70322ae3 |
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 | |
353bc75d |
14 | #include "agedu.h" |
70322ae3 |
15 | #include "malloc.h" |
16 | #include "trie.h" |
17 | |
18 | #define alignof(typ) ( offsetof(struct { char c; typ t; }, t) ) |
19 | |
20 | /* |
21 | * Compare functions for pathnames. Returns the relative order of |
22 | * the names, like strcmp; also passes back the offset of the |
23 | * first differing character if desired. |
24 | */ |
25 | static int trieccmp(unsigned char a, unsigned char b) |
26 | { |
373a02e5 |
27 | a = (a == '\0' ? '\0' : a == pathsep ? '\1' : a+1); |
28 | b = (b == '\0' ? '\0' : b == pathsep ? '\1' : b+1); |
29 | return (int)a - (int)b; |
70322ae3 |
30 | } |
31 | |
32 | static int triencmp(const char *a, size_t alen, |
33 | const char *b, size_t blen, int *offset) |
34 | { |
35 | int off = 0; |
36 | while (off < alen && off < blen && a[off] == b[off]) |
37 | off++; |
38 | if (offset) |
39 | *offset = off; |
40 | if (off == alen || off == blen) return (off == blen) - (off == alen); |
41 | return trieccmp(a[off], b[off]); |
42 | } |
43 | |
44 | static int triecmp(const char *a, const char *b, int *offset) |
45 | { |
46 | return triencmp(a, strlen(a), b, strlen(b), offset); |
47 | } |
48 | |
49 | /* ---------------------------------------------------------------------- |
50 | * Trie node structures. |
51 | * |
52 | * The trie format stored in the file consists of three distinct |
53 | * node types, each with a distinguishing type field at the start. |
54 | * |
55 | * TRIE_LEAF is a leaf node; it contains an actual trie_file |
56 | * structure, and indicates that when you're searching down the |
57 | * trie with a string, you should now expect to encounter |
58 | * end-of-string. |
59 | * |
60 | * TRIE_SWITCH indicates that the set of strings in the trie |
61 | * include strings with more than one distinct character after the |
62 | * prefix leading up to this point. Hence, it stores multiple |
63 | * subnode pointers and a different single character for each one. |
64 | * |
65 | * TRIE_STRING indicates that at this point everything in the trie |
66 | * has the same next few characters; it stores a single mandatory |
67 | * string fragment and exactly one subnode pointer. |
68 | */ |
69 | enum { |
70 | TRIE_LEAF = 0x7fffe000, |
71 | TRIE_SWITCH, |
72 | TRIE_STRING |
73 | }; |
74 | |
75 | struct trie_common { |
76 | int type; |
77 | }; |
78 | |
79 | struct trie_switchentry { |
80 | off_t subnode; |
81 | int subcount; |
82 | }; |
83 | |
84 | struct trie_leaf { |
85 | struct trie_common c; |
86 | struct trie_file file; |
87 | }; |
88 | |
89 | struct trie_switch { |
90 | struct trie_common c; |
91 | /* |
92 | * sw[0] to sw[len-1] give the subnode pointers and element |
93 | * counts. At &sw[len] is stored len single bytes which are |
94 | * the characters corresponding to each subnode. |
95 | */ |
96 | int len; |
97 | struct trie_switchentry sw[]; |
98 | }; |
99 | |
100 | struct trie_string { |
101 | struct trie_common c; |
102 | int stringlen; |
103 | off_t subnode; |
104 | char string[]; |
105 | }; |
106 | |
107 | struct trie_header { |
108 | unsigned long magic; |
109 | off_t root, indexroot; |
110 | int count; |
111 | size_t maxpathlen; |
269fa2d1 |
112 | int pathsep; |
70322ae3 |
113 | }; |
114 | |
115 | /* Union only used for computing alignment */ |
116 | union trie_node { |
117 | struct trie_leaf leaf; |
118 | struct { /* fake trie_switch with indeterminate array length filled in */ |
119 | struct trie_common c; |
120 | int len; |
121 | struct trie_switchentry sw[1]; |
122 | } sw; |
123 | struct { /* fake trie_string with indeterminate array length filled in */ |
