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
|
package main
import (
"bytes"
"database/sql"
"encoding/xml"
"errors"
"flag"
"fmt"
"html"
"io"
"net/http"
"net/url"
"path"
"sort"
"strings"
"time"
chromahtml "github.com/alecthomas/chroma/v2/formatters/html"
"github.com/alecthomas/chroma/v2/styles"
zsqlite "zombiezen.com/go/sqlite"
)
// server holds the shared state for the HTTP handlers.
type server struct {
db *sql.DB // read pool; every request reads through a snapshot on this
writeDB *sql.DB // single-connection writer for the lazy render cache
dbPath string // path to the SQLite file, for opening short-lived blob conns
baseURL string
css []byte
}
// dbReader is the read subset shared by *sql.DB and *sql.Tx, so every
// request-time query helper can take either the pool or (as they all do now) a
// per-request read snapshot.
//
// Handlers open one read-only transaction per request and thread it through, so
// a whole page renders from a single consistent view of the database. This
// matters because a page captures the live generation once (projectHead) and
// then filters every later query on it: without a snapshot, an ingest
// committing mid-render — which flips head_generation and deletes the previous
// generation's rows in one transaction (see gcOldGenerations) — would make the
// remaining queries return nothing, yielding a half-empty page. In WAL mode the
// snapshot is fixed at the transaction's first read and costs nothing to hold.
//
// Writes deliberately do not go through this: the lazy render-cache fill
// (storeRender) runs in autocommit on the separate single-connection writer
// (openWriteDB), so it commits independently and is visible to later requests,
// and a write waiting out an ingest can never eat into this read pool.
type dbReader interface {
QueryRow(query string, args ...any) *sql.Row
Query(query string, args ...any) (*sql.Rows, error)
}
// runServe starts the HTTP server.
func runServe(args []string) error {
fs := flag.NewFlagSet("serve", flag.ContinueOnError)
dbPath := fs.String("db", "", "path to the SQLite database (required)")
addr := fs.String("addr", "127.0.0.1:8790", "listen address")
baseURL := fs.String("base-url", "", "public base URL (for absolute links; optional)")
if err := fs.Parse(args); err != nil {
return err
}
if *dbPath == "" {
return errors.New("--db is required")
}
db, err := openDB(*dbPath)
if err != nil {
return err
}
defer db.Close()
writeDB, err := openWriteDB(*dbPath)
if err != nil {
return err
}
defer writeDB.Close()
css, err := buildCSS()
if err != nil {
return err
}
s := &server{
db: db,
writeDB: writeDB,
dbPath: *dbPath,
baseURL: strings.TrimRight(*baseURL, "/"),
css: css,
}
mux := http.NewServeMux()
// More specific patterns win over "GET /" (Go 1.22+ ServeMux precedence),
// so these are not swallowed by the catch-all project handler.
mux.HandleFunc("GET /robots.txt", s.handleRobots)
mux.HandleFunc("GET /sitemap.xml", s.handleSitemap)
mux.HandleFunc("GET /", s.handleRoot)
fmt.Fprintf(stderr, "source-forge: listening on %s\n", *addr)
return http.ListenAndServe(*addr, mux)
}
// handleRoot dispatches: "/" is the project index; "/<project>/..." is a path
// within a project (directory listing or file view).
func (s *server) handleRoot(w http.ResponseWriter, r *http.Request) {
// One read-only snapshot for the whole request (see dbReader). Tying it to
// the request context means a client disconnecting mid-render releases the
// connection promptly instead of pinning it through a slow first render;
// the deferred Rollback covers the normal return path (we never commit).
tx, err := s.db.BeginTx(r.Context(), &sql.TxOptions{ReadOnly: true})
if err != nil {
httpError(w, err)
return
}
defer tx.Rollback()
reqPath := strings.TrimPrefix(r.URL.Path, "/")
if reqPath == "" {
s.serveProjectIndex(tx, w, r)
return
}
// "/<project>.bundle" serves the clonable git bundle for a project. This is
// a single top-level path with no slash, so handle it before the project
// path split below.
if !strings.Contains(reqPath, "/") && strings.HasSuffix(reqPath, ".bundle") {
s.serveBundle(tx, w, r, strings.TrimSuffix(reqPath, ".bundle"))
return
}
// "/<project>.tar.gz" serves the project's tree as a Nix-flake tarball
// (`nix run https://…/<project>.tar.gz#pkg`). Also a slashless top-level
// path, so handle it before the project split.
if !strings.Contains(reqPath, "/") && strings.HasSuffix(reqPath, ".tar.gz") {
s.serveTarball(tx, w, r, strings.TrimSuffix(reqPath, ".tar.gz"))
return
}
// Split into <project>/<rest>.
project, rest, _ := strings.Cut(reqPath, "/")
rest = strings.Trim(rest, "/")
// A trailing slash (or empty rest) means "directory".
wantDir := rest == "" || strings.HasSuffix(r.URL.Path, "/")
head, branch, ok := s.projectHead(tx, project)
if !ok {
http.NotFound(w, r)
return
}
if wantDir {
// "?full=" renders the whole subtree's file contents on one page.
if r.URL.Query().Has("full") {
s.serveFull(tx, w, r, project, head, rest)
return
}
s.serveDir(tx, w, r, project, branch, head, rest)
return
}
s.serveFile(tx, w, r, project, head, rest)
}
// projectHead returns the live generation and branch for a project.
func (s *server) projectHead(q dbReader, project string) (gen int64, branch string, ok bool) {
err := q.QueryRow(
`SELECT head_generation, branch FROM project WHERE name = ?`, project,
).Scan(&gen, &branch)
if err == sql.ErrNoRows {
return 0, "", false
}
if err != nil {
fmt.Fprintf(stderr, "projectHead(%q): %v\n", project, err)
return 0, "", false
}
return gen, branch, true
}
// baseFor returns the site's absolute base URL (no trailing slash). It prefers
// the configured --base-url and otherwise reconstructs one from the request
// (scheme + Host), so absolute links always resolve.
func (s *server) baseFor(r *http.Request) string {
if s.baseURL != "" {
return s.baseURL
}
scheme := "http"
if r.TLS != nil || strings.EqualFold(r.Header.Get("X-Forwarded-Proto"), "https") {
scheme = "https"
}
return scheme + "://" + r.Host
}
// hostFor returns just the host part of baseFor, for appending to <title> (see
// writePageHeader) so a page copy/pasted or screenshotted out of context still
// shows which site it came from.
func (s *server) hostFor(r *http.Request) string {
base := s.baseFor(r)
if _, host, ok := strings.Cut(base, "://"); ok {
return host
}
return base
}
// bundleURL returns the absolute URL of a project's bundle, so the clone hint
// always shows a working absolute URL.
func (s *server) bundleURL(r *http.Request, project string) string {
return s.baseFor(r) + "/" + project + ".bundle"
}
// hasBundle reports whether a project has a stored bundle at its live
// generation, and its size.
func (s *server) hasBundle(q dbReader, project string, gen int64) (size int64, ok bool) {
err := q.QueryRow(
`SELECT size FROM bundle WHERE project = ? AND generation = ?`, project, gen,
).Scan(&size)
if err != nil {
return 0, false
}
return size, true
}
// serveBundle streams the stored git bundle for a project as a download.
