides a more specific reason. Callers invoke it
/// ONLY on the no-recognized-language branch, so a `Normal` input here always
/// means "unparseable language", never "Tier-1 source".
pub fn unsupported_language_decision(decision: AdmissionDecision) -> AdmissionDecision {
    if decision.tier == AdmissionTier::Normal {
        AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::UnsupportedLanguage)
    } else {
        decision
    }
}

/// Env var gating the SF-009 opt-in "exclude untracked" admission policy.
/// Default OFF — when unset (or set to anything other than a truthy value) the
/// index admits files exactly as before, so admission defaults are unchanged.
pub const EXCLUDE_UNTRACKED_ENV: &str = "SYMFORGE_EXCLUDE_UNTRACKED";

/// Returns `true` when the opt-in `SYMFORGE_EXCLUDE_UNTRACKED` policy is enabled.
///
/// Accepts the usual truthy spellings (`1`, `true`, `yes`, `on`,
/// case-insensitive). Anything else — including unset — is treated as OFF, so
/// the default is a strict no-op. This gate is the ONLY thing that can demote a
/// recognized-extension file to Tier-2 on the basis of git-tracking; with it
/// off, the admission gate behaves identically to before SF-009.
pub fn exclude_untracked_enabled() -> bool {
    std::env::var(EXCLUDE_UNTRACKED_ENV)
        .map(|v| {
            let v = v.trim().to_ascii_lowercase();
            matches!(v.as_str(), "1" | "true" | "yes" | "on")
        })
        .unwrap_or(false)
}

/// Compute the set of git-tracked relative paths (forward-slash normalized) for
/// the repository containing `root`, for the SF-009 exclude-untracked policy.
///
/// **Fails open to `None`** when the policy is disabled, when no git repository
/// is discoverable, or when the git index cannot be read. A `None` result means
/// "do not demote anything" — never "treat every file as untracked". An empty
/// tracked set (readable but empty index) also yields `None` for the same
/// reason, so a freshly `git init`-ed tree does not demote every source file.
///
/// Uses the git index (`git ls-files` semantics) via [`crate::git::GitRepo`],
/// NOT the `ignore` crate — the `ignore` crate models gitignore rules but has no
/// concept of which files are tracked.
pub fn tracked_path_set_for_exclusion(root: &Path) -> Option<std::collections::HashSet<String>> {
    if !exclude_untracked_enabled() {
        return None;
    }
    let git_repo = crate::git::GitRepo::open(root).ok()?;
    let tracked = git_repo.tracked_paths().ok()?;
    if tracked.is_empty() {
        return None;
    }
    Some(tracked.into_iter().collect())
}

/// SF-012(B): git-tracked path set used to RESCUE source files the repo-root
/// build-dir heuristic (`is_under_repo_root_build_dir`) would otherwise skip.
///
/// Unlike [`tracked_path_set_for_exclusion`] this is NOT env-gated: build output
/// is never git-tracked, so a tracked path matching `target-<alnum>` (e.g.
/// tokio's `target-specs/`) is real source, not a build artifact, and must be
/// admitted. **Fails open to `None`** (heuristic decides alone) when no git repo
/// is discoverable, the index cannot be read, or the tracked set is empty — so a
/// non-git tree keeps the conservative build-dir skip exactly as before.
fn tracked_path_set_for_build_dir_rescue(root: &Path) -> Option<std::collections::HashSet<String>> {
    let git_repo = crate::git::GitRepo::open(root).ok()?;
    let tracked = git_repo.tracked_paths().ok()?;
    if tracked.is_empty() {
        return None;
    }
    Some(tracked.into_iter().collect())
}

#[cfg(test)]
mod tests {
    use super::*;
    use std::fs;
    use tempfile::TempDir;

    fn create_file(dir: &Path, name: &str, content: &str) {
        let path = dir.join(name);
        if let Some(parent) = path.parent() {
            fs::create_dir_all(parent).unwrap();
        }
        fs::write(path, content).unwrap();
    }

    #[test]
    fn test_discover_files_finds_rs_py_js() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "main.rs", "fn main() {}");
        create_file(tmp.path(), "script.py", "def foo(): pass");
        create_file(tmp.path(), "app.js", "function bar() {}");

        let files = discover_files(tmp.path()).unwrap();
        let extensions: Vec<&str> = files
            .iter()
            .map(|f| f.relative_path.rsplit('.').next().unwrap())
            .collect();

        assert!(extensions.contains(&"rs"), "should find .rs");
        assert!(extensions.contains(&"py"), "should find .py");
        assert!(extensions.contains(&"js"), "should find .js");
        assert_eq!(files.len(), 3);
    }

    #[test]
    fn test_discover_files_includes_config_files() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "config.json", "{}");
        create_file(tmp.path(), "README.md", "# readme");
        create_file(tmp.path(), "Cargo.toml", "[package]");
        create_file(tmp.path(), "main.rs", "fn main() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 4, "should discover .rs + .json + .md + .toml");
        let paths: Vec<&str> = files.iter().map(|f| f.relative_path.as_str()).collect();
        assert!(paths.contains(&"config.json"), "should find .json");
        assert!(paths.contains(&"README.md"), "should find .md");
        assert!(paths.contains(&"Cargo.toml"), "should find .toml");
        assert!(paths.contains(&"main.rs"), "should find .rs");
    }

