3 * Copyright 2005, Lukas Sandstrom <lukass@etek.chalmers.se>
5 * This file is licensed under the GPL v2.
11 static const char pack_redundant_usage[] =
12 "git-pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
14 int load_all_packs = 0, verbose = 0, alt_odb = 0;
17 struct llist_item *next;
21 struct llist_item *front;
22 struct llist_item *back;
24 } *all_objects; /* all objects which must be present in local packfiles */
27 struct pack_list *next;
28 struct packed_git *pack;
29 struct llist *unique_objects;
30 struct llist *all_objects;
31 } *local_packs = NULL, *altodb_packs = NULL;
38 inline void llist_free(struct llist *list)
40 while((list->back = list->front)) {
41 list->front = list->front->next;
47 inline void llist_init(struct llist **list)
49 *list = xmalloc(sizeof(struct llist));
50 (*list)->front = (*list)->back = NULL;
54 struct llist * llist_copy(struct llist *list)
57 struct llist_item *new, *old, *prev;
61 if ((ret->size = list->size) == 0)
64 new = ret->front = xmalloc(sizeof(struct llist_item));
65 new->sha1 = list->front->sha1;
67 old = list->front->next;
70 new = xmalloc(sizeof(struct llist_item));
72 new->sha1 = old->sha1;
81 inline struct llist_item * llist_insert(struct llist *list,
82 struct llist_item *after, char *sha1)
84 struct llist_item *new = xmalloc(sizeof(struct llist_item));
89 new->next = after->next;
91 if (after == list->back)
93 } else {/* insert in front */
97 new->next = list->front;
104 inline struct llist_item * llist_insert_back(struct llist *list, char *sha1)
106 return llist_insert(list, list->back, sha1);
109 inline struct llist_item * llist_insert_sorted_unique(struct llist *list,
110 char *sha1, struct llist_item *hint)
112 struct llist_item *prev = NULL, *l;
114 l = (hint == NULL) ? list->front : hint;
116 int cmp = memcmp(l->sha1, sha1, 20);
117 if (cmp > 0) { /* we insert before this entry */
118 return llist_insert(list, prev, sha1);
120 if(!cmp) { /* already exists */
126 /* insert at the end */
127 return llist_insert_back(list, sha1);
131 void llist_sorted_difference_inplace(struct llist *A,
134 struct llist_item *prev, *a, *b, *x;
139 while (a != NULL && b != NULL) {
140 int cmp = memcmp(a->sha1, b->sha1, 20);
145 a = prev->next = a->next;
147 if (a == NULL) /* end of list */
162 /* returns a pointer to an item in front of sha1 */
163 inline struct llist_item * llist_sorted_remove(struct llist *list, char *sha1,
164 struct llist_item *hint)
166 struct llist_item *prev, *l;
169 l = (hint == NULL) ? list->front : hint;
172 int cmp = memcmp(l->sha1, sha1, 20);
173 if (cmp > 0) /* not in list, since sorted */
175 if(!cmp) { /* found */
177 if (hint != NULL && hint != list->front) {
178 /* we don't know the previous element */
180 goto redo_from_start;
182 list->front = l->next;
184 prev->next = l->next;
197 inline struct pack_list * pack_list_insert(struct pack_list **pl,
198 struct pack_list *entry)
200 struct pack_list *p = xmalloc(sizeof(struct pack_list));
201 memcpy(p, entry, sizeof(struct pack_list));
207 inline size_t pack_list_size(struct pack_list *pl)
217 struct pack_list * pack_list_difference(struct pack_list *A,
220 struct pack_list *ret, *pl;
227 if (A->pack == pl->pack)
228 return pack_list_difference(A->next, B);
231 ret = xmalloc(sizeof(struct pack_list));
232 memcpy(ret, A, sizeof(struct pack_list));
233 ret->next = pack_list_difference(A->next, B);
237 void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
240 void *p1_base, *p2_base;
241 struct llist_item *p1_hint = NULL, *p2_hint = NULL;
243 p1_off = p2_off = 256 * 4 + 4;
244 p1_base = (void *)p1->pack->index_base;
245 p2_base = (void *)p2->pack->index_base;
247 while (p1_off <= p1->pack->index_size - 3 * 20 &&
248 p2_off <= p2->pack->index_size - 3 * 20)
250 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
253 p1_hint = llist_sorted_remove(p1->unique_objects,
254 p1_base + p1_off, p1_hint);
255 p2_hint = llist_sorted_remove(p2->unique_objects,
256 p1_base + p1_off, p2_hint);
261 if (cmp < 0) { /* p1 has the object, p2 doesn't */
263 } else { /* p2 has the object, p1 doesn't */
269 /* all the permutations have to be free()d at the same time,
270 * since they refer to each other
272 struct pll * get_all_permutations(struct pack_list *list)
274 struct pll *subset, *pll, *new_pll = NULL; /*silence warning*/
279 if (list->next == NULL) {
280 new_pll = xmalloc(sizeof(struct pll));
281 new_pll->next = NULL;
286 pll = subset = get_all_permutations(list->next);
288 new_pll = xmalloc(sizeof(struct pll));
289 new_pll->next = pll->next;
292 new_pll->pl = xmalloc(sizeof(struct pack_list));
293 memcpy(new_pll->pl, list, sizeof(struct pack_list));
294 new_pll->pl->next = pll->pl;
298 /* add ourself to the end */
299 new_pll->next = xmalloc(sizeof(struct pll));
300 new_pll->next->pl = xmalloc(sizeof(struct pack_list));