func (s *server) serveBundle(q dbReader, w http.ResponseWriter, r *http.Request, project string) {
gen, _, ok := s.projectHead(q, project)
if !ok {
http.NotFound(w, r)
return
}
var rowid, size int64
err := q.QueryRow(
`SELECT rowid, size FROM bundle WHERE project = ? AND generation = ?`,
project, gen,
).Scan(&rowid, &size)
if err == sql.ErrNoRows {
http.NotFound(w, r)
return
}
if err != nil {
httpError(w, err)
return
}
w.Header().Set("Content-Type", "application/x-git-bundle")
w.Header().Set("Content-Disposition",
fmt.Sprintf("attachment; filename=%q", project+".bundle"))
w.Header().Set("Cache-Control", "no-cache")
// Stream the blob straight from SQLite so a multi-MiB bundle is never held
// wholly in memory. Once the copy starts the status/headers are committed,
// so mid-stream errors can only be logged.
if err := s.streamBlob(w, "bundle", "data", rowid, size); err != nil {
fmt.Fprintf(stderr, "serveBundle(%q gen %d): %v\n", project, gen, err)
}
}
// tarballURL returns the absolute URL of a project's flake tarball, so the
// "nix run" hint always shows a working absolute URL.
func (s *server) tarballURL(r *http.Request, project string) string {
return s.baseFor(r) + "/" + project + ".tar.gz"
}
// hasTarball reports whether a project has a stored flake tarball at its live
// generation, and its size.
func (s *server) hasTarball(q dbReader, project string, gen int64) (size int64, ok bool) {
err := q.QueryRow(
`SELECT size FROM tarball WHERE project = ? AND generation = ?`, project, gen,
).Scan(&size)
if err != nil {
return 0, false
}
return size, true
}
// serveTarball streams the stored flake tarball for a project as a gzip
// download. It is content-addressed by generation via an ETag: because Nix
// forces tarball-ttl to 0 for flakes, every `nix run` re-validates, so we
// answer a matching If-None-Match with 304 Not Modified (no body, no re-unpack)
// to keep repeat runs nearly free.
func (s *server) serveTarball(q dbReader, w http.ResponseWriter, r *http.Request, project string) {
gen, _, ok := s.projectHead(q, project)
if !ok {
http.NotFound(w, r)
return
}
var rowid, size, modified int64
err := q.QueryRow(
`SELECT rowid, size, modified FROM tarball WHERE project = ? AND generation = ?`,
project, gen,
).Scan(&rowid, &size, &modified)
if err == sql.ErrNoRows {
http.NotFound(w, r)
return
}
if err != nil {
httpError(w, err)
return
}
// The generation is a monotonic content stamp: a new ingest is a new
// generation, and old generations are GC'd, so gen uniquely identifies the
// live bytes. Use it as a strong ETag.
etag := fmt.Sprintf("%q", fmt.Sprintf("%s-gen%d", project, gen))
w.Header().Set("ETag", etag)
w.Header().Set("Content-Type", "application/gzip")
w.Header().Set("Content-Disposition",
fmt.Sprintf("attachment; filename=%q", project+".tar.gz"))
if modified > 0 {
w.Header().Set("Last-Modified", time.Unix(modified, 0).UTC().Format(http.TimeFormat))
}
// Must revalidate every time: flakes pin by narHash, and a mutable
// /<project>.tar.gz can change contents when a new generation is ingested.
w.Header().Set("Cache-Control", "no-cache")
if match := r.Header.Get("If-None-Match"); match != "" && etagMatch(match, etag) {
w.WriteHeader(http.StatusNotModified)
return
}
// Stream the blob straight from SQLite so a large tarball is never held
// wholly in memory. Once the copy starts the status/headers are committed,
// so mid-stream errors can only be logged.
if err := s.streamBlob(w, "tarball", "data", rowid, size); err != nil {
fmt.Fprintf(stderr, "serveTarball(%q gen %d): %v\n", project, gen, err)
}
}
// etagMatch reports whether the ETag `etag` satisfies an If-None-Match header
// value, which is either "*" or a comma-separated list of (possibly weak)
// entity tags. A weak "W/" prefix is ignored for the comparison, which is
// correct for a conditional GET.
func etagMatch(ifNoneMatch, etag string) bool {
strip := func(s string) string { return strings.TrimPrefix(strings.TrimSpace(s), "W/") }
want := strip(etag)
for _, part := range strings.Split(ifNoneMatch, ",") {
got := strip(part)
if got == "*" || got == want {
return true
}
}
return false
}
// streamBlob copies the BLOB in <table>.<column> at the given rowid to w,
// reading it incrementally so large values are never fully buffered in memory.
// It opens a short-lived read-only zombiezen connection per call (the
// database/sql driver in use cannot expose SQLite's incremental blob API): a
// dedicated connection means a slow client draining a big blob never blocks
// other readers, at the cost of one cheap WAL open per download.
//
// Content-Length is set from size before the first byte. The caller must set
// any other headers (Content-Type, Content-Disposition, …) beforehand and must
// not have written a body yet. If an error is returned after copying began, the
// response is already partially written and can only be logged.
func (s *server) streamBlob(w http.ResponseWriter, table, column string, rowid, size int64) error {
conn, err := zsqlite.OpenConn(s.dbPath, zsqlite.OpenReadOnly, zsqlite.OpenWAL)
if err != nil {
return fmt.Errorf("open blob conn: %w", err)
}
defer conn.Close()
blob, err := conn.OpenBlob("main", table, column, rowid, false)
if err != nil {
return fmt.Errorf("open blob %s.%s rowid %d: %w", table, column, rowid, err)
}
defer blob.Close()
w.Header().Set("Content-Length", fmt.Sprintf("%d", size))
if _, err := io.Copy(w, blob); err != nil {
return fmt.Errorf("stream blob %s.%s rowid %d: %w", table, column, rowid, err)
}
return nil
}
// handleRobots serves /robots.txt. Everything is crawlable except the noisy
// surfaces: the recursive ?full= views (which duplicate a whole subtree on one
// page) and the binary .bundle / .tar.gz downloads. It advertises the sitemap
// so crawlers can discover every canonical page.
func (s *server) handleRobots(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "text/plain; charset=utf-8")
fmt.Fprintf(w,
"User-agent: *\n"+
"Disallow: /*?full\n"+
"Disallow: /*.bundle$\n"+
"Disallow: /*.tar.gz$\n"+
"\n"+
"Sitemap: %s/sitemap.xml\n",
s.baseFor(r))
}
// sitemapMaxURLs is the sitemap.org per-file cap (50,000 URLs / 50 MiB). We emit
// a single sitemap and stop at the URL limit; well within the size limit for
// any realistic tree.
const sitemapMaxURLs = 50000
// handleSitemap streams an XML sitemap of every canonical page: the site index,
// each project's directories and files, at their live generation. The noisy
// surfaces excluded in robots.txt (?full=, .bundle) are simply never emitted.
// <lastmod> comes from the git-derived mtime (a file's last-change time; a
// directory's newest descendant), omitted when unknown (0).
func (s *server) handleSitemap(w http.ResponseWriter, r *http.Request) {
base := s.baseFor(r)
// One read-only snapshot for the whole sitemap (see dbReader): the project
// list and each project's file walk below must agree on the same
// generation, or an ingest mid-crawl would drop a project's entries.
q, err := s.db.BeginTx(r.Context(), &sql.TxOptions{ReadOnly: true})
if err != nil {
httpError(w, err)
return
}
defer q.Rollback()
// Fallible work first: gather projects (name + live generation + root time)
// before any bytes are written, so a query error can still 500 cleanly.