    #[test]
    fn test_discover_files_respects_gitignore() {
        let tmp = TempDir::new().unwrap();
        // Must create .git dir for gitignore to be respected
        fs::create_dir(tmp.path().join(".git")).unwrap();
        fs::write(tmp.path().join(".gitignore"), "ignored.rs\n").unwrap();

        create_file(tmp.path(), "main.rs", "fn main() {}");
        create_file(tmp.path(), "ignored.rs", "fn ignored() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 1, "ignored.rs should be excluded");
        assert_eq!(files[0].relative_path, "main.rs");
    }

    #[test]
    fn test_discover_files_normalizes_backslashes() {
        let tmp = TempDir::new().unwrap();
        // Create a file in a subdirectory — the path separator will be OS-native
        create_file(tmp.path(), "src/lib.rs", "pub fn lib() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 1);
        // Must use forward slashes regardless of OS
        assert!(
            !files[0].relative_path.contains('\\'),
            "should have no backslashes: {:?}",
            files[0].relative_path
        );
        assert!(files[0].relative_path.contains('/') || files[0].relative_path == "src/lib.rs");
    }

    #[test]
    fn test_discover_files_deterministic_sorted_order() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "Zoo.rs", "fn zoo() {}");
        create_file(tmp.path(), "apple.rs", "fn apple() {}");
        create_file(tmp.path(), "Mango.rs", "fn mango() {}");

        let files = discover_files(tmp.path()).unwrap();
        assert_eq!(files.len(), 3);
        // Case-insensitive alphabetical order
        let names: Vec<&str> = files.iter().map(|f| f.relative_path.as_str()).collect();
        // "apple" < "Mango" < "Zoo" case-insensitively
        let lower: Vec<String> = names.iter().map(|n| n.to_lowercase()).collect();
        let mut sorted = lower.clone();
        sorted.sort();
        assert_eq!(
            lower, sorted,
            "files should be in case-insensitive sorted order"
        );
    }

    #[test]
    fn test_discover_files_assigns_classification_tags_from_path() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "tests/unit_spec.rs", "fn spec_case() {}");
        create_file(tmp.path(), "vendor/pkg/lib.rs", "fn vendored() {}");
        create_file(
            tmp.path(),
            "src/generated/client.generated.rs",
            "fn generated() {}",
        );

        let files = discover_files(tmp.path()).unwrap();
        let by_path: std::collections::HashMap<&str, &DiscoveredFile> = files
            .iter()
            .map(|file| (file.relative_path.as_str(), file))
            .collect();

        assert!(
            by_path["tests/unit_spec.rs"].classification.is_test,
            "tests path should set is_test"
        );
        assert!(
            by_path["vendor/pkg/lib.rs"].classification.is_vendor,
            "vendor path should set is_vendor"
        );
        assert!(
            by_path["src/generated/client.generated.rs"]
                .classification
                .is_generated,
            "generated path should set is_generated"
        );
    }

    // ── repo-root Cargo build-dir skip (X6) ──
    //
    // Build dirs at the repo root (`target`, `target-wsl`, `CARGO_TARGET_DIR`)
    // must be skipped independently of `.gitignore`, while normal source and a
    // legitimately-named non-build dir stay indexed. The skip inspects only the
    // first path component, so a nested `src/target/` source dir is preserved.
    mod cargo_build_dir_skip {
        use super::*;

        #[test]
        fn build_dir_name_matcher_classifies_correctly() {
            // Matches: bare `target` and `target-<alnum/underscore suffix>`.
            assert!(is_cargo_build_dir_name("target"));
            assert!(is_cargo_build_dir_name("target-wsl"));
            assert!(is_cargo_build_dir_name("target-debug"));
            assert!(is_cargo_build_dir_name("target-x86_64"));
            assert!(is_cargo_build_dir_name("target-CI_2"));
            // Non-matches: lookalikes that are legitimate source dir names.
            assert!(!is_cargo_build_dir_name("targets"));
            assert!(!is_cargo_build_dir_name("target_dir"));
            assert!(!is_cargo_build_dir_name("target-"));
            assert!(!is_cargo_build_dir_name("target-foo/bar"));
            assert!(!is_cargo_build_dir_name("my-target"));
            assert!(!is_cargo_build_dir_name("src"));
        }

        #[test]
        fn under_repo_root_build_dir_only_matches_root_child() {
            // Root-level build dirs are skipped.
            assert!(is_under_repo_root_build_dir("target/debug/foo.rs", None));
            assert!(is_under_repo_root_build_dir(
                "target-wsl/release/x.rs",
                None
            ));
            // A nested source dir literally named `target` is NOT skipped.
            assert!(!is_under_repo_root_build_dir("src/target/mod.rs", None));
            // A root-level FILE (no separator) is not a build dir.
            assert!(!is_under_repo_root_build_dir("target", None));
            // Normal source is untouched.
            assert!(!is_under_repo_root_build_dir("src/lib.rs", None));
            // CARGO_TARGET_DIR child is skipped when supplied.
            assert!(is_under_repo_root_build_dir(
                "build-out/app.rs",
                Some("build-out")
            ));
            assert!(!is_under_repo_root_build_dir(
                "src/build-out/app.rs",
                Some("build-out")
            ));
        }