301 new_pll->next->next = NULL;
302 memcpy(new_pll->next->pl, list, sizeof(struct pack_list));
303 new_pll->next->pl->next = NULL;
308 int is_superset(struct pack_list *pl, struct llist *list)
312 diff = llist_copy(list);
315 llist_sorted_difference_inplace(diff,
317 if (diff->size == 0) { /* we're done */
327 size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
331 void *p1_base, *p2_base;
333 p1_off = p2_off = 256 * 4 + 4;
334 p1_base = (void *)p1->index_base;
335 p2_base = (void *)p2->index_base;
337 while (p1_off <= p1->index_size - 3 * 20 &&
338 p2_off <= p2->index_size - 3 * 20)
340 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
348 if (cmp < 0) { /* p1 has the object, p2 doesn't */
350 } else { /* p2 has the object, p1 doesn't */
357 /* another O(n^2) function ... */
358 size_t get_pack_redundancy(struct pack_list *pl)
360 struct pack_list *subset;
366 while ((subset = pl->next)) {
368 ret += sizeof_union(pl->pack, subset->pack);
369 subset = subset->next;
376 inline size_t pack_set_bytecount(struct pack_list *pl)
380 ret += pl->pack->pack_size;
381 ret += pl->pack->index_size;
387 void minimize(struct pack_list **min)
389 struct pack_list *pl, *unique = NULL,
390 *non_unique = NULL, *min_perm = NULL;
391 struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
392 struct llist *missing;
393 size_t min_perm_size = (size_t)-1, perm_size;
397 if(pl->unique_objects->size)
398 pack_list_insert(&unique, pl);
400 pack_list_insert(&non_unique, pl);
403 /* find out which objects are missing from the set of unique packs */
404 missing = llist_copy(all_objects);
407 llist_sorted_difference_inplace(missing,
412 /* return if there are no objects missing from the unique set */
413 if (missing->size == 0) {
418 /* find the permutations which contain all missing objects */
419 perm_all = perm = get_all_permutations(non_unique);
421 if (is_superset(perm->pl, missing)) {
422 new_perm = xmalloc(sizeof(struct pll));
423 new_perm->pl = perm->pl;
424 new_perm->next = perm_ok;
431 die("Internal error: No complete sets found!\n");
433 /* find the permutation with the smallest size */
436 perm_size = pack_set_bytecount(perm->pl);
437 if (min_perm_size > perm_size) {
438 min_perm_size = perm_size;
444 /* add the unique packs to the list */
447 pack_list_insert(min, pl);
452 void load_all_objects()
454 struct pack_list *pl = local_packs;
455 struct llist_item *hint, *l;
458 llist_init(&all_objects);
463 l = pl->all_objects->front;
465 hint = llist_insert_sorted_unique(all_objects,
471 /* remove objects present in remote packs */
474 llist_sorted_difference_inplace(all_objects, pl->all_objects);
479 /* this scales like O(n^2) */
482 struct pack_list *subset, *pl = local_packs;
484 while ((subset = pl)) {
485 while((subset = subset->next))
486 cmp_two_packs(pl, subset);
492 subset = local_packs;
494 llist_sorted_difference_inplace(subset->unique_objects,
496 subset = subset->next;
502 struct pack_list * add_pack(struct packed_git *p)
509 llist_init(&l.all_objects);
512 base = (void *)p->index_base;
513 while (off <= p->index_size - 3 * 20) {
514 llist_insert_back(l.all_objects, base + off);
517 /* this list will be pruned in cmp_two_packs later */
518 l.unique_objects = llist_copy(l.all_objects);
520 return pack_list_insert(&local_packs, &l);
522 return alt_odb ? pack_list_insert(&altodb_packs, &l) : NULL;
525 struct pack_list * add_pack_file(char *filename)
527 struct packed_git *p = packed_git;
529 if (strlen(filename) < 40)
530 die("Bad pack filename: %s\n", filename);
533 if (strstr(p->pack_name, filename))
537 die("Filename %s not found in packed_git\n", filename);
542 struct packed_git *p = packed_git;
550 int main(int argc, char **argv)
553 struct pack_list *min, *red, *pl;
555 for (i = 1; i < argc; i++) {
556 const char *arg = argv[i];
557 if(!strcmp(arg, "--")) {
561 if(!strcmp(arg, "--all")) {
565 if(!strcmp(arg, "--verbose")) {
569 if(!strcmp(arg, "--alt-odb")) {
574 usage(pack_redundant_usage);
579 prepare_packed_git();
584 while (*(argv + i) != NULL)
585 add_pack_file(*(argv + i++));
587 if (local_packs == NULL)
588 die("Zero packs found!\n");
596 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
597 (unsigned long)pack_list_size(altodb_packs));
598 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
601 fprintf(stderr, "\t%s\n", pl->pack->pack_name);
604 fprintf(stderr, "containing %lu duplicate objects "
605 "with a total size of %lukb.\n",
606 (unsigned long)get_pack_redundancy(min),
607 (unsigned long)pack_set_bytecount(min)/1024);
608 fprintf(stderr, "A total of %lu unique objects were considered.\n",
609 (unsigned long)all_objects->size);
610 fprintf(stderr, "Redundant packs (with indexes):\n");
612 pl = red = pack_list_difference(local_packs, min);
615 sha1_pack_index_name(pl->pack->sha1),
616 pl->pack->pack_name);