| 1 | /* |
| 2 | * trie.c: implementation of trie.h. |
| 3 | */ |
| 4 | |
| 5 | #include <assert.h> |
| 6 | #include <stdio.h> |
| 7 | #include <string.h> |
| 8 | #include <stdlib.h> |
| 9 | #include <errno.h> |
| 10 | |
| 11 | #include <sys/types.h> |
| 12 | #include <unistd.h> |
| 13 | |
| 14 | #include "malloc.h" |
| 15 | #include "trie.h" |
| 16 | |
| 17 | #define alignof(typ) ( offsetof(struct { char c; typ t; }, t) ) |
| 18 | |
| 19 | /* |
| 20 | * Compare functions for pathnames. Returns the relative order of |
| 21 | * the names, like strcmp; also passes back the offset of the |
| 22 | * first differing character if desired. |
| 23 | */ |
| 24 | static int trieccmp(unsigned char a, unsigned char b) |
| 25 | { |
| 26 | a = (a == '\0' ? '\0' : a == '/' ? '\1' : a+1); |
| 27 | b = (b == '\0' ? '\0' : b == '/' ? '\1' : b+1); |
| 28 | return a - b; |
| 29 | } |
| 30 | |
| 31 | static int triencmp(const char *a, size_t alen, |
| 32 | const char *b, size_t blen, int *offset) |
| 33 | { |
| 34 | int off = 0; |
| 35 | while (off < alen && off < blen && a[off] == b[off]) |
| 36 | off++; |
| 37 | if (offset) |
| 38 | *offset = off; |
| 39 | if (off == alen || off == blen) return (off == blen) - (off == alen); |
| 40 | return trieccmp(a[off], b[off]); |
| 41 | } |
| 42 | |
| 43 | static int triecmp(const char *a, const char *b, int *offset) |
| 44 | { |
| 45 | return triencmp(a, strlen(a), b, strlen(b), offset); |
| 46 | } |
| 47 | |
| 48 | /* ---------------------------------------------------------------------- |
| 49 | * Trie node structures. |
| 50 | * |
| 51 | * The trie format stored in the file consists of three distinct |
| 52 | * node types, each with a distinguishing type field at the start. |
| 53 | * |
| 54 | * TRIE_LEAF is a leaf node; it contains an actual trie_file |
| 55 | * structure, and indicates that when you're searching down the |
| 56 | * trie with a string, you should now expect to encounter |
| 57 | * end-of-string. |
| 58 | * |
| 59 | * TRIE_SWITCH indicates that the set of strings in the trie |
| 60 | * include strings with more than one distinct character after the |
| 61 | * prefix leading up to this point. Hence, it stores multiple |
| 62 | * subnode pointers and a different single character for each one. |
| 63 | * |
| 64 | * TRIE_STRING indicates that at this point everything in the trie |
| 65 | * has the same next few characters; it stores a single mandatory |
| 66 | * string fragment and exactly one subnode pointer. |
| 67 | */ |
| 68 | enum { |
| 69 | TRIE_LEAF = 0x7fffe000, |
| 70 | TRIE_SWITCH, |
| 71 | TRIE_STRING |
| 72 | }; |
| 73 | |
| 74 | struct trie_common { |
| 75 | int type; |
| 76 | }; |
| 77 | |
| 78 | struct trie_switchentry { |
| 79 | off_t subnode; |
| 80 | int subcount; |
| 81 | }; |
| 82 | |
| 83 | struct trie_leaf { |
| 84 | struct trie_common c; |
| 85 | struct trie_file file; |
| 86 | }; |
| 87 | |
| 88 | struct trie_switch { |
| 89 | struct trie_common c; |
| 90 | /* |
| 91 | * sw[0] to sw[len-1] give the subnode pointers and element |
| 92 | * counts. At &sw[len] is stored len single bytes which are |
| 93 | * the characters corresponding to each subnode. |
