Merge commit 'git/master'
[git/git-bigfiles.git] / pack-redundant.c
blob48a12bc1352ad53fbc19ec8c5982a91673a098e1
1 /*
3 * Copyright 2005, Lukas Sandstrom <lukass@etek.chalmers.se>
5 * This file is licensed under the GPL v2.
7 */
9 #include "cache.h"
10 #include "exec_cmd.h"
12 #define BLKSIZE 512
14 static const char pack_redundant_usage[] =
15 "git pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
17 static int load_all_packs, verbose, alt_odb;
19 struct llist_item {
20 struct llist_item *next;
21 const unsigned char *sha1;
23 static struct llist {
24 struct llist_item *front;
25 struct llist_item *back;
26 size_t size;
27 } *all_objects; /* all objects which must be present in local packfiles */
29 static struct pack_list {
30 struct pack_list *next;
31 struct packed_git *pack;
32 struct llist *unique_objects;
33 struct llist *all_objects;
34 } *local_packs = NULL, *altodb_packs = NULL;
36 struct pll {
37 struct pll *next;
38 struct pack_list *pl;
41 static struct llist_item *free_nodes;
43 static inline void llist_item_put(struct llist_item *item)
45 item->next = free_nodes;
46 free_nodes = item;
49 static inline struct llist_item *llist_item_get(void)
51 struct llist_item *new;
52 if ( free_nodes ) {
53 new = free_nodes;
54 free_nodes = free_nodes->next;
55 } else {
56 int i = 1;
57 new = xmalloc(sizeof(struct llist_item) * BLKSIZE);
58 for(;i < BLKSIZE; i++) {
59 llist_item_put(&new[i]);
62 return new;
65 static void llist_free(struct llist *list)
67 while((list->back = list->front)) {
68 list->front = list->front->next;
69 llist_item_put(list->back);
71 free(list);
74 static inline void llist_init(struct llist **list)
76 *list = xmalloc(sizeof(struct llist));
77 (*list)->front = (*list)->back = NULL;
78 (*list)->size = 0;
81 static struct llist * llist_copy(struct llist *list)
83 struct llist *ret;
84 struct llist_item *new, *old, *prev;
86 llist_init(&ret);
88 if ((ret->size = list->size) == 0)
89 return ret;
91 new = ret->front = llist_item_get();
92 new->sha1 = list->front->sha1;
94 old = list->front->next;
95 while (old) {
96 prev = new;
97 new = llist_item_get();
98 prev->next = new;
99 new->sha1 = old->sha1;
100 old = old->next;
102 new->next = NULL;
103 ret->back = new;
105 return ret;
108 static inline struct llist_item *llist_insert(struct llist *list,
109 struct llist_item *after,
110 const unsigned char *sha1)
112 struct llist_item *new = llist_item_get();
113 new->sha1 = sha1;
114 new->next = NULL;
116 if (after != NULL) {
117 new->next = after->next;
118 after->next = new;
119 if (after == list->back)
120 list->back = new;
121 } else {/* insert in front */
122 if (list->size == 0)
123 list->back = new;
124 else
125 new->next = list->front;
126 list->front = new;
128 list->size++;
129 return new;
132 static inline struct llist_item *llist_insert_back(struct llist *list,
133 const unsigned char *sha1)
135 return llist_insert(list, list->back, sha1);
138 static inline struct llist_item *llist_insert_sorted_unique(struct llist *list,
139 const unsigned char *sha1, struct llist_item *hint)
141 struct llist_item *prev = NULL, *l;
143 l = (hint == NULL) ? list->front : hint;
144 while (l) {
145 int cmp = hashcmp(l->sha1, sha1);
146 if (cmp > 0) { /* we insert before this entry */
147 return llist_insert(list, prev, sha1);
149 if(!cmp) { /* already exists */
150 return l;
152 prev = l;
153 l = l->next;
155 /* insert at the end */
156 return llist_insert_back(list, sha1);
159 /* returns a pointer to an item in front of sha1 */
160 static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
162 struct llist_item *prev, *l;
164 redo_from_start:
165 l = (hint == NULL) ? list->front : hint;
166 prev = NULL;
167 while (l) {
168 int cmp = hashcmp(l->sha1, sha1);
169 if (cmp > 0) /* not in list, since sorted */
170 return prev;
171 if(!cmp) { /* found */
172 if (prev == NULL) {
173 if (hint != NULL && hint != list->front) {
174 /* we don't know the previous element */