        #[test]
        fn discover_files_skips_target_wsl_keeps_source() {
            let tmp = TempDir::new().unwrap();
            // A build-dir variant that is NOT gitignored here.
            create_file(
                tmp.path(),
                "target-wsl/debug/build_artifact.rs",
                "fn a() {}",
            );
            // Bare `target` build dir.
            create_file(tmp.path(), "target/debug/other.rs", "fn b() {}");
            // Normal source MUST be indexed.
            create_file(tmp.path(), "src/lib.rs", "pub fn lib() {}");
            // A legitimately-named non-build dir must NOT be over-skipped.
            create_file(tmp.path(), "targets/config.rs", "fn cfg() {}");
            // A nested source dir literally named `target` must be preserved.
            create_file(tmp.path(), "src/target/mod.rs", "fn nested() {}");

            let files = discover_files(tmp.path()).unwrap();
            let paths: Vec<&str> = files.iter().map(|f| f.relative_path.as_str()).collect();

            assert!(
                !paths.contains(&"target-wsl/debug/build_artifact.rs"),
                "target-wsl/ build output must be skipped: {paths:?}"
            );
            assert!(
                !paths.contains(&"target/debug/other.rs"),
                "target/ build output must be skipped: {paths:?}"
            );
            assert!(
                paths.contains(&"src/lib.rs"),
                "normal source must be indexed: {paths:?}"
            );
            assert!(
                paths.contains(&"targets/config.rs"),
                "non-build dir `targets/` must not be over-skipped: {paths:?}"
            );
            assert!(
                paths.contains(&"src/target/mod.rs"),
                "nested source dir `src/target/` must be preserved: {paths:?}"
            );
        }

        #[test]
        fn discover_all_files_skips_target_wsl_keeps_source() {
            let tmp = TempDir::new().unwrap();
            create_file(tmp.path(), "target-wsl/debug/artifact.rs", "fn a() {}");
            create_file(tmp.path(), "src/main.rs", "fn main() {}");
            create_file(tmp.path(), "targets/x.rs", "fn x() {}");

            let entries = discover_all_files(tmp.path()).unwrap();
            let paths: Vec<&str> = entries.iter().map(|e| e.relative_path.as_str()).collect();

            assert!(
                !paths.contains(&"target-wsl/debug/artifact.rs"),
                "target-wsl/ build output must be skipped in full discovery: {paths:?}"
            );
            assert!(
                paths.contains(&"src/main.rs"),
                "normal source must be discovered: {paths:?}"
            );
            assert!(
                paths.contains(&"targets/x.rs"),
                "non-build dir `targets/` must not be over-skipped: {paths:?}"
            );
        }
    }

    // ── SF-004 / SF-012(A): unsupported-language admission demotion ──
    //
    // A small, non-binary file with an extension that maps to no supported
    // grammar must be admitted Tier-2 (metadata-only / unsupported-language),
    // NOT stored with a contradictory Tier-1/Normal decision that vanishes from
    // tier accounting and mints a false "File not found".
    mod unsupported_language {
        use super::*;

        #[test]
        fn unsupported_language_decision_demotes_normal_to_metadata_only() {
            // classify_admission returns Normal for a small non-binary file (it
            // never inspects language); the helper must demote it honestly.
            let normal = AdmissionDecision::normal();
            let demoted = unsupported_language_decision(normal);
            assert_eq!(demoted.tier, AdmissionTier::MetadataOnly);
            assert_eq!(demoted.reason, Some(SkipReason::UnsupportedLanguage));
        }

        #[test]
        fn unsupported_language_decision_preserves_specific_skip_reasons() {
            // A real size/denylist/binary skip must pass through unchanged — the
            // helper only rewrites the contradictory Normal-but-unparseable state.
            for original in [
                AdmissionDecision::skip(AdmissionTier::HardSkip, SkipReason::SizeCeiling),
                AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::SizeThreshold),
                AdmissionDecision::skip(AdmissionTier::MetadataOnly, SkipReason::BinaryContent),
                AdmissionDecision::skip(
                    AdmissionTier::MetadataOnly,
                    SkipReason::DependencyLockfile,
                ),
            ] {
                assert_eq!(
                    unsupported_language_decision(original),
                    original,
                    "non-Normal decision must be returned unchanged"
                );
            }
        }

        #[test]
        fn unsupported_language_reason_renders_honestly() {
            assert_eq!(
                SkipReason::UnsupportedLanguage.to_string(),
                "unsupported language"
            );
        }
    }

    // ── SF-025: hidden-path scan-policy predicate ──
    //
    // The bulk walk skips hidden dotfiles/dotdirs; the single-file (re)index path
    // must mirror that so index membership is deterministic from scan policy.
    mod hidden_path {
        use super::*;

        #[test]
        fn detects_hidden_directory_component() {
            assert!(path_has_hidden_component(".github/workflows/ci.yml"));
            assert!(path_has_hidden_component(
                "deps/hiredis/.github/release.yml"
            ));
            assert!(path_has_hidden_component(".travis.yml"));
            assert!(path_has_hidden_component("a/b/.hidden"));
        }

        #[test]
        fn allows_visible_paths_and_traversal_segments() {
            assert!(!path_has_hidden_component("src/main.rs"));
            assert!(!path_has_hidden_component("README.md"));
            // `.`/`..` traversal segments are not "hidden" file names.
            assert!(!path_has_hidden_component("./src/main.rs"));
            assert!(!path_has_hidden_component("../sibling/main.rs"));
            // A dot inside a name (not a leading-dot component) is visible.
            assert!(!path_has_hidden_component("src/a.b.c/main.rs"));
        }