| 94 | */ |
| 95 | int len; |
| 96 | struct trie_switchentry sw[]; |
| 97 | }; |
| 98 | |
| 99 | struct trie_string { |
| 100 | struct trie_common c; |
| 101 | int stringlen; |
| 102 | off_t subnode; |
| 103 | char string[]; |
| 104 | }; |
| 105 | |
| 106 | struct trie_header { |
| 107 | unsigned long magic; |
| 108 | off_t root, indexroot; |
| 109 | int count; |
| 110 | size_t maxpathlen; |
| 111 | }; |
| 112 | |
| 113 | /* Union only used for computing alignment */ |
| 114 | union trie_node { |
| 115 | struct trie_leaf leaf; |
| 116 | struct { /* fake trie_switch with indeterminate array length filled in */ |
| 117 | struct trie_common c; |
| 118 | int len; |
| 119 | struct trie_switchentry sw[1]; |
| 120 | } sw; |
| 121 | struct { /* fake trie_string with indeterminate array length filled in */ |
| 122 | struct trie_common c; |
| 123 | int stringlen; |
| 124 | off_t subnode; |
| 125 | char string[1]; |
| 126 | } str; |
| 127 | }; |
| 128 | #define TRIE_MAGIC 0x75646761UL |
| 129 | #define TRIE_ALIGN alignof(union trie_node) |
| 130 | |
| 131 | /* ---------------------------------------------------------------------- |
| 132 | * Trie-building functions. |
| 133 | */ |
| 134 | |
| 135 | struct tbswitch { |
| 136 | int len; |
| 137 | char c[256]; |
| 138 | off_t off[256]; |
| 139 | int count[256]; |
| 140 | }; |
| 141 | |
| 142 | struct triebuild { |
| 143 | int fd; |
| 144 | off_t offset; |
| 145 | char *lastpath; |
| 146 | int lastlen, lastsize; |
| 147 | off_t lastoff; |
| 148 | struct tbswitch *switches; |
| 149 | int switchsize; |
| 150 | size_t maxpathlen; |
| 151 | }; |
| 152 | |
| 153 | static void tb_seek(triebuild *tb, off_t off) |
| 154 | { |
| 155 | tb->offset = off; |
| 156 | if (lseek(tb->fd, off, SEEK_SET) < 0) { |
| 157 | fprintf(stderr, "agedu: lseek: %s\n", strerror(errno)); |
| 158 | exit(1); |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | static void tb_write(triebuild *tb, const void *buf, size_t len) |
| 163 | { |
| 164 | tb->offset += len; |
| 165 | while (len > 0) { |
| 166 | int ret = write(tb->fd, buf, len); |
| 167 | if (ret < 0) { |
| 168 | fprintf(stderr, "agedu: write: %s\n", strerror(errno)); |
| 169 | exit(1); |
| 170 | } |
| 171 | len -= ret; |
| 172 | buf = (const void *)((const char *)buf + ret); |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | static char trie_align_zeroes[TRIE_ALIGN]; |
| 177 | |
| 178 | static void tb_align(triebuild *tb) |
| 179 | { |
| 180 | int off = (TRIE_ALIGN - ((tb)->offset % TRIE_ALIGN)) % TRIE_ALIGN; |
| 181 | tb_write(tb, trie_align_zeroes, off); |
| 182 | } |
| 183 | |
| 184 | triebuild *triebuild_new(int fd) |
| 185 | { |
| 186 | triebuild *tb = snew(triebuild); |
| 187 | struct trie_header th; |
| 188 | |
| 189 | tb->fd = fd; |
| 190 | tb->lastpath = NULL; |
| 191 | tb->lastlen = tb->lastsize = 0; |
| 192 | tb->lastoff = 0; |
| 193 | tb->switches = NULL; |
| 194 | tb->switchsize = 0; |
| 195 | tb->maxpathlen = 0; |
| 196 | |
| 197 | th.magic = TRIE_MAGIC; |
| 198 | th.root = th.count = 0; |
| 199 | th.indexroot = 0; |
| 200 | th.maxpathlen = 0; |
| 201 | |
| 202 | tb_seek(tb, 0); |
| 203 | tb_write(tb, &th, sizeof(th)); |
| 204 | |