175 hint = NULL;
176 goto redo_from_start;
178 list->front = l->next;
179 } else
180 prev->next = l->next;
181 if (l == list->back)
182 list->back = prev;
183 llist_item_put(l);
184 list->size--;
185 return prev;
187 prev = l;
188 l = l->next;
190 return prev;
193 /* computes A\B */
194 static void llist_sorted_difference_inplace(struct llist *A,
195 struct llist *B)
197 struct llist_item *hint, *b;
199 hint = NULL;
200 b = B->front;
202 while (b) {
203 hint = llist_sorted_remove(A, b->sha1, hint);
204 b = b->next;
208 static inline struct pack_list * pack_list_insert(struct pack_list **pl,
209 struct pack_list *entry)
211 struct pack_list *p = xmalloc(sizeof(struct pack_list));
212 memcpy(p, entry, sizeof(struct pack_list));
213 p->next = *pl;
214 *pl = p;
215 return p;
218 static inline size_t pack_list_size(struct pack_list *pl)
220 size_t ret = 0;
221 while(pl) {
222 ret++;
223 pl = pl->next;
225 return ret;
228 static struct pack_list * pack_list_difference(const struct pack_list *A,
229 const struct pack_list *B)
231 struct pack_list *ret;
232 const struct pack_list *pl;
234 if (A == NULL)
235 return NULL;
237 pl = B;
238 while (pl != NULL) {
239 if (A->pack == pl->pack)
240 return pack_list_difference(A->next, B);
241 pl = pl->next;
243 ret = xmalloc(sizeof(struct pack_list));
244 memcpy(ret, A, sizeof(struct pack_list));
245 ret->next = pack_list_difference(A->next, B);
246 return ret;
249 static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
251 unsigned long p1_off = 0, p2_off = 0, p1_step, p2_step;
252 const unsigned char *p1_base, *p2_base;
253 struct llist_item *p1_hint = NULL, *p2_hint = NULL;
255 p1_base = p1->pack->index_data;
256 p2_base = p2->pack->index_data;
257 p1_base += 256 * 4 + ((p1->pack->index_version < 2) ? 4 : 8);
258 p2_base += 256 * 4 + ((p2->pack->index_version < 2) ? 4 : 8);
259 p1_step = (p1->pack->index_version < 2) ? 24 : 20;
260 p2_step = (p2->pack->index_version < 2) ? 24 : 20;
262 while (p1_off < p1->pack->num_objects * p1_step &&
263 p2_off < p2->pack->num_objects * p2_step)
265 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
266 /* cmp ~ p1 - p2 */
267 if (cmp == 0) {
268 p1_hint = llist_sorted_remove(p1->unique_objects,
269 p1_base + p1_off, p1_hint);
270 p2_hint = llist_sorted_remove(p2->unique_objects,
271 p1_base + p1_off, p2_hint);
272 p1_off += p1_step;
273 p2_off += p2_step;
274 continue;
276 if (cmp < 0) { /* p1 has the object, p2 doesn't */
277 p1_off += p1_step;
278 } else { /* p2 has the object, p1 doesn't */
279 p2_off += p2_step;
284 static void pll_free(struct pll *l)
286 struct pll *old;
287 struct pack_list *opl;
289 while (l) {
290 old = l;
291 while (l->pl) {
292 opl = l->pl;
293 l->pl = opl->next;
294 free(opl);
296 l = l->next;
297 free(old);
301 /* all the permutations have to be free()d at the same time,
302 * since they refer to each other
304 static struct pll * get_permutations(struct pack_list *list, int n)
306 struct pll *subset, *ret = NULL, *new_pll = NULL, *pll;
308 if (list == NULL || pack_list_size(list) < n || n == 0)
309 return NULL;
311 if (n == 1) {
312 while (list) {
313 new_pll = xmalloc(sizeof(pll));
314 new_pll->pl = NULL;
315 pack_list_insert(&new_pll->pl, list);
316 new_pll->next = ret;
317 ret = new_pll;
318 list = list->next;
320 return ret;
323 while (list->next) {
324 subset = get_permutations(list->next, n - 1);
325 while (subset) {
326 new_pll = xmalloc(sizeof(pll));
327 new_pll->pl = subset->pl;
328 pack_list_insert(&new_pll->pl, list);
329 new_pll->next = ret;
330 ret = new_pll;
331 subset = subset->next;
333 list = list->next;
335 return ret;
338 static int is_superset(struct pack_list *pl, struct llist *list)
340 struct llist *diff;
342 diff = llist_copy(list);
344 while (pl) {
345 llist_sorted_difference_inplace(diff, pl->all_objects);
346 if (diff->size == 0) { /* we're done */
347 llist_free(diff);
348 return 1;
350 pl = pl->next;
352 llist_free(diff);
353 return 0;