        #[test]
        fn discover_all_files_skips_hidden_supported_extension_files() {
            let tmp = TempDir::new().unwrap();
            // A hidden dir with a SUPPORTED extension inside — the exact SF-025
            // shape (e.g. `.github/workflows/ci.yml`).
            create_file(tmp.path(), ".github/workflows/ci.yml", "name: ci\n");
            create_file(tmp.path(), ".travis.yml", "language: rust\n");
            // Visible source must still be discovered.
            create_file(tmp.path(), "src/main.rs", "fn main() {}");

            let entries = discover_all_files(tmp.path()).unwrap();
            let paths: Vec<&str> = entries.iter().map(|e| e.relative_path.as_str()).collect();

            assert!(
                !paths.contains(&".github/workflows/ci.yml"),
                "hidden-dir file must not be discovered by the bulk walk: {paths:?}"
            );
            assert!(
                !paths.contains(&".travis.yml"),
                "hidden dotfile must not be discovered by the bulk walk: {paths:?}"
            );
            assert!(
                paths.contains(&"src/main.rs"),
                "visible source must be discovered: {paths:?}"
            );
        }
    }

    // ── SF-012(B): build-dir heuristic rescues tracked source dirs ──
    mod build_dir_tracked_rescue {
        use super::*;
        use std::process::Command;

        #[test]
        fn discover_all_files_rescues_tracked_target_specs_dir() {
            let tmp = TempDir::new().unwrap();
            let root = tmp.path();
            let run = |args: &[&str]| {
                Command::new("git")
                    .args(args)
                    .current_dir(root)
                    .output()
                    .expect("git command");
            };
            run(&["init"]);
            run(&["config", "user.email", "test@test.com"]);
            run(&["config", "user.name", "Test"]);

            // tokio's real shape: a tracked `target-specs/` source dir whose name
            // matches the build-dir heuristic, plus a genuine `target/` build dir.
            create_file(root, "target-specs/i686.json", "{}\n");
            create_file(root, "target-specs/README.md", "# specs\n");
            create_file(root, "src/main.rs", "fn main() {}");
            // Stage+commit ONLY the source and the target-specs dir — the build
            // dir below is left untracked, exactly like real build output.
            run(&["add", "target-specs", "src"]);
            run(&["commit", "-m", "initial"]);
            // A genuine (untracked) build dir matching the heuristic.
            create_file(root, "target/debug/artifact.rs", "fn a() {}");

            let entries = discover_all_files(root).unwrap();
            let paths: Vec<&str> = entries.iter().map(|e| e.relative_path.as_str()).collect();

            assert!(
                paths.contains(&"target-specs/i686.json"),
                "tracked target-specs/ source must be rescued: {paths:?}"
            );
            assert!(
                paths.contains(&"target-specs/README.md"),
                "tracked target-specs/ source must be rescued: {paths:?}"
            );
            assert!(
                paths.contains(&"src/main.rs"),
                "normal source must be discovered: {paths:?}"
            );
            assert!(
                !paths.contains(&"target/debug/artifact.rs"),
                "untracked build output must still be skipped: {paths:?}"
            );
        }

        #[test]
        fn discover_all_files_without_git_keeps_conservative_build_dir_skip() {
            // No git repo: the rescue helper fails open to None, so the heuristic
            // decides alone and a `target-*` dir is skipped exactly as before.
            let tmp = TempDir::new().unwrap();
            create_file(tmp.path(), "target-wsl/debug/artifact.rs", "fn a() {}");
            create_file(tmp.path(), "src/main.rs", "fn main() {}");

            let entries = discover_all_files(tmp.path()).unwrap();
            let paths: Vec<&str> = entries.iter().map(|e| e.relative_path.as_str()).collect();

            assert!(
                !paths.contains(&"target-wsl/debug/artifact.rs"),
                "non-git tree must still skip target-* build dirs: {paths:?}"
            );
            assert!(
                paths.contains(&"src/main.rs"),
                "normal source must be discovered: {paths:?}"
            );
        }
    }

    #[test]
    fn test_is_forbidden_root_blocks_home_dir() {
        let home = home_dir();
        if let Some(h) = home {
            assert!(is_forbidden_root(&h), "home directory should be forbidden");
        }
    }

    #[test]
    fn test_is_forbidden_root_blocks_drive_root() {
        #[cfg(target_os = "windows")]
        assert!(is_forbidden_root(Path::new("C:\\")));
        #[cfg(not(target_os = "windows"))]
        assert!(is_forbidden_root(Path::new("/")));
    }

    #[test]
    fn test_is_forbidden_root_blocks_system_dirs() {
        assert!(is_forbidden_root(Path::new("/tmp")));
        assert!(is_forbidden_root(Path::new("/home")));
    }

    #[test]
    fn test_is_forbidden_root_allows_project_dirs() {
        let tmp = TempDir::new().unwrap();
        assert!(
            !is_forbidden_root(tmp.path()),
            "temp project dir should be allowed"
        );
    }

    #[test]
    fn test_is_forbidden_root_allows_project_named_tmp() {
        let tmp = TempDir::new().unwrap();
        let project = tmp.path().join("projects").join("tmp");
        std::fs::create_dir_all(&project).unwrap();
        assert!(
            !is_forbidden_root(&project),
            "project at C:\\projects\\tmp must not be rejected by basename"
        );
    }