| 205 | return tb; |
| 206 | } |
| 207 | |
| 208 | static off_t triebuild_unwind(triebuild *tb, int targetdepth, int *outcount) |
| 209 | { |
| 210 | off_t offset; |
| 211 | int count, depth; |
| 212 | |
| 213 | if (tb->lastoff == 0) { |
| 214 | *outcount = 0; |
| 215 | return 0; |
| 216 | } |
| 217 | |
| 218 | offset = tb->lastoff; |
| 219 | count = 1; |
| 220 | depth = tb->lastlen + 1; |
| 221 | |
| 222 | assert(depth >= targetdepth); |
| 223 | |
| 224 | while (depth > targetdepth) { |
| 225 | int odepth = depth; |
| 226 | while (depth > targetdepth && |
| 227 | (depth-1 > tb->switchsize || tb->switches[depth-1].len == 0)) |
| 228 | depth--; |
| 229 | if (odepth > depth) { |
| 230 | /* |
| 231 | * Write out a string node. |
| 232 | */ |
| 233 | size_t nodesize = sizeof(struct trie_string) + odepth - depth; |
| 234 | struct trie_string *st = (struct trie_string *)smalloc(nodesize); |
| 235 | st->c.type = TRIE_STRING; |
| 236 | st->stringlen = odepth - depth; |
| 237 | st->subnode = offset; |
| 238 | memcpy(st->string, tb->lastpath + depth, odepth - depth); |
| 239 | tb_align(tb); |
| 240 | offset = tb->offset; |
| 241 | tb_write(tb, st, nodesize); |
| 242 | sfree(st); |
| 243 | } |
| 244 | |
| 245 | assert(depth >= targetdepth); |
| 246 | if (depth <= targetdepth) |
| 247 | break; |
| 248 | |
| 249 | /* |
| 250 | * Now we expect to be sitting just below a switch node. |
| 251 | * Add our final entry to it and write it out. |
| 252 | */ |
| 253 | depth--; |
| 254 | { |
| 255 | struct trie_switch *sw; |
| 256 | char *chars; |
| 257 | size_t nodesize; |
| 258 | int swlen = tb->switches[depth].len; |
| 259 | int i; |
| 260 | |
| 261 | assert(swlen > 0); |
| 262 | |
| 263 | tb->switches[depth].c[swlen] = tb->lastpath[depth]; |
| 264 | tb->switches[depth].off[swlen] = offset; |
| 265 | tb->switches[depth].count[swlen] = count; |
| 266 | swlen++; |
| 267 | |
| 268 | nodesize = sizeof(struct trie_switch) + |
| 269 | swlen * sizeof(struct trie_switchentry) + swlen; |
| 270 | sw = (struct trie_switch *)smalloc(nodesize); |
| 271 | chars = (char *)&sw->sw[swlen]; |
| 272 | |
| 273 | sw->c.type = TRIE_SWITCH; |
| 274 | sw->len = swlen; |
| 275 | count = 0; |
| 276 | for (i = 0; i < swlen; i++) { |
| 277 | sw->sw[i].subnode = tb->switches[depth].off[i]; |
| 278 | sw->sw[i].subcount = tb->switches[depth].count[i]; |
| 279 | chars[i] = tb->switches[depth].c[i]; |
| 280 | |
| 281 | count += tb->switches[depth].count[i]; |
| 282 | } |
| 283 | |
| 284 | tb_align(tb); |
| 285 | offset = tb->offset; |
| 286 | tb_write(tb, sw, nodesize); |
| 287 | sfree(sw); |
| 288 | |
| 289 | tb->switches[depth].len = 0; /* clear this node */ |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | *outcount = count; |
| 294 | return offset; |
| 295 | } |
| 296 | |
| 297 | void triebuild_add(triebuild *tb, const char *pathname, |
| 298 | const struct trie_file *file) |
| 299 | { |
| 300 | int pathlen = strlen(pathname); |
| 301 | int depth; |
| 302 | |
| 303 | if (tb->maxpathlen < pathlen+1) |
| 304 | tb->maxpathlen = pathlen+1; |
| 305 | |
| 306 | if (tb->lastpath) { |
| 307 | off_t offset; |
| 308 | int count; |
| 309 | |
| 310 | /* |