124 | struct trie_common c; |
125 | int stringlen; |
126 | off_t subnode; |
127 | char string[1]; |
128 | } str; |
129 | }; |
130 | #define TRIE_MAGIC 0x75646761UL |
131 | #define TRIE_ALIGN alignof(union trie_node) |
132 | |
133 | /* ---------------------------------------------------------------------- |
134 | * Trie-building functions. |
135 | */ |
136 | |
137 | struct tbswitch { |
138 | int len; |
139 | char c[256]; |
140 | off_t off[256]; |
141 | int count[256]; |
142 | }; |
143 | |
144 | struct triebuild { |
145 | int fd; |
146 | off_t offset; |
147 | char *lastpath; |
148 | int lastlen, lastsize; |
149 | off_t lastoff; |
150 | struct tbswitch *switches; |
151 | int switchsize; |
152 | size_t maxpathlen; |
153 | }; |
154 | |
155 | static void tb_seek(triebuild *tb, off_t off) |
156 | { |
157 | tb->offset = off; |
158 | if (lseek(tb->fd, off, SEEK_SET) < 0) { |
bf53e756 |
159 | fprintf(stderr, PNAME ": lseek: %s\n", strerror(errno)); |
70322ae3 |
160 | exit(1); |
161 | } |
162 | } |
163 | |
164 | static void tb_write(triebuild *tb, const void *buf, size_t len) |
165 | { |
166 | tb->offset += len; |
167 | while (len > 0) { |
168 | int ret = write(tb->fd, buf, len); |
169 | if (ret < 0) { |
bf53e756 |
170 | fprintf(stderr, PNAME ": write: %s\n", strerror(errno)); |
70322ae3 |
171 | exit(1); |
172 | } |
173 | len -= ret; |
174 | buf = (const void *)((const char *)buf + ret); |
175 | } |
176 | } |
177 | |
178 | static char trie_align_zeroes[TRIE_ALIGN]; |
179 | |
180 | static void tb_align(triebuild *tb) |
181 | { |
182 | int off = (TRIE_ALIGN - ((tb)->offset % TRIE_ALIGN)) % TRIE_ALIGN; |
183 | tb_write(tb, trie_align_zeroes, off); |
184 | } |
185 | |
186 | triebuild *triebuild_new(int fd) |
187 | { |
188 | triebuild *tb = snew(triebuild); |
189 | struct trie_header th; |
190 | |
191 | tb->fd = fd; |
192 | tb->lastpath = NULL; |
193 | tb->lastlen = tb->lastsize = 0; |
194 | tb->lastoff = 0; |
195 | tb->switches = NULL; |
196 | tb->switchsize = 0; |
197 | tb->maxpathlen = 0; |
198 | |
199 | th.magic = TRIE_MAGIC; |
200 | th.root = th.count = 0; |
201 | th.indexroot = 0; |
202 | th.maxpathlen = 0; |
269fa2d1 |
203 | th.pathsep = (unsigned char)pathsep; |
70322ae3 |
204 | |
205 | tb_seek(tb, 0); |
206 | tb_write(tb, &th, sizeof(th)); |
207 | |
208 | return tb; |
209 | } |
210 | |
211 | static off_t triebuild_unwind(triebuild *tb, int targetdepth, int *outcount) |
212 | { |
213 | off_t offset; |
214 | int count, depth; |
215 | |
216 | if (tb->lastoff == 0) { |
217 | *outcount = 0; |
218 | return 0; |
219 | } |
220 | |
221 | offset = tb->lastoff; |
222 | count = 1; |
223 | depth = tb->lastlen + 1; |
224 | |
225 | assert(depth >= targetdepth); |
226 | |
227 | while (depth > targetdepth) { |
228 | int odepth = depth; |
229 | while (depth > targetdepth && |
230 | (depth-1 > tb->switchsize || tb->switches[depth-1].len == 0)) |
231 | depth--; |
232 | if (odepth > depth) { |
233 | /* |
234 | * Write out a string node. |
235 | */ |
236 | size_t nodesize = sizeof(struct trie_string) + odepth - depth; |
237 | struct trie_string *st = (struct trie_string *)smalloc(nodesize); |
238 | st->c.type = TRIE_STRING; |
239 | st->stringlen = odepth - depth; |
240 | st->subnode = offset; |
241 | memcpy(st->string, tb->lastpath + depth, odepth - depth); |
242 | tb_align(tb); |
243 | offset = tb->offset; |
244 | tb_write(tb, st, nodesize); |
245 | sfree(st); |
246 | } |
247 | |
248 | assert(depth >= targetdepth); |
249 | if (depth <= targetdepth) |
250 | break; |
251 | |
252 | /* |
253 | * Now we expect to be sitting just below a switch node. |
254 | * Add our final entry to it and write it out. |
255 | */ |
256 | depth--; |
257 | { |