    #[test]
    fn test_is_forbidden_root_allows_project_named_var() {
        let tmp = TempDir::new().unwrap();
        let project = tmp.path().join("workspace").join("var");
        std::fs::create_dir_all(&project).unwrap();
        assert!(
            !is_forbidden_root(&project),
            "project at workspace/var must not be rejected by basename"
        );
    }

    #[test]
    fn test_is_forbidden_root_still_blocks_top_level_tmp_on_unix() {
        // Skip on Windows where /tmp doesn't apply
        #[cfg(unix)]
        {
            // /tmp itself is a real path; canonicalize will succeed.
            let path = std::path::Path::new("/tmp");
            if path.exists() {
                assert!(
                    is_forbidden_root(path),
                    "/tmp must still be blocked as system path"
                );
            }
        }
    }

    #[test]
    fn test_is_forbidden_root_still_blocks_windows_system_paths() {
        #[cfg(target_os = "windows")]
        {
            let path = std::path::Path::new(r"C:\Windows\System32");
            if path.exists() {
                assert!(
                    is_forbidden_root(path),
                    "C:\\Windows\\System32 must remain blocked"
                );
            }
        }
    }

    #[test]
    fn test_binary_sniff_detects_null_bytes() {
        let content = b"hello\x00world";
        assert!(is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_allows_pure_utf8() {
        let content = b"fn main() { println!(\"hello\"); }";
        assert!(!is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_empty_file() {
        assert!(!is_binary_content(b""));
    }

    #[test]
    fn test_binary_sniff_detects_invalid_utf8() {
        let content: &[u8] = &[0x80, 0x81, 0x82, 0x83, 0x84];
        assert!(is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_detects_high_control_ratio() {
        let mut content = Vec::new();
        content.extend(std::iter::repeat_n(0x01, 80)); // SOH — control char
        content.extend(std::iter::repeat_n(b'A', 20)); // printable
        // 80% control bytes > 30% threshold -> binary
        assert!(is_binary_content(&content));
    }

    #[test]
    fn test_binary_sniff_allows_low_control_ratio() {
        let content = b"line1\tvalue1\nline2\tvalue2\nline3\tvalue3\n";
        assert!(!is_binary_content(content));
    }

    #[test]
    fn test_binary_sniff_allows_common_whitespace_controls() {
        let content = b"col1\tcol2\tcol3\r\nval1\tval2\tval3\r\n";
        assert!(!is_binary_content(content));
    }

    // ── classify_admission tests ──

    use crate::domain::index::{AdmissionDecision, AdmissionTier, SkipReason};

    #[test]
    fn test_huge_text_file_is_hard_skip() {
        let decision =
            classify_admission(std::path::Path::new("huge.txt"), 150 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::HardSkip);
        assert_eq!(decision.reason, Some(SkipReason::SizeCeiling));
    }

    #[test]
    fn test_small_ckpt_is_metadata_only() {
        let decision = classify_admission(std::path::Path::new("model.ckpt"), 50 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::DenylistedExtension));
    }

    #[test]
    fn test_huge_ckpt_is_hard_skip() {
        let decision = classify_admission(std::path::Path::new("big.ckpt"), 4_200_000_000, None);
        assert_eq!(decision.tier, AdmissionTier::HardSkip);
        assert_eq!(decision.reason, Some(SkipReason::SizeCeiling));
    }

    #[test]
    fn test_large_json_is_metadata_only() {
        let decision = classify_admission(std::path::Path::new("big.json"), 2 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::SizeThreshold));
    }

    #[test]
    fn test_small_txt_is_normal() {
        let decision = classify_admission(std::path::Path::new("readme.txt"), 50 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_medium_rust_source_is_normal() {
        let decision = classify_admission(std::path::Path::new("big_module.rs"), 500 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_binary_content_is_metadata_only() {
        let content = b"ELF\x00\x00\x00binary";
        let decision =
            classify_admission(std::path::Path::new("unknown_file"), 1024, Some(content));
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::BinaryContent));
    }

    #[test]
    fn test_svg_not_denylisted() {
        let decision = classify_admission(std::path::Path::new("icon.svg"), 50 * 1024, None);
        assert_eq!(decision, AdmissionDecision::normal());
    }

    #[test]
    fn test_large_svg_is_metadata_only_by_size() {
        let decision = classify_admission(std::path::Path::new("huge.svg"), 2 * 1024 * 1024, None);
        assert_eq!(decision.tier, AdmissionTier::MetadataOnly);
        assert_eq!(decision.reason, Some(SkipReason::SizeThreshold));
    }

    // ── discover_all_files + admission gate integration tests ──

    #[test]
    fn test_discovery_skips_denylisted_extension() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "main.rs", "fn main() {}");
        // Write a fake .safetensors file (extension is on the denylist)
        fs::write(tmp.path().join("model.safetensors"), b"fake model bytes").unwrap();

        let entries = discover_all_files(tmp.path()).unwrap();