| 311 | * Find the first differing character between this pathname |
| 312 | * and the previous one. |
| 313 | */ |
| 314 | int ret = triecmp(tb->lastpath, pathname, &depth); |
| 315 | assert(ret < 0); |
| 316 | |
| 317 | /* |
| 318 | * Finalise all nodes above this depth. |
| 319 | */ |
| 320 | offset = triebuild_unwind(tb, depth+1, &count); |
| 321 | |
| 322 | /* |
| 323 | * Add the final node we just acquired to the switch node |
| 324 | * at our chosen depth, creating it if it isn't already |
| 325 | * there. |
| 326 | */ |
| 327 | if (tb->switchsize <= depth) { |
| 328 | int oldsize = tb->switchsize; |
| 329 | tb->switchsize = depth * 3 / 2 + 64; |
| 330 | tb->switches = sresize(tb->switches, tb->switchsize, |
| 331 | struct tbswitch); |
| 332 | while (oldsize < tb->switchsize) |
| 333 | tb->switches[oldsize++].len = 0; |
| 334 | } |
| 335 | |
| 336 | tb->switches[depth].c[tb->switches[depth].len] = tb->lastpath[depth]; |
| 337 | tb->switches[depth].off[tb->switches[depth].len] = offset; |
| 338 | tb->switches[depth].count[tb->switches[depth].len] = count; |
| 339 | tb->switches[depth].len++; |
| 340 | } |
| 341 | |
| 342 | /* |
| 343 | * Write out a leaf node for the new file, and remember its |
| 344 | * file offset. |
| 345 | */ |
| 346 | { |
| 347 | struct trie_leaf leaf; |
| 348 | |
| 349 | leaf.c.type = TRIE_LEAF; |
| 350 | leaf.file = *file; /* structure copy */ |
| 351 | |
| 352 | tb_align(tb); |
| 353 | tb->lastoff = tb->offset; |
| 354 | tb_write(tb, &leaf, sizeof(leaf)); |
| 355 | } |
| 356 | |
| 357 | /* |
| 358 | * Store this pathname for comparison with the next one. |
| 359 | */ |
| 360 | if (tb->lastsize < pathlen+1) { |
| 361 | tb->lastsize = pathlen * 3 / 2 + 64; |
| 362 | tb->lastpath = sresize(tb->lastpath, tb->lastsize, char); |
| 363 | } |
| 364 | strcpy(tb->lastpath, pathname); |
| 365 | tb->lastlen = pathlen; |
| 366 | } |
| 367 | |
| 368 | int triebuild_finish(triebuild *tb) |
| 369 | { |
| 370 | struct trie_header th; |
| 371 | |
| 372 | th.magic = TRIE_MAGIC; |
| 373 | th.root = triebuild_unwind(tb, 0, &th.count); |
| 374 | th.indexroot = 0; |
| 375 | th.maxpathlen = tb->maxpathlen; |
| 376 | |
| 377 | tb_seek(tb, 0); |
| 378 | tb_write(tb, &th, sizeof(th)); |
| 379 | |
| 380 | return th.count; |
| 381 | } |
| 382 | |
| 383 | void triebuild_free(triebuild *tb) |
| 384 | { |
| 385 | sfree(tb->switches); |
| 386 | sfree(tb->lastpath); |
| 387 | sfree(tb); |
| 388 | } |
| 389 | |
| 390 | /* ---------------------------------------------------------------------- |
| 391 | * Querying functions. |
| 392 | */ |
| 393 | |
| 394 | #define NODE(t, off, type) \ |
| 395 | ((const struct type *)((const char *)(t) + (off))) |
| 396 | |
| 397 | size_t trie_maxpathlen(const void *t) |
| 398 | { |
| 399 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 400 | return hdr->maxpathlen; |
| 401 | } |
| 402 | |
| 403 | unsigned long trie_before(const void *t, const char *pathname) |
| 404 | { |
| 405 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 406 | int ret = 0, lastcount = hdr->count; |
| 407 | int len = 1 + strlen(pathname), depth = 0; |
| 408 | off_t off = hdr->root; |
| 409 | |
| 410 | while (1) { |