258 | struct trie_switch *sw; |
259 | char *chars; |
260 | size_t nodesize; |
261 | int swlen = tb->switches[depth].len; |
262 | int i; |
263 | |
264 | assert(swlen > 0); |
265 | |
266 | tb->switches[depth].c[swlen] = tb->lastpath[depth]; |
267 | tb->switches[depth].off[swlen] = offset; |
268 | tb->switches[depth].count[swlen] = count; |
269 | swlen++; |
270 | |
271 | nodesize = sizeof(struct trie_switch) + |
272 | swlen * sizeof(struct trie_switchentry) + swlen; |
273 | sw = (struct trie_switch *)smalloc(nodesize); |
274 | chars = (char *)&sw->sw[swlen]; |
275 | |
276 | sw->c.type = TRIE_SWITCH; |
277 | sw->len = swlen; |
278 | count = 0; |
279 | for (i = 0; i < swlen; i++) { |
280 | sw->sw[i].subnode = tb->switches[depth].off[i]; |
281 | sw->sw[i].subcount = tb->switches[depth].count[i]; |
282 | chars[i] = tb->switches[depth].c[i]; |
283 | |
284 | count += tb->switches[depth].count[i]; |
285 | } |
286 | |
287 | tb_align(tb); |
288 | offset = tb->offset; |
289 | tb_write(tb, sw, nodesize); |
290 | sfree(sw); |
291 | |
292 | tb->switches[depth].len = 0; /* clear this node */ |
293 | } |
294 | } |
295 | |
296 | *outcount = count; |
297 | return offset; |
298 | } |
299 | |
300 | void triebuild_add(triebuild *tb, const char *pathname, |
301 | const struct trie_file *file) |
302 | { |
303 | int pathlen = strlen(pathname); |
304 | int depth; |
305 | |
306 | if (tb->maxpathlen < pathlen+1) |
307 | tb->maxpathlen = pathlen+1; |
308 | |
309 | if (tb->lastpath) { |
310 | off_t offset; |
311 | int count; |
312 | |
313 | /* |
314 | * Find the first differing character between this pathname |
315 | * and the previous one. |
316 | */ |
317 | int ret = triecmp(tb->lastpath, pathname, &depth); |
318 | assert(ret < 0); |
319 | |
320 | /* |
321 | * Finalise all nodes above this depth. |
322 | */ |
323 | offset = triebuild_unwind(tb, depth+1, &count); |
324 | |
325 | /* |
326 | * Add the final node we just acquired to the switch node |
327 | * at our chosen depth, creating it if it isn't already |
328 | * there. |
329 | */ |
330 | if (tb->switchsize <= depth) { |
331 | int oldsize = tb->switchsize; |
332 | tb->switchsize = depth * 3 / 2 + 64; |
333 | tb->switches = sresize(tb->switches, tb->switchsize, |
334 | struct tbswitch); |
335 | while (oldsize < tb->switchsize) |
336 | tb->switches[oldsize++].len = 0; |
337 | } |
338 | |
339 | tb->switches[depth].c[tb->switches[depth].len] = tb->lastpath[depth]; |
340 | tb->switches[depth].off[tb->switches[depth].len] = offset; |
341 | tb->switches[depth].count[tb->switches[depth].len] = count; |
342 | tb->switches[depth].len++; |
343 | } |
344 | |
345 | /* |
346 | * Write out a leaf node for the new file, and remember its |
347 | * file offset. |
348 | */ |
349 | { |
350 | struct trie_leaf leaf; |
351 | |
352 | leaf.c.type = TRIE_LEAF; |
353 | leaf.file = *file; /* structure copy */ |
354 | |
355 | tb_align(tb); |
356 | tb->lastoff = tb->offset; |
357 | tb_write(tb, &leaf, sizeof(leaf)); |
358 | } |
359 | |
360 | /* |
361 | * Store this pathname for comparison with the next one. |
362 | */ |
363 | if (tb->lastsize < pathlen+1) { |
364 | tb->lastsize = pathlen * 3 / 2 + 64; |
365 | tb->lastpath = sresize(tb->lastpath, tb->lastsize, char); |
366 | } |
367 | strcpy(tb->lastpath, pathname); |
368 | tb->lastlen = pathlen; |
369 | } |
370 | |
371 | int triebuild_finish(triebuild *tb) |
372 | { |
373 | struct trie_header th; |
374 | |
375 | th.magic = TRIE_MAGIC; |
376 | th.root = triebuild_unwind(tb, 0, &th.count); |
377 | th.indexroot = 0; |
378 | th.maxpathlen = tb->maxpathlen; |
269fa2d1 |
379 | th.pathsep = (unsigned char)pathsep; |
70322ae3 |
380 | |
381 | tb_seek(tb, 0); |