        // Classify each entry and collect skipped ones
        let mut rs_found = false;
        let mut safetensors_skipped = false;
        let mut safetensors_reason = None;

        for entry in &entries {
            let size = entry.file_size;
            let decision = classify_admission(&entry.absolute_path, size, None);
            if entry.relative_path == "main.rs" {
                assert_eq!(
                    decision.tier,
                    AdmissionTier::Normal,
                    ".rs file should be Normal"
                );
                rs_found = true;
            }
            if entry.relative_path == "model.safetensors" {
                assert_eq!(
                    decision.tier,
                    AdmissionTier::MetadataOnly,
                    ".safetensors should be MetadataOnly"
                );
                safetensors_skipped = true;
                safetensors_reason = decision.reason;
            }
        }

        assert!(rs_found, ".rs file must appear in discovered entries");
        assert!(
            safetensors_skipped,
            ".safetensors must appear in discovered entries and be skipped"
        );
        assert_eq!(
            safetensors_reason,
            Some(SkipReason::DenylistedExtension),
            ".safetensors skip reason must be DenylistedExtension"
        );
    }

    #[test]
    fn test_discovery_deferred_binary_sniff_reclassifies() {
        let tmp = TempDir::new().unwrap();
        create_file(tmp.path(), "lib.rs", "pub fn hello() {}");

        // Write a .dat file with NUL-heavy content: not on denylist, under 1MB,
        // but binary sniff (NUL bytes) should reclassify to MetadataOnly.
        let mut binary_content = vec![0u8; 512]; // NUL bytes — triggers binary sniff
        binary_content.extend_from_slice(b"some trailing text");
        fs::write(tmp.path().join("custom.dat"), &binary_content).unwrap();

        let entries = discover_all_files(tmp.path()).unwrap();

        let mut rs_normal = false;
        let mut dat_skipped = false;
        let mut dat_reason = None;

        for entry in &entries {
            let size = entry.file_size;

            // Phase 1: pre-content check
            let pre = classify_admission(&entry.absolute_path, size, None);

            if entry.relative_path == "lib.rs" {
                assert_eq!(pre.tier, AdmissionTier::Normal);
                rs_normal = true;
            }

            if entry.relative_path == "custom.dat" {
                // Pre-content: should be Normal (not denylisted, under 1MB)
                assert_eq!(
                    pre.tier,
                    AdmissionTier::Normal,
                    "custom.dat should be Normal before binary sniff"
                );

                // Phase 2: with content — binary sniff should reclassify
                let content = fs::read(&entry.absolute_path).unwrap();
                let post = classify_admission(&entry.absolute_path, size, Some(&content));
                assert_eq!(
                    post.tier,
                    AdmissionTier::MetadataOnly,
                    "custom.dat should be MetadataOnly after binary sniff"
                );
                dat_skipped = true;
                dat_reason = post.reason;
            }
        }

        assert!(rs_normal, "lib.rs must be Normal");
        assert!(
            dat_skipped,
            "custom.dat must be discovered and reclassified"
        );
        assert_eq!(
            dat_reason,
            Some(SkipReason::BinaryContent),
            "custom.dat skip reason must be BinaryContent"
        );
    }

    // ── bounded discovery (resource ceilings) ──
    //
    // These guard against OOM/panic on a huge but NON-sensitive tree: discovery
    // must refuse with a graceful, explicit error before committing the full set
    // to the in-memory index build. Env-mutating cases are serialized by a mutex
    // and restore the prior value on drop so they don't race other env readers.
    mod bounded_discovery {
        use super::*;
        use std::ffi::OsString;
        use std::sync::Mutex;

        // Serializes the env-mutating limit tests against each other. Discovery
        // env vars are process-global, so two tests setting them concurrently
        // would interfere.
        static ENV_LOCK: Mutex<()> = Mutex::new(());

        struct LimitEnvGuard {
            files_prev: Option<OsString>,
            bytes_prev: Option<OsString>,
        }

        #[allow(unsafe_code)] // test-only env guard; mutation is serialized by ENV_LOCK.
        impl LimitEnvGuard {
            /// Set both limit env vars (any `None` clears that var) and capture
            /// the prior values for restoration on drop.
            fn set(max_files: Option<&str>, max_bytes: Option<&str>) -> Self {
                let files_prev = std::env::var_os(MAX_INDEX_FILES_ENV);
                let bytes_prev = std::env::var_os(MAX_INDEX_BYTES_ENV);
                // SAFETY: env mutation is serialized by ENV_LOCK held by the caller;
                // no concurrent env readers in this single-threaded test section.
                unsafe {
                    match max_files {
                        Some(v) => std::env::set_var(MAX_INDEX_FILES_ENV, v),
                        None => std::env::remove_var(MAX_INDEX_FILES_ENV),
                    }
                    match max_bytes {
                        Some(v) => std::env::set_var(MAX_INDEX_BYTES_ENV, v),
                        None => std::env::remove_var(MAX_INDEX_BYTES_ENV),
                    }
                }
                Self {
                    files_prev,
                    bytes_prev,
                }
            }
        }

        #[allow(unsafe_code)] // test-only env guard; restores serialized env mutation.
        impl Drop for LimitEnvGuard {
            fn drop(&mut self) {
                // SAFETY: env mutation is serialized by ENV_LOCK; restore prior state.
                unsafe {
                    match &self.files_prev {
                        Some(v) => std::env::set_var(MAX_INDEX_FILES_ENV, v),
                        None => std::env::remove_var(MAX_INDEX_FILES_ENV),
                    }
                    match &self.bytes_prev {
                        Some(v) => std::env::set_var(MAX_INDEX_BYTES_ENV, v),
                        None => std::env::remove_var(MAX_INDEX_BYTES_ENV),
                    }
                }
            }
        }