| 411 | const struct trie_common *node = NODE(t, off, trie_common); |
| 412 | if (node->type == TRIE_LEAF) { |
| 413 | if (depth < len) |
| 414 | ret += lastcount; /* _shouldn't_ happen, but in principle */ |
| 415 | return ret; |
| 416 | } else if (node->type == TRIE_STRING) { |
| 417 | const struct trie_string *st = NODE(t, off, trie_string); |
| 418 | |
| 419 | int offset; |
| 420 | int cmp = triencmp(st->string, st->stringlen, |
| 421 | pathname + depth, len-depth, &offset); |
| 422 | |
| 423 | if (offset < st->stringlen) { |
| 424 | if (cmp < 0) |
| 425 | ret += lastcount; |
| 426 | return ret; |
| 427 | } |
| 428 | |
| 429 | depth += st->stringlen; |
| 430 | off = st->subnode; |
| 431 | } else if (node->type == TRIE_SWITCH) { |
| 432 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
| 433 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 434 | int i; |
| 435 | |
| 436 | for (i = 0; i < sw->len; i++) { |
| 437 | int c = chars[i]; |
| 438 | int cmp = trieccmp(pathname[depth], c); |
| 439 | if (cmp > 0) |
| 440 | ret += sw->sw[i].subcount; |
| 441 | else if (cmp < 0) |
| 442 | return ret; |
| 443 | else { |
| 444 | off = sw->sw[i].subnode; |
| 445 | lastcount = sw->sw[i].subcount; |
| 446 | depth++; |
| 447 | break; |
| 448 | } |
| 449 | } |
| 450 | if (i == sw->len) |
| 451 | return ret; |
| 452 | } |
| 453 | } |
| 454 | } |
| 455 | |
| 456 | void trie_getpath(const void *t, unsigned long n, char *buf) |
| 457 | { |
| 458 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 459 | int depth = 0; |
| 460 | off_t off = hdr->root; |
| 461 | |
| 462 | while (1) { |
| 463 | const struct trie_common *node = NODE(t, off, trie_common); |
| 464 | if (node->type == TRIE_LEAF) { |
| 465 | assert(depth > 0 && buf[depth-1] == '\0'); |
| 466 | return; |
| 467 | } else if (node->type == TRIE_STRING) { |
| 468 | const struct trie_string *st = NODE(t, off, trie_string); |
| 469 | |
| 470 | memcpy(buf + depth, st->string, st->stringlen); |
| 471 | depth += st->stringlen; |
| 472 | off = st->subnode; |
| 473 | } else if (node->type == TRIE_SWITCH) { |
| 474 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
| 475 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 476 | int i; |
| 477 | |
| 478 | for (i = 0; i < sw->len; i++) { |
| 479 | if (n < sw->sw[i].subcount) { |
| 480 | buf[depth++] = chars[i]; |
| 481 | off = sw->sw[i].subnode; |
| 482 | break; |
| 483 | } else |
| 484 | n -= sw->sw[i].subcount; |
| 485 | } |
| 486 | assert(i < sw->len); |
| 487 | } |
| 488 | } |
| 489 | } |
| 490 | |
| 491 | unsigned long trie_count(const void *t) |
| 492 | { |
| 493 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
| 494 | return hdr->count; |
| 495 | } |
| 496 | |
| 497 | struct triewalk_switch { |
| 498 | const struct trie_switch *sw; |
| 499 | int pos, depth, count; |
| 500 | }; |
| 501 | struct triewalk { |
| 502 | const void *t; |
| 503 | struct triewalk_switch *switches; |
| 504 | int nswitches, switchsize; |
| 505 | int count; |
| 506 | }; |
| 507 | triewalk *triewalk_new(const void *vt) |
| 508 | { |
| 509 | triewalk *tw = snew(triewalk); |
| 510 | |
| 511 | tw->t = (const char *)vt; |
| 512 | tw->switches = NULL; |
| 513 | tw->switchsize = 0; |