// head_generation = 0 means declared but never pushed (see `project add` in
// project.go): there is no tree to walk, so such projects are excluded
// here rather than emitting a lone, contentless project-root URL.
type proj struct {
name string
gen int64
mtime int64
}
rows, err := q.Query(
`SELECT name, head_generation, root_mtime FROM project
WHERE head_generation > 0 ORDER BY name`)
if err != nil {
httpError(w, err)
return
}
var projs []proj
for rows.Next() {
var p proj
if err := rows.Scan(&p.name, &p.gen, &p.mtime); err != nil {
rows.Close()
httpError(w, err)
return
}
projs = append(projs, p)
}
rows.Close()
w.Header().Set("Content-Type", "application/xml; charset=utf-8")
io.WriteString(w, `<?xml version="1.0" encoding="UTF-8"?>`+"\n")
io.WriteString(w, `<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">`+"\n")
n := 0
emit := func(loc string, mtime int64) bool {
if n >= sitemapMaxURLs {
return false
}
writeSitemapURL(w, loc, mtime)
n++
return true
}
// The site index.
emit(base+"/", 0)
for _, p := range projs {
if n >= sitemapMaxURLs {
break
}
// Project root directory.
if !emit(base+"/"+sitemapPath(p.name)+"/", p.mtime) {
break
}
// Every directory and file in the live tree. Directories get a trailing
// slash to match how they are linked and served (serveFile redirects a
// dir path to the slash form).
fr, err := q.Query(
`SELECT path, is_dir, mtime FROM file
WHERE project = ? AND generation = ?
ORDER BY path`,
p.name, p.gen)
if err != nil {
// Header already committed; log and stop this project's entries.
fmt.Fprintf(stderr, "sitemap(%q): %v\n", p.name, err)
break
}
stop := false
for fr.Next() {
var path string
var isDir int
var mtime int64
if err := fr.Scan(&path, &isDir, &mtime); err != nil {
fmt.Fprintf(stderr, "sitemap(%q) scan: %v\n", p.name, err)
break
}
loc := base + "/" + sitemapPath(p.name) + "/" + sitemapPath(path)
if isDir == 1 {
loc += "/"
}
if !emit(loc, mtime) {
stop = true
break
}
}
fr.Close()
if stop {
break
}
}
if n >= sitemapMaxURLs {
fmt.Fprintf(stderr, "sitemap: hit %d-URL cap; some pages omitted\n", sitemapMaxURLs)
}
io.WriteString(w, "</urlset>\n")
}
// sitemapPath percent-encodes a slash-separated tree path for use in a URL,
// escaping each segment (so spaces, '#', '?', non-ASCII, etc. are RFC-3986
// safe) while keeping the '/' separators literal.
func sitemapPath(p string) string {
parts := strings.Split(p, "/")
for i, seg := range parts {
parts[i] = url.PathEscape(seg)
}
return strings.Join(parts, "/")
}
// writeSitemapURL writes one <url> entry, XML-escaping the (already
// percent-encoded) location and adding <lastmod> when mtime is known.
func writeSitemapURL(w io.Writer, loc string, mtime int64) {
io.WriteString(w, " <url><loc>")
xml.EscapeText(w, []byte(loc))
io.WriteString(w, "</loc>")
if mtime > 0 {
fmt.Fprintf(w, "<lastmod>%s</lastmod>",
time.Unix(mtime, 0).UTC().Format(time.RFC3339))
}
io.WriteString(w, "</url>\n")
}
// statsLabel renders recursive directory stats as "SIZE · N dirs, M files",
// omitting either count when it is zero (e.g. "SIZE · M files").
func statsLabel(dirs, files, size int64) string {
var parts []string
if dirs > 0 {
parts = append(parts, fmt.Sprintf("%d dirs", dirs))
}
if files > 0 {
parts = append(parts, fmt.Sprintf("%d files", files))
}
if len(parts) == 0 {
return humanSize(size)
}
return humanSize(size) + " · " + strings.Join(parts, ", ")
}
// serveProjectIndex lists all published projects, including any that have
// been declared (via `project add`) but never pushed to yet — recognisable by
// head_generation = 0, since generations start at 1 (see nextGeneration in
// ingest.go). These are shown with a "pending" marker instead of a
// misleading "0 files", so a botched or forgotten push is obvious from the
// index rather than looking identical to an empty project.
func (s *server) serveProjectIndex(q dbReader, w http.ResponseWriter, r *http.Request) {
// Collect all rows first (fallible) before writing any output, so a query
// error can still produce a clean 500 instead of a half-written page.
type projRow struct {
name, branch, description string
headGen int64
rootDirs, rootFiles, rootCodeSize int64
}
rows, err := q.Query(
`SELECT name, branch, head_generation, root_dirs, root_files, root_code_size, description
FROM project ORDER BY name`)
if err != nil {
httpError(w, err)
return
}
defer rows.Close()
var projs []projRow
for rows.Next() {
var p projRow
if err := rows.Scan(&p.name, &p.branch, &p.headGen, &p.rootDirs, &p.rootFiles, &p.rootCodeSize, &p.description); err != nil {
httpError(w, err)
return
}
projs = append(projs, p)
}
s.writePageHeader(w, r, "Projects", nil, false)
// The list is named by the page's own heading rather than by a duplicated
// aria-label, so the two can never drift apart.
io.WriteString(w,
"<h1 id=\"page-title\">Projects</h1>\n"+
"<ul class=\"listing\" aria-labelledby=\"page-title\">\n")
for _, p := range projs {
label := statsLabel(p.rootDirs, p.rootFiles, p.rootCodeSize)
if p.headGen == 0 {
label = "pending — not yet pushed"
}
fmt.Fprintf(w,
"<li class=\"dir\"><a href=\"/%s/\">%s/</a> "+
"<span class=\"muted\">(%s) — %s</span>",
html.EscapeString(p.name), html.EscapeString(p.name),
html.EscapeString(p.branch), html.EscapeString(label))
// The description is stored and emitted UNESCAPED — see the
// description column's comment in schema.go for why that is an
// accepted trust boundary here.
if p.description != "" {
fmt.Fprintf(w, "<div class=\"proj-desc\">%s</div>", p.description)
}
io.WriteString(w, "</li>\n")
}
if len(projs) == 0 {
io.WriteString(w, "<li class=\"muted\">no projects declared yet</li>\n")
}
io.WriteString(w, "</ul>\n")
s.writePageFooter(w)
}
// dirEntry is one row in a directory listing: the file or directory actually
// displayed, which need not be a direct child of the directory being browsed.