        #[test]
        fn default_limits_are_generous() {
            let limits = DiscoveryLimits::default();
            assert_eq!(limits.max_files, DEFAULT_MAX_INDEX_FILES);
            assert_eq!(limits.max_bytes, DEFAULT_MAX_INDEX_BYTES);
            // 200k files is comfortably above a very large real monorepo.
            assert!(limits.max_files >= 200_000);
        }

        #[test]
        fn parse_positive_env_rejects_zero_empty_and_garbage() {
            let _lock = ENV_LOCK.lock().unwrap();
            let _guard = LimitEnvGuard::set(Some("0"), Some("not-a-number"));
            // Zero and non-numeric overrides are ignored, so the defaults stand —
            // a typo can never silently disable indexing.
            assert_eq!(parse_positive_env(MAX_INDEX_FILES_ENV), None);
            assert_eq!(parse_positive_env(MAX_INDEX_BYTES_ENV), None);
            let limits = DiscoveryLimits::from_env();
            assert_eq!(limits.max_files, DEFAULT_MAX_INDEX_FILES);
            assert_eq!(limits.max_bytes, DEFAULT_MAX_INDEX_BYTES);
        }

        #[test]
        fn from_env_honors_valid_override() {
            let _lock = ENV_LOCK.lock().unwrap();
            let _guard = LimitEnvGuard::set(Some("5"), Some("4096"));
            let limits = DiscoveryLimits::from_env();
            assert_eq!(limits.max_files, 5);
            assert_eq!(limits.max_bytes, 4096);
        }

        #[test]
        fn normal_repo_indexes_under_default_cap() {
            // No env override: the generous default cap must not interfere with a
            // small, ordinary project.
            let _lock = ENV_LOCK.lock().unwrap();
            let _guard = LimitEnvGuard::set(None, None);
            let tmp = TempDir::new().unwrap();
            create_file(tmp.path(), "main.rs", "fn main() {}");
            create_file(tmp.path(), "lib.rs", "pub fn f() {}");
            create_file(tmp.path(), "README.md", "# hi");

            let files = discover_files(tmp.path()).expect("normal repo indexes fine");
            assert_eq!(files.len(), 3);

            let entries = discover_all_files(tmp.path()).expect("normal repo full-discovery fine");
            assert!(entries.len() >= 3);
        }

        #[test]
        fn over_file_cap_yields_graceful_error_not_panic() {
            let _lock = ENV_LOCK.lock().unwrap();
            // Cap at 2 files; create 5 source files to exceed it.
            let _guard = LimitEnvGuard::set(Some("2"), None);
            let tmp = TempDir::new().unwrap();
            for i in 0..5 {
                create_file(tmp.path(), &format!("f{i}.rs"), "fn x() {}");
            }

            let err = discover_files(tmp.path()).expect_err("over file cap must error");
            let msg = err.to_string();
            assert!(
                msg.contains("tree too large to index"),
                "error must be the graceful over-cap message: {msg}"
            );
            assert!(
                msg.contains(MAX_INDEX_FILES_ENV),
                "error must name the override knob: {msg}"
            );

            let err2 = discover_all_files(tmp.path()).expect_err("full discovery over cap errors");
            assert!(err2.to_string().contains("tree too large to index"));
        }

        #[test]
        fn over_byte_cap_yields_graceful_error() {
            let _lock = ENV_LOCK.lock().unwrap();
            // Very high file cap, tiny byte cap (8 bytes). A single non-empty file
            // exceeds the byte ceiling, exercising the cumulative-bytes path that
            // only `discover_all_files` enforces.
            let _guard = LimitEnvGuard::set(Some("1000000"), Some("8"));
            let tmp = TempDir::new().unwrap();
            create_file(
                tmp.path(),
                "big.rs",
                "fn this_is_more_than_eight_bytes() {}",
            );

            let err = discover_all_files(tmp.path()).expect_err("over byte cap must error");
            let msg = err.to_string();
            assert!(
                msg.contains("tree too large to index"),
                "error must be the graceful over-cap message: {msg}"
            );
            assert!(
                msg.contains(MAX_INDEX_BYTES_ENV),
                "error must name the byte override knob: {msg}"
            );
        }

        #[test]
        fn raised_cap_lets_a_previously_over_cap_tree_index() {
            let _lock = ENV_LOCK.lock().unwrap();
            let tmp = TempDir::new().unwrap();
            for i in 0..4 {
                create_file(tmp.path(), &format!("f{i}.rs"), "fn x() {}");
            }
            // Low cap: refused.
            {
                let _guard = LimitEnvGuard::set(Some("2"), None);
                assert!(discover_files(tmp.path()).is_err());
            }
            // Raised cap: accepted — the limit is genuinely configurable.
            {
                let _guard = LimitEnvGuard::set(Some("100"), None);
                let files = discover_files(tmp.path()).expect("raised cap indexes the tree");
                assert_eq!(files.len(), 4);
            }
        }
    }

    // ── `.git` fast-path sensitive-root guard (find_project_root) ──
    //
    // A `.git` planted under a forbidden/sensitive ancestor must NOT be selected
    // as the project root; a genuine project `.git` still is. We exercise the
    // guard helper `is_forbidden_root` that the fast-path now consults, on
    // synthetic sensitive shapes, plus a positive case for an ordinary repo.
    mod git_fast_path_guard {
        use super::*;