356 static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
358 size_t ret = 0;
359 unsigned long p1_off = 0, p2_off = 0, p1_step, p2_step;
360 const unsigned char *p1_base, *p2_base;
362 p1_base = p1->index_data;
363 p2_base = p2->index_data;
364 p1_base += 256 * 4 + ((p1->index_version < 2) ? 4 : 8);
365 p2_base += 256 * 4 + ((p2->index_version < 2) ? 4 : 8);
366 p1_step = (p1->index_version < 2) ? 24 : 20;
367 p2_step = (p2->index_version < 2) ? 24 : 20;
369 while (p1_off < p1->num_objects * p1_step &&
370 p2_off < p2->num_objects * p2_step)
372 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
373 /* cmp ~ p1 - p2 */
374 if (cmp == 0) {
375 ret++;
376 p1_off += p1_step;
377 p2_off += p2_step;
378 continue;
380 if (cmp < 0) { /* p1 has the object, p2 doesn't */
381 p1_off += p1_step;
382 } else { /* p2 has the object, p1 doesn't */
383 p2_off += p2_step;
386 return ret;
389 /* another O(n^2) function ... */
390 static size_t get_pack_redundancy(struct pack_list *pl)
392 struct pack_list *subset;
393 size_t ret = 0;
395 if (pl == NULL)
396 return 0;
398 while ((subset = pl->next)) {
399 while(subset) {
400 ret += sizeof_union(pl->pack, subset->pack);
401 subset = subset->next;
403 pl = pl->next;
405 return ret;
408 static inline off_t pack_set_bytecount(struct pack_list *pl)
410 off_t ret = 0;
411 while (pl) {
412 ret += pl->pack->pack_size;
413 ret += pl->pack->index_size;
414 pl = pl->next;
416 return ret;
419 static void minimize(struct pack_list **min)
421 struct pack_list *pl, *unique = NULL,
422 *non_unique = NULL, *min_perm = NULL;
423 struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
424 struct llist *missing;
425 off_t min_perm_size = 0, perm_size;
426 int n;
428 pl = local_packs;
429 while (pl) {
430 if(pl->unique_objects->size)
431 pack_list_insert(&unique, pl);
432 else
433 pack_list_insert(&non_unique, pl);
434 pl = pl->next;
436 /* find out which objects are missing from the set of unique packs */
437 missing = llist_copy(all_objects);
438 pl = unique;
439 while (pl) {
440 llist_sorted_difference_inplace(missing, pl->all_objects);
441 pl = pl->next;
444 /* return if there are no objects missing from the unique set */
445 if (missing->size == 0) {
446 *min = unique;
447 return;
450 /* find the permutations which contain all missing objects */
451 for (n = 1; n <= pack_list_size(non_unique) && !perm_ok; n++) {
452 perm_all = perm = get_permutations(non_unique, n);
453 while (perm) {
454 if (is_superset(perm->pl, missing)) {
455 new_perm = xmalloc(sizeof(struct pll));
456 memcpy(new_perm, perm, sizeof(struct pll));
457 new_perm->next = perm_ok;
458 perm_ok = new_perm;
460 perm = perm->next;
462 if (perm_ok)
463 break;
464 pll_free(perm_all);
466 if (perm_ok == NULL)
467 die("Internal error: No complete sets found!");
469 /* find the permutation with the smallest size */
470 perm = perm_ok;
471 while (perm) {
472 perm_size = pack_set_bytecount(perm->pl);
473 if (!min_perm_size || min_perm_size > perm_size) {
474 min_perm_size = perm_size;
475 min_perm = perm->pl;
477 perm = perm->next;
479 *min = min_perm;
480 /* add the unique packs to the list */
481 pl = unique;
482 while(pl) {
483 pack_list_insert(min, pl);
484 pl = pl->next;
488 static void load_all_objects(void)
490 struct pack_list *pl = local_packs;
491 struct llist_item *hint, *l;
493 llist_init(&all_objects);
495 while (pl) {
496 hint = NULL;
497 l = pl->all_objects->front;
498 while (l) {
499 hint = llist_insert_sorted_unique(all_objects,
500 l->sha1, hint);
501 l = l->next;
503 pl = pl->next;
505 /* remove objects present in remote packs */
506 pl = altodb_packs;
507 while (pl) {
508 llist_sorted_difference_inplace(all_objects, pl->all_objects);
509 pl = pl->next;
513 /* this scales like O(n^2) */
514 static void cmp_local_packs(void)
516 struct pack_list *subset, *pl = local_packs;
518 while ((subset = pl)) {
519 while((subset = subset->next))
520 cmp_two_packs(pl, subset);
521 pl = pl->next;