//
// A chain of single-child directories is shown as one entry pointing at the
// deepest non-branching path (so "users" links straight to "users/Profpatsch"),
// and a directory's .source-forge may add `shortcut` entries pointing further
// down as well. All of that is resolved at ingest into the listing table, so
// the fields below already describe exactly what to show.
type dirEntry struct {
// Full tree path of the displayed entry, e.g. "users/Profpatsch".
path string
isDir bool
size int64
// Last-change time: a file's own, a directory's newest descendant.
mtime int64
// Recursive aggregates (only meaningful when the entry is a directory).
subFiles int64
subDirs int64
subCodeSize int64
// Rendered HTML blurb from the directory's .source-forge, "" when none.
description string
}
// dirSize fetches a directory's total (recursive) byte size, which gates the
// expensive ?full= render. This is the raw size including binaries, not the
// code-only figure shown in listings: it measures how much there would be to
// render, not how much of it is source.
func (s *server) dirSize(q dbReader, project string, gen int64, dir string) (size int64) {
_ = q.QueryRow(
`SELECT subtree_size FROM file
WHERE project = ? AND path = ? AND generation = ?`,
project, dir, gen,
).Scan(&size)
return
}
// maxDirManpages caps how many of a directory's manpages are rendered below
// its listing. Each one costs a mandoc subprocess on a cache miss, so a
// directory holding a whole man/ tree would otherwise fan a single request out
// into dozens of them — the same bounded-work concern as fullMaxSize. Pages
// past the cap are still listed, and still browseable as ordinary files; only
// the rendered prose below the listing stops.
const maxDirManpages = 8
// sortManpages orders a directory's manpages for rendering below its listing,
// in place. The page a directory is "about" comes first: one whose basename
// matches the directory (or the project, at the root), so e.g. timetrack.1
// precedes sfttime.7 under timetrack/. Ties fall back to the lower section
// number, then the lower path, so the order is total and stable regardless of
// what order the listing produced the candidates in.
func sortManpages(pages []string, project, dir string) {
// The name this directory would ideally have a manpage for.
want := project
if dir != "" {
want = path.Base(dir)
}
nameMatches := func(p string) bool {
base := path.Base(p)
return base[:len(base)-len(path.Ext(base))] == want
}
sort.Slice(pages, func(i, j int) bool {
a, b := pages[i], pages[j]
if ma, mb := nameMatches(a), nameMatches(b); ma != mb {
return ma
}
// Same match status: prefer the lower section, then the lower path.
if se := strings.Compare(path.Ext(a), path.Ext(b)); se != 0 {
return se < 0
}
return a < b
})
}
// serveDir renders a directory listing, plus any README rendered above it.
func (s *server) serveDir(q dbReader, w http.ResponseWriter, r *http.Request, project, branch string, gen int64, dir string) {
// gen == 0 means the project has been declared (see `project add` in
// project.go) but never pushed to: there is no generation to have any file
// rows at all, so every path within it is "not found" except the project
// root itself, which gets a short pending notice instead of the ordinary
// (and here misleadingly empty) listing.
if gen == 0 {
if dir != "" {
http.NotFound(w, r)
return
}
description := s.projectDescription(q, project)
s.writePageHeader(w, r, project, nil, true)
io.WriteString(w, breadcrumb(project, "", ""))
fmt.Fprintf(w, "<h1 class=\"sr-only\">%s</h1>\n", html.EscapeString(project))
if description != "" {
io.WriteString(w, description)
}
fmt.Fprintf(w,
"<p class=\"muted\">Declared, branch <code>%s</code> — "+
"pending, nothing has been pushed yet.</p>\n",
html.EscapeString(branch))
s.writePageFooter(w)
return
}
// Verify the directory exists (root dir "" always exists implicitly), and
// take its total size and its own description in the same lookup: the
// "full contents" link below needs the size and the header below shows the
// description, so this saves two further queries.
//
// At the root, the project's description plays that part instead: the root
// has no parent listing to be described in, so its .source-forge blurb is
// stored on the project row (see storeDescriptions in ingest.go).
var totalSize int64
var description string
if dir != "" {
var isDir int
err := q.QueryRow(
`SELECT is_dir, subtree_size, description FROM file
WHERE project = ? AND path = ? AND generation = ?`,
project, dir, gen,
).Scan(&isDir, &totalSize, &description)
if err == sql.ErrNoRows || (err == nil && isDir == 0) {
http.NotFound(w, r)
return
}
if err != nil {
httpError(w, err)
return
}
} else {
_, _, totalSize = s.projectStats(q, project)
description = s.projectDescription(q, project)
}
// The listing is stored, already ordered and already resolved: collapsed
// single-child chains and any `shortcut` entries a directory declared in
// its .source-forge were all worked out at ingest (see buildListing). Each
// row names the file to display, so this is a plain join on the primary
// key with no ordering or path arithmetic left to do at request time.
rows, err := q.Query(
`SELECT c.path, c.is_dir, c.size, c.mtime,
c.subtree_files, c.subtree_dirs, c.subtree_code_size, c.description
FROM listing l
JOIN file c
ON c.project = l.project
AND c.generation = l.generation
AND c.path = l.target
WHERE l.project = ? AND l.generation = ? AND l.dir = ?
ORDER BY l.seq`,
project, gen, dir,
)
if err != nil {
httpError(w, err)
return
}
defer rows.Close()
var entries []dirEntry
var readme string // path of a README.md in this dir, if any
var manpages []string // paths of this dir's manpages, rendered after it
for rows.Next() {
var e dirEntry
var isDir int
if err := rows.Scan(&e.path, &isDir, &e.size, &e.mtime,
&e.subFiles, &e.subDirs, &e.subCodeSize, &e.description); err != nil {
httpError(w, err)
return
}
e.isDir = isDir == 1
entries = append(entries, e)
// The README / manpages rendered below the listing must be THIS
// directory's own files: an entry pointing deeper (a collapsed chain or
// a shortcut) belongs to some other directory and is skipped. A file
// that is a direct child always appears as itself, so none is missed.
parent := path.Dir(e.path)
if parent == "." {
parent = ""
}
if e.isDir || parent != dir {
continue
}
base := path.Base(e.path)
if strings.EqualFold(base, "README.md") {
readme = e.path
}
if isManpage(base) {
manpages = append(manpages, e.path)
}
}
sortManpages(manpages, project, dir)
// Breadcrumb + title.
title := project
if dir != "" {
title = project + "/" + dir
}
// All fallible queries above are done; begin streaming the page.
s.writePageHeader(w, r, title, nil, false)
io.WriteString(w, breadcrumb(project, dir, ""))
// The heading is visually hidden (the breadcrumb above plays its part on
// screen) but names the listing below it, via aria-labelledby rather than
// a duplicated aria-label so the two can never drift apart.
fmt.Fprintf(w, "<h1 class=\"sr-only\" id=\"page-title\">%s</h1>\n", html.EscapeString(title))
// The directory's description (at the root, the project's) sits between
// the header and the listing — above the clone/tarball instructions, where
// those appear. Fetched above; stored and emitted UNESCAPED — see the
// description column's comment in schema.go for why that is an accepted
// trust boundary here.
//
// A directory's is inline HTML, since its main home is inside a listing
// entry's line, so it is wrapped in a paragraph to stand on its own here.
// The project's (dir == "") is stored as block content already.
if description != "" {
if dir != "" {
fmt.Fprintf(w, "<p class=\"dir-desc\">%s</p>\n", description)
} else {
io.WriteString(w, description)
}
}
// At the project root, offer the clonable git bundle if one exists.