382 | tb_write(tb, &th, sizeof(th)); |
383 | |
384 | return th.count; |
385 | } |
386 | |
387 | void triebuild_free(triebuild *tb) |
388 | { |
389 | sfree(tb->switches); |
390 | sfree(tb->lastpath); |
391 | sfree(tb); |
392 | } |
393 | |
394 | /* ---------------------------------------------------------------------- |
395 | * Querying functions. |
396 | */ |
397 | |
398 | #define NODE(t, off, type) \ |
399 | ((const struct type *)((const char *)(t) + (off))) |
400 | |
401 | size_t trie_maxpathlen(const void *t) |
402 | { |
403 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
404 | return hdr->maxpathlen; |
405 | } |
406 | |
407 | unsigned long trie_before(const void *t, const char *pathname) |
408 | { |
409 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
410 | int ret = 0, lastcount = hdr->count; |
411 | int len = 1 + strlen(pathname), depth = 0; |
412 | off_t off = hdr->root; |
413 | |
414 | while (1) { |
415 | const struct trie_common *node = NODE(t, off, trie_common); |
416 | if (node->type == TRIE_LEAF) { |
417 | if (depth < len) |
418 | ret += lastcount; /* _shouldn't_ happen, but in principle */ |
419 | return ret; |
420 | } else if (node->type == TRIE_STRING) { |
421 | const struct trie_string *st = NODE(t, off, trie_string); |
422 | |
423 | int offset; |
424 | int cmp = triencmp(st->string, st->stringlen, |
425 | pathname + depth, len-depth, &offset); |
426 | |
427 | if (offset < st->stringlen) { |
428 | if (cmp < 0) |
429 | ret += lastcount; |
430 | return ret; |
431 | } |
432 | |
433 | depth += st->stringlen; |
434 | off = st->subnode; |
435 | } else if (node->type == TRIE_SWITCH) { |
436 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
437 | const char *chars = (const char *)&sw->sw[sw->len]; |
438 | int i; |
439 | |
440 | for (i = 0; i < sw->len; i++) { |
441 | int c = chars[i]; |
442 | int cmp = trieccmp(pathname[depth], c); |
443 | if (cmp > 0) |
444 | ret += sw->sw[i].subcount; |
445 | else if (cmp < 0) |
446 | return ret; |
447 | else { |
448 | off = sw->sw[i].subnode; |
449 | lastcount = sw->sw[i].subcount; |
450 | depth++; |
451 | break; |
452 | } |
453 | } |
454 | if (i == sw->len) |
455 | return ret; |
456 | } |
457 | } |
458 | } |
459 | |
460 | void trie_getpath(const void *t, unsigned long n, char *buf) |
461 | { |
462 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
463 | int depth = 0; |
464 | off_t off = hdr->root; |
465 | |
466 | while (1) { |
467 | const struct trie_common *node = NODE(t, off, trie_common); |
468 | if (node->type == TRIE_LEAF) { |
469 | assert(depth > 0 && buf[depth-1] == '\0'); |
470 | return; |
471 | } else if (node->type == TRIE_STRING) { |
472 | const struct trie_string *st = NODE(t, off, trie_string); |
473 | |
474 | memcpy(buf + depth, st->string, st->stringlen); |
475 | depth += st->stringlen; |
476 | off = st->subnode; |
477 | } else if (node->type == TRIE_SWITCH) { |
478 | const struct trie_switch *sw = NODE(t, off, trie_switch); |
479 | const char *chars = (const char *)&sw->sw[sw->len]; |
480 | int i; |
481 | |
482 | for (i = 0; i < sw->len; i++) { |
483 | if (n < sw->sw[i].subcount) { |
484 | buf[depth++] = chars[i]; |
485 | off = sw->sw[i].subnode; |
486 | break; |
487 | } else |
488 | n -= sw->sw[i].subcount; |
489 | } |
490 | assert(i < sw->len); |
491 | } |
492 | } |
493 | } |
494 | |
495 | unsigned long trie_count(const void *t) |
496 | { |
497 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
498 | return hdr->count; |
499 | } |
500 | |
269fa2d1 |
501 | char trie_pathsep(const void *t) |
502 | { |
503 | const struct trie_header *hdr = NODE(t, 0, trie_header); |
504 | return (char)hdr->pathsep; |
505 | } |
506 | |
70322ae3 |