| 514 | tw->nswitches = -1; |
| 515 | tw->count = 0; |
| 516 | |
| 517 | return tw; |
| 518 | } |
| 519 | const struct trie_file *triewalk_next(triewalk *tw, char *buf) |
| 520 | { |
| 521 | off_t off; |
| 522 | int depth; |
| 523 | |
| 524 | if (tw->nswitches < 0) { |
| 525 | const struct trie_header *hdr = NODE(tw->t, 0, trie_header); |
| 526 | off = hdr->root; |
| 527 | depth = 0; |
| 528 | tw->nswitches = 0; |
| 529 | } else { |
| 530 | while (1) { |
| 531 | int swpos; |
| 532 | const struct trie_switch *sw; |
| 533 | const char *chars; |
| 534 | |
| 535 | if (tw->nswitches == 0) { |
| 536 | assert(tw->count == NODE(tw->t, 0, trie_header)->count); |
| 537 | return NULL; /* run out of trie */ |
| 538 | } |
| 539 | |
| 540 | swpos = tw->switches[tw->nswitches-1].pos; |
| 541 | sw = tw->switches[tw->nswitches-1].sw; |
| 542 | chars = (const char *)&sw->sw[sw->len]; |
| 543 | |
| 544 | if (swpos < sw->len) { |
| 545 | depth = tw->switches[tw->nswitches-1].depth; |
| 546 | off = sw->sw[swpos].subnode; |
| 547 | if (buf) |
| 548 | buf[depth++] = chars[swpos]; |
| 549 | assert(tw->count == tw->switches[tw->nswitches-1].count); |
| 550 | tw->switches[tw->nswitches-1].count += sw->sw[swpos].subcount; |
| 551 | tw->switches[tw->nswitches-1].pos++; |
| 552 | break; |
| 553 | } |
| 554 | |
| 555 | tw->nswitches--; |
| 556 | } |
| 557 | } |
| 558 | |
| 559 | while (1) { |
| 560 | const struct trie_common *node = NODE(tw->t, off, trie_common); |
| 561 | if (node->type == TRIE_LEAF) { |
| 562 | const struct trie_leaf *lf = NODE(tw->t, off, trie_leaf); |
| 563 | if (buf) |
| 564 | assert(depth > 0 && buf[depth-1] == '\0'); |
| 565 | tw->count++; |
| 566 | return &lf->file; |
| 567 | } else if (node->type == TRIE_STRING) { |
| 568 | const struct trie_string *st = NODE(tw->t, off, trie_string); |
| 569 | |
| 570 | if (buf) |
| 571 | memcpy(buf + depth, st->string, st->stringlen); |
| 572 | depth += st->stringlen; |
| 573 | off = st->subnode; |
| 574 | } else if (node->type == TRIE_SWITCH) { |
| 575 | const struct trie_switch *sw = NODE(tw->t, off, trie_switch); |
| 576 | const char *chars = (const char *)&sw->sw[sw->len]; |
| 577 | |
| 578 | if (tw->nswitches >= tw->switchsize) { |
| 579 | tw->switchsize = tw->nswitches * 3 / 2 + 32; |
| 580 | tw->switches = sresize(tw->switches, tw->switchsize, |
| 581 | struct triewalk_switch); |
| 582 | } |
| 583 | |
| 584 | tw->switches[tw->nswitches].sw = sw; |
| 585 | tw->switches[tw->nswitches].pos = 1; |
| 586 | tw->switches[tw->nswitches].depth = depth; |
| 587 | tw->switches[tw->nswitches].count = tw->count + sw->sw[0].subcount; |
| 588 | off = sw->sw[0].subnode; |
| 589 | if (buf) |
| 590 | buf[depth++] = chars[0]; |
| 591 | tw->nswitches++; |
| 592 | } |
| 593 | } |
| 594 | } |
| 595 | void triewalk_free(triewalk *tw) |
| 596 | { |
| 597 | sfree(tw->switches); |
| 598 | sfree(tw); |
| 599 | } |
| 600 | |
| 601 | void trie_set_index_offset(void *t, off_t ptr) |
| 602 | { |
| 603 | ((struct trie_header *)t)->indexroot = ptr; |
| 604 | } |
| 605 | off_t trie_get_index_offset(const void *t) |
| 606 | { |
| 607 | return ((const struct trie_header *)t)->indexroot; |
| 608 | } |