// A git bundle is a single file holding the full history; git cannot clone
// it over HTTP directly (that triggers the smart-HTTP protocol), so the
// instructions download it first, then clone/fetch from the local file.
if dir == "" {
if bundleSize, ok := s.hasBundle(q, project, gen); ok {
bundleURL := s.bundleURL(r, project)
fmt.Fprintf(w,
"<div class=\"clone\">Clone the full history "+
"(<a href=\"/%s.bundle\">%s.bundle</a>, a git bundle, %s):"+
"<pre class=\"chroma\">curl -O %s\ngit clone %s.bundle %s</pre>"+
"<p class=\"muted\">The clone keeps the bundle as its "+
"<code>origin</code>; replace the file and "+
"<code>git fetch</code> to update.</p></div>\n",
html.EscapeString(project), html.EscapeString(project),
html.EscapeString(humanSize(bundleSize)),
html.EscapeString(bundleURL),
html.EscapeString(project), html.EscapeString(project))
}
// Offer the Nix-flake tarball if one exists. Nix's tarball input scheme
// fetches this static archive directly — no git server needed — so a
// package can be run straight from the URL.
if tarballSize, ok := s.hasTarball(q, project, gen); ok {
tarballURL := s.tarballURL(r, project)
fmt.Fprintf(w,
"<div class=\"clone\">Run a package with Nix "+
"(<a href=\"/%s.tar.gz\">%s.tar.gz</a>, a source tarball, %s):"+
"<pre class=\"chroma\">nix run %s#<package></pre>"+
"<p class=\"muted\">The archive is a Nix flake; "+
"<code>nix flake show %s</code> lists its packages.</p></div>\n",
html.EscapeString(project), html.EscapeString(project),
html.EscapeString(humanSize(tarballSize)),
html.EscapeString(tarballURL),
html.EscapeString(tarballURL))
}
}
// Listing, named by the page heading emitted above.
io.WriteString(w, "<ul class=\"listing\" aria-labelledby=\"page-title\">\n")
for _, e := range entries {
// The label is the entry's path relative to the directory being
// browsed, so a collapsed chain or a shortcut reads as the path it
// jumps to (e.g. "Profpatsch/git-blimey/") rather than just its
// basename.
label := strings.TrimPrefix(e.path, dir+"/")
if dir == "" {
label = e.path
}
if e.isDir {
fmt.Fprintf(w,
"<li class=\"dir\"><a href=\"/%s/%s/\">%s/</a> "+
"<span class=\"muted\">%s%s</span>%s</li>\n",
html.EscapeString(project), html.EscapeString(e.path),
html.EscapeString(label),
html.EscapeString(statsLabel(e.subDirs, e.subFiles, e.subCodeSize)),
// The directory's own blurb, on the same line and in the same
// muted style as the stats it follows. Inline HTML, stored and
// emitted UNESCAPED — see the description column's comment in
// schema.go for why that is an accepted trust boundary here.
descSuffix(e.description),
agoTime(e.mtime))
} else {
fmt.Fprintf(w,
"<li class=\"file\"><a href=\"/%s/%s\">%s</a> "+
"<span class=\"muted\">%s</span>%s</li>\n",
html.EscapeString(project), html.EscapeString(e.path),
html.EscapeString(label), humanSize(e.size), agoTime(e.mtime))
}
}
io.WriteString(w, "</ul>\n")
// Offer the recursive "full contents" view only for subtrees within the
// size limit, so we never advertise a page that would render a huge tree.
// The handler enforces the same limit for direct-URL requests.
if totalSize > 0 && totalSize <= fullMaxSize {
io.WriteString(w,
"<p class=\"full-link\"><a href=\"?full=\">View full contents ↓</a></p>\n")
}
s.writeProse(q, w, project, gen, readme, manpages)
s.writePageFooter(w)
}
// writeProse renders a directory's prose below (or, in the ?full= view, above)
// its listing: the README first when there is one, then every manpage the
// directory holds, each under a heading naming it. Both go through the render
// cache (markdown prose / mandoc prose respectively).
//
// A manpage is rendered whether or not a README is present: the two describe
// different things — a README introduces the directory, a manpage specifies the
// program — and a project that has both had, until now, its manpage hidden
// entirely by the README.
//
// Rendering is capped at maxDirManpages; see it for why. Manpages must already
// be ordered (sortManpages).
func (s *server) writeProse(q dbReader, w io.Writer, project string, gen int64, readme string, manpages []string) {
if readme != "" {
if h, ok := s.markdownFragment(q, project, gen, readme); ok {
io.WriteString(w, h)
}
}
if len(manpages) > maxDirManpages {
manpages = manpages[:maxDirManpages]
}
for _, p := range manpages {
if h, ok := s.manpageFragment(q, project, gen, p); ok {
io.WriteString(w, h)
}
}
}
// fullMaxSize is the maximum total subtree size for which the recursive "full
// contents" view is offered and rendered. Directories above this limit hide the
// link AND refuse the ?full= page (a small "too large" notice is shown instead),
// so a scraper appending ?full= to every directory cannot trigger huge renders.
const fullMaxSize = 1 << 20 // 1 MiB
// projectStats returns the whole-tree recursive stats stored on the project row.
func (s *server) projectStats(q dbReader, project string) (dirs, files, size int64) {
_ = q.QueryRow(
`SELECT root_dirs, root_files, root_size FROM project WHERE name = ?`,
project,
).Scan(&dirs, &files, &size)
return
}
// projectDescription returns a project's HTML blurb, declared by `description`
// in the root .source-forge and applied at ingest (see storeDescriptions in
// ingest.go and the description column's comment in schema.go), or "" when
// none is set. Callers emit it UNESCAPED — this function does no
// sanitisation, deliberately.
func (s *server) projectDescription(q dbReader, project string) string {
var description string
_ = q.QueryRow(
`SELECT description FROM project WHERE name = ?`, project,
).Scan(&description)
return description
}
// serveFull renders the contents of every file under a directory (recursively)
// on a single page: text/source highlighted, markdown as prose, images inline,
// and other binaries as a download stub. The header shows the summed size.
//
// The subtree must be within fullMaxSize; larger directories get a small notice
// instead, so the expensive recursive render is never triggered by a crawler.
func (s *server) serveFull(q dbReader, w http.ResponseWriter, r *http.Request, project string, gen int64, dir string) {
// Confirm the directory exists (root "" is always valid).
if dir != "" {
var isDir int
err := q.QueryRow(
`SELECT is_dir FROM file WHERE project = ? AND path = ? AND generation = ?`,
project, dir, gen,
).Scan(&isDir)
if err == sql.ErrNoRows || (err == nil && isDir == 0) {
http.NotFound(w, r)
return
}
if err != nil {
httpError(w, err)
return
}
}
title := project
if dir != "" {
title = project + "/" + dir
}
// Size guard: read the precomputed subtree size (one indexed lookup) and
// refuse to render trees over the limit. This happens BEFORE the recursive
// scan below, so an over-limit ?full= costs only this cheap check.
var totalSize int64
if dir == "" {
_, _, totalSize = s.projectStats(q, project)
} else {
totalSize = s.dirSize(q, project, gen, dir)
}
if totalSize > fullMaxSize {
s.writePageHeader(w, r, title, nil, true)
io.WriteString(w, breadcrumb(project, dir, ""))
fmt.Fprintf(w, "<h1 class=\"sr-only\">%s — full contents</h1>\n",
html.EscapeString(title))
fmt.Fprintf(w,
"<p class=\"muted\">This directory is %s, larger than the %s limit "+
"for the single-page full view. <a href=\".\">← browse the "+
"listing</a> instead.</p>\n",
html.EscapeString(humanSize(totalSize)),
html.EscapeString(humanSize(fullMaxSize)))
s.writePageFooter(w)
return
}
// Metadata pass: collect the file list (path/size/type only — no content),
// so the header can show an accurate count. This is fallible, so it runs
// before any output is written.