507 | struct triewalk_switch { |
508 | const struct trie_switch *sw; |
509 | int pos, depth, count; |
510 | }; |
511 | struct triewalk { |
512 | const void *t; |
513 | struct triewalk_switch *switches; |
514 | int nswitches, switchsize; |
515 | int count; |
516 | }; |
517 | triewalk *triewalk_new(const void *vt) |
518 | { |
519 | triewalk *tw = snew(triewalk); |
520 | |
521 | tw->t = (const char *)vt; |
522 | tw->switches = NULL; |
523 | tw->switchsize = 0; |
524 | tw->nswitches = -1; |
525 | tw->count = 0; |
526 | |
527 | return tw; |
528 | } |
529 | const struct trie_file *triewalk_next(triewalk *tw, char *buf) |
530 | { |
531 | off_t off; |
532 | int depth; |
533 | |
534 | if (tw->nswitches < 0) { |
535 | const struct trie_header *hdr = NODE(tw->t, 0, trie_header); |
536 | off = hdr->root; |
537 | depth = 0; |
538 | tw->nswitches = 0; |
539 | } else { |
540 | while (1) { |
541 | int swpos; |
542 | const struct trie_switch *sw; |
543 | const char *chars; |
544 | |
545 | if (tw->nswitches == 0) { |
546 | assert(tw->count == NODE(tw->t, 0, trie_header)->count); |
547 | return NULL; /* run out of trie */ |
548 | } |
549 | |
550 | swpos = tw->switches[tw->nswitches-1].pos; |
551 | sw = tw->switches[tw->nswitches-1].sw; |
552 | chars = (const char *)&sw->sw[sw->len]; |
553 | |
554 | if (swpos < sw->len) { |
555 | depth = tw->switches[tw->nswitches-1].depth; |
556 | off = sw->sw[swpos].subnode; |
557 | if (buf) |
558 | buf[depth++] = chars[swpos]; |
559 | assert(tw->count == tw->switches[tw->nswitches-1].count); |
560 | tw->switches[tw->nswitches-1].count += sw->sw[swpos].subcount; |
561 | tw->switches[tw->nswitches-1].pos++; |
562 | break; |
563 | } |
564 | |
565 | tw->nswitches--; |
566 | } |
567 | } |
568 | |
569 | while (1) { |
570 | const struct trie_common *node = NODE(tw->t, off, trie_common); |
571 | if (node->type == TRIE_LEAF) { |
572 | const struct trie_leaf *lf = NODE(tw->t, off, trie_leaf); |
573 | if (buf) |
574 | assert(depth > 0 && buf[depth-1] == '\0'); |
575 | tw->count++; |
576 | return &lf->file; |
577 | } else if (node->type == TRIE_STRING) { |
578 | const struct trie_string *st = NODE(tw->t, off, trie_string); |
579 | |
580 | if (buf) |
581 | memcpy(buf + depth, st->string, st->stringlen); |
582 | depth += st->stringlen; |
583 | off = st->subnode; |
584 | } else if (node->type == TRIE_SWITCH) { |
585 | const struct trie_switch *sw = NODE(tw->t, off, trie_switch); |
586 | const char *chars = (const char *)&sw->sw[sw->len]; |
587 | |
588 | if (tw->nswitches >= tw->switchsize) { |
589 | tw->switchsize = tw->nswitches * 3 / 2 + 32; |
590 | tw->switches = sresize(tw->switches, tw->switchsize, |
591 | struct triewalk_switch); |
592 | } |
593 | |
594 | tw->switches[tw->nswitches].sw = sw; |
595 | tw->switches[tw->nswitches].pos = 1; |
596 | tw->switches[tw->nswitches].depth = depth; |
597 | tw->switches[tw->nswitches].count = tw->count + sw->sw[0].subcount; |
598 | off = sw->sw[0].subnode; |
599 | if (buf) |
600 | buf[depth++] = chars[0]; |
601 | tw->nswitches++; |
602 | } |
603 | } |
604 | } |
605 | void triewalk_free(triewalk *tw) |
606 | { |
607 | sfree(tw->switches); |
608 | sfree(tw); |
609 | } |
610 | |
611 | void trie_set_index_offset(void *t, off_t ptr) |
612 | { |
613 | ((struct trie_header *)t)->indexroot = ptr; |
614 | } |
615 | off_t trie_get_index_offset(const void *t) |
616 | { |
617 | return ((const struct trie_header *)t)->indexroot; |
618 | } |
256c29a2 |
619 | |
620 | void make_successor(char *pathbuf) |
621 | { |
622 | int len = strlen(pathbuf); |
623 | if (len > 0 && pathbuf[len-1] == pathsep) |
624 | len--; |
625 | pathbuf[len] = '\001'; |
626 | pathbuf[len+1] = '\0'; |
627 | } |