        #[test]
        fn ordinary_project_with_git_is_not_forbidden() {
            // A normal temp project dir is not sensitive/forbidden, so a `.git`
            // there would be selected by the fast-path.
            let tmp = TempDir::new().unwrap();
            fs::create_dir_all(tmp.path().join(".git")).unwrap();
            assert!(
                !is_forbidden_root(tmp.path()),
                "an ordinary project dir must remain selectable as a git root"
            );
        }

        #[cfg(not(target_os = "windows"))]
        #[test]
        fn sensitive_unix_root_with_git_is_forbidden() {
            // `/etc` is sensitive; a planted `/etc/.git` must NOT be selected.
            // We assert the guard the fast-path consults rejects the root itself,
            // independent of whether the path exists on the test host.
            assert!(
                is_forbidden_root(Path::new("/etc")),
                "/etc must be forbidden even if a `.git` is planted there"
            );
            assert!(crate::paths::is_sensitive_path(Path::new("/etc")));
        }

        #[cfg(target_os = "windows")]
        #[test]
        fn sensitive_windows_root_with_git_is_forbidden() {
            // A bare drive root and the Windows user container are sensitive; a
            // planted `.git` there must NOT be selected by the fast-path.
            assert!(is_forbidden_root(Path::new("C:\\Windows")));
            assert!(crate::paths::is_sensitive_path(Path::new("C:\\Windows")));
        }
    }

    // WSL DrvFs Windows-profile / drive-root guard (rule 4c).
    //
    // These exercise the pure path-shape helper `is_wsl_windows_container_path`,
    // which is independent of the WSL probe and therefore deterministic on any
    // non-Windows host (CI, macOS, native Linux). The helper only exists on
    // non-Windows targets, so the whole group is gated to match.
    #[cfg(not(target_os = "windows"))]
    mod wsl_drvfs {
        use super::*;

        // --- forbidden: the broad container roots that caused the hang ---

        #[test]
        fn blocks_bare_drive_root() {
            assert!(is_wsl_windows_container_path(Path::new("/mnt/c")));
            assert!(is_wsl_windows_container_path(Path::new("/mnt/d")));
        }

        #[test]
        fn blocks_users_container() {
            assert!(is_wsl_windows_container_path(Path::new("/mnt/c/Users")));
        }

        #[test]
        fn blocks_bare_profile_root() {
            // The exact reported hang path.
            assert!(is_wsl_windows_container_path(Path::new(
                "/mnt/c/Users/poslj"
            )));
        }

        #[test]
        fn blocks_other_drive_profile() {
            assert!(is_wsl_windows_container_path(Path::new(
                "/mnt/d/Users/alice"
            )));
        }

        #[test]
        fn blocks_case_insensitive_users_segment() {
            // DrvFs is case-insensitive but canonicalize is case-preserving, so
            // `cd /mnt/c/users/...` reaches the identical Windows tree. All
            // casings of the profile container/root must be caught.
            assert!(is_wsl_windows_container_path(Path::new("/mnt/c/users")));
            assert!(is_wsl_windows_container_path(Path::new(
                "/mnt/c/USERS/poslj"
            )));
            assert!(is_wsl_windows_container_path(Path::new(
                "/mnt/c/UsErS/poslj"
            )));
        }

        // --- allowed: deep projects and lookalikes must stay indexable ---

        #[test]
        fn allows_deep_project_under_profile() {
            // A non-git project kept under the profile must NOT be forbidden;
            // the .git fast-path handles real repos, and deep dirs are scoped.
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/Users/poslj/dev/my-lib"
            )));
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/Users/poslj/Documents/project"
            )));
        }

        #[test]
        fn allows_non_users_mount_project() {
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/code/proj"
            )));
        }

        #[test]
        fn allows_users_named_deeper() {
            // A dir literally named Users but NOT at the /mnt/<drive>/Users
            // position must stay allowed (guards against over-broad matching).
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/code/Users"
            )));
        }

        #[test]
        fn allows_non_mnt_paths() {
            // Genuine Linux paths with a Users dir are not under /mnt.
            assert!(!is_wsl_windows_container_path(Path::new("/srv/Users/bob")));
            assert!(!is_wsl_windows_container_path(Path::new("/home/robert")));
        }

        #[test]
        fn allows_multichar_second_segment() {
            // comps[1] must be a single ASCII letter; multi-char (a real mount
            // name, not a drive) is allowed.
            assert!(!is_wsl_windows_container_path(Path::new("/mnt/cc/Users/x")));
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/wsl/Users/x"
            )));
        }

        #[test]
        fn allows_lookalike_prefixes() {
            // Substring/prefix lookalikes must not collide with `Users`.
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/Users-data/proj"
            )));
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/UserStuff/proj"
            )));
        }

        #[test]
        fn allows_bare_mnt() {
            assert!(!is_wsl_windows_container_path(Path::new("/mnt")));
        }

        #[test]
        fn parent_dir_escape_does_not_misfire() {
            // `..` is popped lexically rather than dropped, so a path whose real
            // target is a non-Users dir is not falsely forbidden.
            assert!(!is_wsl_windows_container_path(Path::new(
                "/mnt/c/Users/../code/proj"
            )));
            // `..` popping past the root yields no match (not a panic / false true).
            assert!(!is_wsl_windows_container_path(Path::new("/mnt/c/../..")));
        }
    }
}