var rows *sql.Rows
var err error
if dir == "" {
rows, err = q.Query(
`SELECT path, size, is_binary, mime_type FROM file
WHERE project = ? AND generation = ? AND is_dir = 0
ORDER BY path`,
project, gen,
)
} else {
rows, err = q.Query(
`SELECT path, size, is_binary, mime_type FROM file
WHERE project = ? AND generation = ? AND is_dir = 0
AND (path = ? OR path LIKE ? ESCAPE '\')
ORDER BY path`,
project, gen, dir, likePrefix(dir)+"/%",
)
}
if err != nil {
httpError(w, err)
return
}
type fullFile struct {
path string
size int64
isBin bool
mime string
}
var files []fullFile
for rows.Next() {
var f fullFile
var isBin int
if err := rows.Scan(&f.path, &f.size, &isBin, &f.mime); err != nil {
rows.Close()
httpError(w, err)
return
}
f.isBin = isBin == 1
files = append(files, f)
}
rows.Close()
// Collect the view-root's own README / manpages (direct children of dir) to
// render at the top, mirroring the directory listing. Derived from the
// already-collected file list, so no extra query. The promoted README is
// also shown verbatim (highlighted source) at its lexicographic slot below;
// so are the manpages, which are not markdown and so already render as
// source there.
var readme string
var manpages []string
for _, f := range files {
parent := path.Dir(f.path)
if parent == "." {
parent = ""
}
if parent != dir {
continue
}
base := path.Base(f.path)
if strings.EqualFold(base, "README.md") {
readme = f.path
}
if isManpage(base) {
manpages = append(manpages, f.path)
}
}
sortManpages(manpages, project, dir)
// Stream the page. Each file's rendered fragment is fetched and written one
// at a time, so we never hold the whole assembled page in memory.
flusher, _ := w.(http.Flusher)
s.writePageHeader(w, r, title, nil, true)
io.WriteString(w, breadcrumb(project, dir, ""))
fmt.Fprintf(w, "<h1 class=\"sr-only\">%s — full contents</h1>\n",
html.EscapeString(title))
fmt.Fprintf(w,
"<p class=\"muted\">%d files · %s total. "+
"<a href=\".\">← back to listing</a></p>\n",
len(files), html.EscapeString(humanSize(totalSize)))
// Rendered README and manpages promoted above the file dump, exactly as in
// the directory listing.
s.writeProse(q, w, project, gen, readme, manpages)
if flusher != nil {
flusher.Flush()
}
for _, f := range files {
anchor := "f-" + f.path
fmt.Fprintf(w,
"<section class=\"full-file\" id=\"%s\">\n"+
"<h2 class=\"full-path\"><a href=\"/%s/%s\">%s</a> "+
"<span class=\"muted\">%s</span></h2>\n",
html.EscapeString(anchor),
html.EscapeString(project), html.EscapeString(f.path),
html.EscapeString(f.path), html.EscapeString(humanSize(f.size)))
switch {
case f.isBin && strings.HasPrefix(f.mime, "image/"):
// Inline the image via the existing raw file route.
fmt.Fprintf(w,
"<img class=\"full-image\" src=\"/%s/%s\" alt=\"%s\">\n",
html.EscapeString(project), html.EscapeString(f.path),
html.EscapeString(f.path))
case f.isBin:
// Non-image binary: a download stub, no inline content.
fmt.Fprintf(w,
"<p class=\"muted\">binary file (%s) — "+
"<a href=\"/%s/%s\">download</a></p>\n",
html.EscapeString(f.mime),
html.EscapeString(project), html.EscapeString(f.path))
default:
// The promoted README is already rendered as prose at the top, so
// here it appears verbatim (highlighted source) at its lexicographic
// slot; every other text file gets its canonical display rendering
// (nested markdown stays prose).
var frag string
var ok bool
if f.path == readme {
frag, ok = s.highlightFragment(q, project, gen, f.path)
} else {
frag, ok = s.displayFragment(q, project, gen, f.path)
}
if ok {
io.WriteString(w, frag)
io.WriteString(w, "\n")
}
}
io.WriteString(w, "</section>\n")
if flusher != nil {
flusher.Flush()
}
}
s.writePageFooter(w)
}
// likePrefix escapes SQLite LIKE metacharacters (% _ \) in a literal path so it
// can be used safely as a prefix with `ESCAPE '\'`.
func likePrefix(p string) string {
r := strings.NewReplacer(`\`, `\\`, `%`, `\%`, `_`, `\_`)
return r.Replace(p)
}
// serveFile renders a single file: raw bytes for binaries, cached highlighted
// HTML (or markdown) for text.
func (s *server) serveFile(q dbReader, w http.ResponseWriter, r *http.Request, project string, gen int64, p string) {
var isDir, isBin int
var mimeType string
var size, rowid int64
err := q.QueryRow(
`SELECT is_dir, is_binary, mime_type, size, rowid FROM file
WHERE project = ? AND path = ? AND generation = ?`,
project, p, gen,
).Scan(&isDir, &isBin, &mimeType, &size, &rowid)
if err == sql.ErrNoRows {
http.NotFound(w, r)
return
}
if err != nil {
httpError(w, err)
return
}
if isDir == 1 {
// Redirect to the trailing-slash directory URL.
http.Redirect(w, r, "/"+project+"/"+p+"/", http.StatusFound)
return
}
// Binary: serve raw with the MIME detected at ingest, streamed straight from
// the DB so large files (images, PDFs) are not buffered in memory. The
// full-contents view inlines every image through this route, so this is the
// hot path for binary payloads.
if isBin == 1 {
w.Header().Set("Content-Type", mimeType)
w.Header().Set("Cache-Control", "public, max-age=3600")
if err := s.streamBlob(w, "file", "content", rowid, size); err != nil {
fmt.Fprintf(stderr, "serveFile(%q/%q): %v\n", project, p, err)
}
return
}
// Text: render (via cache) and serve.
htmlFrag, ok := s.displayFragment(q, project, gen, p)
if !ok {
http.NotFound(w, r)
return
}
s.writePage(w, r, project+"/"+p, &fileMeta{project: project, path: p, size: size}, htmlFrag)
}
// displayFragment returns the canonical "file view" rendering of a text file:
// markdown as prose, everything else chroma-highlighted. This is what a direct
// file request and the non-promoted entries of the ?full= view show.
func (s *server) displayFragment(q dbReader, project string, gen int64, p string) (string, bool) {
if isMarkdown(p) {
return s.markdownFragment(q, project, gen, p)
}
return s.highlightFragment(q, project, gen, p)
}
// highlightFragment returns a file's raw bytes chroma-highlighted (kind
// 'highlight'), using the render cache and populating it on a miss. This is the
// verbatim source rendering, used both for non-markdown file views and for the
// promoted README's in-place slot in the ?full= view.
func (s *server) highlightFragment(q dbReader, project string, gen int64, p string) (string, bool) {
if htmlFrag, ok := s.cachedRender(q, project, gen, p, kindHighlight); ok {
return htmlFrag, true
}
content, ok := s.fileContent(q, project, gen, p)
if !ok {
return "", false
}
htmlFrag := ""
if h, err := highlightSource(p, content); err == nil {
htmlFrag = h
} else {
htmlFrag = plainFallback(content)
}
s.storeRender(project, gen, p, kindHighlight, htmlFrag)
return htmlFrag, true
}
// markdownFragment returns a markdown file rendered as prose (kind 'markdown'),
// wrapped in <section class="readme"> and cached. Used for the file view of a
// .md and for a directory's promoted README.
func (s *server) markdownFragment(q dbReader, project string, gen int64, p string) (string, bool) {
if htmlFrag, ok := s.cachedRender(q, project, gen, p, kindMarkdown); ok {
return htmlFrag, true
}
content, ok := s.fileContent(q, project, gen, p)
if !ok {
return "", false
}
htmlFrag := ""
if h, err := renderMarkdown(content, project, p); err == nil {
htmlFrag = "<section class=\"readme\">\n" + h + "</section>\n"
} else {
htmlFrag = plainFallback(content)
}
s.storeRender(project, gen, p, kindMarkdown, htmlFrag)
return htmlFrag, true
}
// manpageFragment returns the rendered manpage HTML fragment for a .N file,
// using the render cache (kind 'manpage', distinct from the file's own
// highlighted source under kind 'highlight') when available and populating it
// on a miss. The fragment is wrapped in <section class="readme"> so it reuses
// the README styling, mirroring how markdown READMEs are rendered.