525 static void scan_alt_odb_packs(void)
527 struct pack_list *local, *alt;
529 alt = altodb_packs;
530 while (alt) {
531 local = local_packs;
532 while (local) {
533 llist_sorted_difference_inplace(local->unique_objects,
534 alt->all_objects);
535 local = local->next;
537 llist_sorted_difference_inplace(all_objects, alt->all_objects);
538 alt = alt->next;
542 static struct pack_list * add_pack(struct packed_git *p)
544 struct pack_list l;
545 unsigned long off = 0, step;
546 const unsigned char *base;
548 if (!p->pack_local && !(alt_odb || verbose))
549 return NULL;
551 l.pack = p;
552 llist_init(&l.all_objects);
554 if (open_pack_index(p))
555 return NULL;
557 base = p->index_data;
558 base += 256 * 4 + ((p->index_version < 2) ? 4 : 8);
559 step = (p->index_version < 2) ? 24 : 20;
560 while (off < p->num_objects * step) {
561 llist_insert_back(l.all_objects, base + off);
562 off += step;
564 /* this list will be pruned in cmp_two_packs later */
565 l.unique_objects = llist_copy(l.all_objects);
566 if (p->pack_local)
567 return pack_list_insert(&local_packs, &l);
568 else
569 return pack_list_insert(&altodb_packs, &l);
572 static struct pack_list * add_pack_file(char *filename)
574 struct packed_git *p = packed_git;
576 if (strlen(filename) < 40)
577 die("Bad pack filename: %s", filename);
579 while (p) {
580 if (strstr(p->pack_name, filename))
581 return add_pack(p);
582 p = p->next;
584 die("Filename %s not found in packed_git", filename);
587 static void load_all(void)
589 struct packed_git *p = packed_git;
591 while (p) {
592 add_pack(p);
593 p = p->next;
597 int main(int argc, char **argv)
599 int i;
600 struct pack_list *min, *red, *pl;
601 struct llist *ignore;
602 unsigned char *sha1;
603 char buf[42]; /* 40 byte sha1 + \n + \0 */
605 git_extract_argv0_path(argv[0]);
607 setup_git_directory();
609 for (i = 1; i < argc; i++) {
610 const char *arg = argv[i];
611 if(!strcmp(arg, "--")) {
612 i++;
613 break;
615 if(!strcmp(arg, "--all")) {
616 load_all_packs = 1;
617 continue;
619 if(!strcmp(arg, "--verbose")) {
620 verbose = 1;
621 continue;
623 if(!strcmp(arg, "--alt-odb")) {
624 alt_odb = 1;
625 continue;
627 if(*arg == '-')
628 usage(pack_redundant_usage);
629 else
630 break;
633 prepare_packed_git();
635 if (load_all_packs)
636 load_all();
637 else
638 while (*(argv + i) != NULL)
639 add_pack_file(*(argv + i++));
641 if (local_packs == NULL)
642 die("Zero packs found!");
644 load_all_objects();
646 cmp_local_packs();
647 if (alt_odb)
648 scan_alt_odb_packs();
650 /* ignore objects given on stdin */
651 llist_init(&ignore);
652 if (!isatty(0)) {
653 while (fgets(buf, sizeof(buf), stdin)) {
654 sha1 = xmalloc(20);
655 if (get_sha1_hex(buf, sha1))
656 die("Bad sha1 on stdin: %s", buf);
657 llist_insert_sorted_unique(ignore, sha1, NULL);
660 llist_sorted_difference_inplace(all_objects, ignore);
661 pl = local_packs;
662 while (pl) {
663 llist_sorted_difference_inplace(pl->unique_objects, ignore);
664 pl = pl->next;
667 minimize(&min);
669 if (verbose) {
670 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
671 (unsigned long)pack_list_size(altodb_packs));
672 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
673 pl = min;
674 while (pl) {
675 fprintf(stderr, "\t%s\n", pl->pack->pack_name);
676 pl = pl->next;
678 fprintf(stderr, "containing %lu duplicate objects "
679 "with a total size of %lukb.\n",
680 (unsigned long)get_pack_redundancy(min),
681 (unsigned long)pack_set_bytecount(min)/1024);
682 fprintf(stderr, "A total of %lu unique objects were considered.\n",
683 (unsigned long)all_objects->size);
684 fprintf(stderr, "Redundant packs (with indexes):\n");
686 pl = red = pack_list_difference(local_packs, min);
687 while (pl) {
688 printf("%s\n%s\n",
689 sha1_pack_index_name(pl->pack->sha1),
690 pl->pack->pack_name);
691 pl = pl->next;
693 if (verbose)
694 fprintf(stderr, "%luMB of redundant packs in total.\n",
695 (unsigned long)pack_set_bytecount(red)/(1024*1024));
697 return 0;