//
// The title heading (manpageTitleHTML) goes INSIDE that section, and so into
// the cached bytes: the section carries the border-top separating one prose
// block from the next, and a title emitted before it would be cut off from its
// own prose by that border, appearing to belong to whatever came above.
func (s *server) manpageFragment(q dbReader, project string, gen int64, p string) (string, bool) {
if htmlFrag, ok := s.cachedRender(q, project, gen, p, kindManpage); ok {
return htmlFrag, true
}
content, ok := s.fileContent(q, project, gen, p)
if !ok {
return "", false
}
h, err := renderManpage(content, project, p)
if err != nil {
fmt.Fprintf(stderr, "renderManpage(%q): %v\n", p, err)
return "", false
}
htmlFrag := "<section class=\"readme\">\n" + manpageTitleHTML(project, p) + h + "</section>\n"
s.storeRender(project, gen, p, kindManpage, htmlFrag)
return htmlFrag, true
}
// fileContent fetches a file's raw content.
func (s *server) fileContent(q dbReader, project string, gen int64, p string) ([]byte, bool) {
var content []byte
err := q.QueryRow(
`SELECT content FROM file WHERE project = ? AND path = ? AND generation = ?`,
project, p, gen,
).Scan(&content)
if err != nil {
if err != sql.ErrNoRows {
fmt.Fprintf(stderr, "fileContent(%q): %v\n", p, err)
}
return nil, false
}
return content, true
}
// renderKind names one way a file's bytes are rendered into cached HTML, one
// value per distinct HTML output. It is stored verbatim in render_cache.kind
// (see the `kind` column), so a file can hold several cached renderings at once
// without colliding: e.g. a README.md is cached both as markdown prose (for the
// file view and the directory README) and as highlighted source (for its
// verbatim slot in the ?full= view). Append new kinds freely; never rename an
// existing value (old rows would orphan — harmless, they are re-rendered).
type renderKind string
const (
kindHighlight renderKind = "highlight" // chroma-highlighted raw bytes
kindMarkdown renderKind = "markdown" // goldmark prose, section.readme-wrapped
kindManpage renderKind = "manpage" // mandoc prose, section.readme-wrapped
)
// cachedRender returns the cached HTML fragment for a file at a generation and
// render kind.
func (s *server) cachedRender(q dbReader, project string, gen int64, p string, kind renderKind) (string, bool) {
var htmlFrag string
err := q.QueryRow(
`SELECT html FROM render_cache
WHERE project = ? AND path = ? AND generation = ? AND kind = ?`,
project, p, gen, string(kind),
).Scan(&htmlFrag)
if err != nil {
return "", false
}
return htmlFrag, true
}
// storeRender caches a rendered HTML fragment under a render kind. Errors are
// non-fatal (we just re-render next time).
//
// This deliberately writes through the pool rather than the request's read
// snapshot: that transaction is read-only, and the cache fill should commit on
// its own so later requests see it immediately rather than being tied to the
// lifetime of whichever page happened to render the file first.
//
// The write goes to the dedicated single-connection writer, so a write waiting
// out an ingest can never consume connections from the read pool, and our own
// writes queue behind each other instead of contending. Waiting for the lock is
// left entirely to that connection's busy_timeout, which is long enough to
// outlast an ingest; there is no retry loop on top, as that would just be a
// second, worse timeout.
//
// Failure is not silent: a miss here means the next request has to re-run the
// whole render (chroma, or a mandoc subprocess), so it is worth complaining
// about rather than quietly degrading.
func (s *server) storeRender(project string, gen int64, p string, kind renderKind, htmlFrag string) {
_, err := s.writeDB.Exec(
`INSERT OR REPLACE INTO render_cache (project, path, generation, kind, html)
VALUES (?, ?, ?, ?, ?)`,
project, p, gen, string(kind), htmlFrag,
)
if err != nil {
fmt.Fprintf(stderr, "storeRender(%q): %v (file will be re-rendered on the next request)\n", p, err)
}
}
// buildCSS produces the combined stylesheet: the (unconditional) page layout
// plus one chroma block per theme, each guarded by a prefers-color-scheme
// media query.
//
// Both blocks must be guarded, and the two queries must be mutually exclusive.
// Chroma only emits a rule for the tokens a style actually overrides, so the
// two themes do not produce the same set of rules: github-dark says nothing
// about NameOther, Punctuation, NameAttribute, NameBuiltin or NameBuiltinPseudo,
// for instance. Emitting the light block unscoped (relying on the dark block to
// shadow it at equal specificity) therefore left exactly those tokens at their
// light colour on a dark background — identifiers and punctuation at #1f2328 on
// #0d1117, a contrast ratio of 1.2:1, i.e. all but invisible.
//
// The light query is spelled "not all and (prefers-color-scheme: dark)" rather
// than "(prefers-color-scheme: light)" so that a browser which supports neither
// query still gets the light theme, instead of unhighlighted text.
func buildCSS() ([]byte, error) {
var buf bytes.Buffer
buf.WriteString(pageCSS)
// Same options as the formatter that produces the markup, or the
// stylesheet does not match it; see chromaFormatterOptions.
formatter := chromahtml.New(chromaFormatterOptions...)
writeTheme := func(name, query string) error {
style := styles.Get(name)
if style == nil {
return fmt.Errorf("unknown chroma style %q", name)
}
// Rendered to a scratch buffer first so a failure cannot leave a
// half-written, unbalanced media block in the served stylesheet.
var themeBuf bytes.Buffer
if err := formatter.WriteCSS(&themeBuf, style); err != nil {
return fmt.Errorf("write %s css: %w", name, err)
}
buf.WriteString("\n@media " + query + " {\n")
buf.Write(themeBuf.Bytes())
buf.WriteString("}\n")
return nil
}
if err := writeTheme(lightStyle, "not all and (prefers-color-scheme: dark)"); err != nil {
return nil, err
}
if err := writeTheme(darkStyle, "(prefers-color-scheme: dark)"); err != nil {
return nil, err
}
return buf.Bytes(), nil
}
// --- HTML page shell ---------------------------------------------------------
type fileMeta struct {
project string // project name, for building the breadcrumb links
path string // file path within the project (e.g. "dir/sub/file.nix")
size int64
}
// writePageHeader writes the HTML page shell up to the opening of <main> and
// sets the response headers. Handlers must perform all fallible work (queries
// that may 500) BEFORE calling this, since once the header is written the
// status code is committed and the body is streamed incrementally.
func (s *server) writePageHeader(w http.ResponseWriter, r *http.Request, title string, meta *fileMeta, noindex bool) {
w.Header().Set("Content-Type", "text/html; charset=utf-8")
w.Header().Set("Cache-Control", "no-cache")
io.WriteString(w, "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n")
io.WriteString(w, "<meta charset=\"utf-8\">\n")
io.WriteString(w, "<meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">\n")
// Canonical listing/file pages are indexable; the recursive ?full= views are
// kept out of the index (they duplicate every file on one page) but still
// followed so crawlers reach the canonical per-file pages.
if noindex {
io.WriteString(w, "<meta name=\"robots\" content=\"noindex, follow\">\n")
} else {
io.WriteString(w, "<meta name=\"robots\" content=\"index, follow\">\n")
}
// The host is appended so a page copy/pasted, screenshotted, or opened from
// browser history still shows which site (and, for anyone running their own
// instance, which instance) it came from.
fmt.Fprintf(w, "<title>%s — %s</title>\n", html.EscapeString(title), html.EscapeString(s.hostFor(r)))
// The stylesheet is inlined rather than linked so every page is fully
// self-contained (one request, no relative-URL pitfalls under nested paths).
io.WriteString(w, "<style>\n")
w.Write(s.css)
io.WriteString(w, "\n</style>\n")
io.WriteString(w, "</head>\n<body>\n<main>\n")
if meta != nil {
// The file page's navigable path: parent directories are linked and the
// filename is the current (aria-current) crumb.
dir, file := path.Split(meta.path)
io.WriteString(w, "<div class=\"filehead\">")
io.WriteString(w, breadcrumb(meta.project, strings.Trim(dir, "/"), file))
io.WriteString(w, "</div>\n")
}
}
// writePageFooter closes the streamed page shell.
func (s *server) writePageFooter(w http.ResponseWriter) {
io.WriteString(w, "</main>\n</body>\n</html>\n")
}
// writePage wraps a fully-built HTML body fragment in the page shell. Kept for
// the simpler pages; the body string is written in one shot.
func (s *server) writePage(w http.ResponseWriter, r *http.Request, title string, meta *fileMeta, body string) {
s.writePageHeader(w, r, title, meta, false)
io.WriteString(w, body)
s.writePageFooter(w)
}
// --- helpers -----------------------------------------------------------------
func httpError(w http.ResponseWriter, err error) {
fmt.Fprintf(stderr, "source-forge: %v\n", err)
http.Error(w, "internal error", http.StatusInternalServerError)
}
// breadcrumb builds an accessible breadcrumb navigation for a path within a
// project, following the WAI-ARIA breadcrumb pattern: a <nav aria-label> around
// an ordered list of crumbs. Every crumb except the last is a link; the final
// crumb (the current location) is a plain <span> marked aria-current="page".
// The "/" separators are emitted as real, aria-hidden text nodes: hidden from
// screen readers (which announce only the crumb labels) but part of the DOM
// text, so selecting the breadcrumb copies the full slash-separated path.
//
// leaf selects what the current (non-link) crumb is:
// - leaf == "": a directory view. The last segment of dir (or the project
// name when dir is empty) is the current crumb.
// - leaf != "": a file view. Every segment of dir is linked and leaf (the
// filename) is the current crumb.
func breadcrumb(project, dir, leaf string) string {
// crumb is one entry in the trail: label plus the href of its directory
// (empty href marks the current, non-link crumb).
type crumb struct{ label, href string }
var crumbs []crumb
crumbs = append(crumbs, crumb{project, "/" + project + "/"})
if dir != "" {
acc := ""
for _, part := range strings.Split(dir, "/") {
acc = pathJoin(acc, part)
crumbs = append(crumbs, crumb{part, "/" + project + "/" + acc + "/"})
}
}
if leaf != "" {
crumbs = append(crumbs, crumb{leaf, ""})
}
// The final crumb is always the current location: drop its link.
crumbs[len(crumbs)-1].href = ""
var b strings.Builder
b.WriteString("<nav class=\"breadcrumb\" aria-label=\"Breadcrumb\"><ol>")
for i, c := range crumbs {
b.WriteString("<li>")
if i > 0 {
// Real separator text so the copied selection keeps the slashes;
// aria-hidden keeps it out of the screen-reader announcement.
b.WriteString("<span class=\"sep\" aria-hidden=\"true\">/</span>")
}
if c.href == "" {
fmt.Fprintf(&b, "<span aria-current=\"page\">%s</span>",
html.EscapeString(c.label))
} else {
fmt.Fprintf(&b, "<a href=\"%s\">%s</a>",
html.EscapeString(c.href), html.EscapeString(c.label))
}
b.WriteString("</li>")
}
b.WriteString("</ol></nav>")
return b.String()
}
func pathJoin(dir, name string) string {
if dir == "" {
return name
}
return dir + "/" + name
}
func humanSize(n int64) string {
const unit = 1024
if n < unit {
return fmt.Sprintf("%d B", n)
}
div, exp := int64(unit), 0
for m := n / unit; m >= unit; m /= unit {
div *= unit
exp++
}
return fmt.Sprintf("%.1f %ciB", float64(n)/float64(div), "KMGTPE"[exp])
}
// agoTime renders a trailing " <time class="ago">…</time>" for a listing entry,
// or "" when the mtime is unknown (0) so old data degrades gracefully.
//
// <time> is the element for exactly this: a human-readable instant whose
// machine-readable form differs from its text. The datetime attribute carries
// the full instant, so the "3 days ago" text — which is only true at the moment
// the page is rendered — is anchored to something exact for assistive tech and
// anything else parsing the page. The title attribute keeps the absolute date
// visible on hover.
func agoTime(mtime int64) string {
if mtime <= 0 {
return ""
}
t := time.Unix(mtime, 0).UTC()
return fmt.Sprintf(" <time class=\"ago\" datetime=\"%s\" title=\"%s\">%s</time>",
html.EscapeString(t.Format(time.RFC3339)),
html.EscapeString(t.Format("2006-01-02")),
html.EscapeString(humanAgo(time.Since(t))))
}
// descSuffix renders a listing entry's description as a continuation of the
// muted stats that precede it ("143.1 KiB · 14 files — An interactive …"), or
// "" when there is none.
//
// The separator is a real text node rather than CSS ::before content so that
// selecting and copying a listing keeps it, and aria-hidden so a screen reader
// reads the stats and the description as two phrases instead of announcing a
// dash between them — the same treatment the breadcrumb gives its "/".
//
// The description is inline HTML (see meta.go) and is emitted UNESCAPED, which
// is what lets it carry <code> and links; see the description column's comment
// in schema.go for why that trust boundary is accepted.
func descSuffix(description string) string {
if description == "" {
return ""
}
return "<span class=\"sep\" aria-hidden=\"true\"> — </span>" + description
}
// humanAgo renders a duration as a coarse "N unit(s) ago" string, picking the
// largest whole unit (years down to seconds). Future or ~now times render as
// "just now" (clock skew shouldn't produce alarming negative labels).
func humanAgo(d time.Duration) string {
secs := int64(d.Seconds())
if secs < 60 {
return "just now"
}
type unit struct {
name string
secs int64
}
for _, u := range []unit{
{"year", 365 * 24 * 3600},
{"month", 30 * 24 * 3600},
{"week", 7 * 24 * 3600},
{"day", 24 * 3600},
{"hour", 3600},
{"minute", 60},
} {
if n := secs / u.secs; n >= 1 {
if n == 1 {
return "1 " + u.name + " ago"
}
return fmt.Sprintf("%d %ss ago", n, u.name)
}
}
return "just now"
}
|