3 * Copyright 2005, Lukas Sandstrom <lukass@etek.chalmers.se>
5 * This file is licensed under the GPL v2.
13 static const char pack_redundant_usage[] =
14 "git-pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
16 static int load_all_packs = 0, verbose = 0, alt_odb = 0;
19 struct llist_item *next;
23 struct llist_item *front;
24 struct llist_item *back;
26 } *all_objects; /* all objects which must be present in local packfiles */
28 static struct pack_list {
29 struct pack_list *next;
30 struct packed_git *pack;
31 struct llist *unique_objects;
32 struct llist *all_objects;
33 } *local_packs = NULL, *altodb_packs = NULL;
40 static struct llist_item *free_nodes = NULL;
42 static inline void llist_item_put(struct llist_item *item)
44 item->next = free_nodes;
48 static inline struct llist_item *llist_item_get(void)
50 struct llist_item *new;
53 free_nodes = free_nodes->next;
56 new = xmalloc(sizeof(struct llist_item) * BLKSIZE);
57 for(;i < BLKSIZE; i++) {
58 llist_item_put(&new[i]);
64 static void llist_free(struct llist *list)
66 while((list->back = list->front)) {
67 list->front = list->front->next;
68 llist_item_put(list->back);
73 static inline void llist_init(struct llist **list)
75 *list = xmalloc(sizeof(struct llist));
76 (*list)->front = (*list)->back = NULL;
80 static struct llist * llist_copy(struct llist *list)
83 struct llist_item *new, *old, *prev;
87 if ((ret->size = list->size) == 0)
90 new = ret->front = llist_item_get();
91 new->sha1 = list->front->sha1;
93 old = list->front->next;
96 new = llist_item_get();
98 new->sha1 = old->sha1;
107 static inline struct llist_item * llist_insert(struct llist *list,
108 struct llist_item *after,
111 struct llist_item *new = llist_item_get();
116 new->next = after->next;
118 if (after == list->back)
120 } else {/* insert in front */
124 new->next = list->front;
131 static inline struct llist_item *llist_insert_back(struct llist *list, unsigned char *sha1)
133 return llist_insert(list, list->back, sha1);
136 static inline struct llist_item *llist_insert_sorted_unique(struct llist *list, unsigned char *sha1, struct llist_item *hint)
138 struct llist_item *prev = NULL, *l;
140 l = (hint == NULL) ? list->front : hint;
142 int cmp = memcmp(l->sha1, sha1, 20);
143 if (cmp > 0) { /* we insert before this entry */
144 return llist_insert(list, prev, sha1);
146 if(!cmp) { /* already exists */
152 /* insert at the end */
153 return llist_insert_back(list, sha1);
156 /* returns a pointer to an item in front of sha1 */
157 static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
159 struct llist_item *prev, *l;
162 l = (hint == NULL) ? list->front : hint;
165 int cmp = memcmp(l->sha1, sha1, 20);
166 if (cmp > 0) /* not in list, since sorted */
168 if(!cmp) { /* found */
170 if (hint != NULL && hint != list->front) {
171 /* we don't know the previous element */
173 goto redo_from_start;
175 list->front = l->next;
177 prev->next = l->next;
191 static void llist_sorted_difference_inplace(struct llist *A,
194 struct llist_item *hint, *b;
200 hint = llist_sorted_remove(A, b->sha1, hint);
205 static inline struct pack_list * pack_list_insert(struct pack_list **pl,
206 struct pack_list *entry)
208 struct pack_list *p = xmalloc(sizeof(struct pack_list));
209 memcpy(p, entry, sizeof(struct pack_list));
215 static inline size_t pack_list_size(struct pack_list *pl)
225 static struct pack_list * pack_list_difference(const struct pack_list *A,
226 const struct pack_list *B)
228 struct pack_list *ret;
229 const struct pack_list *pl;
236 if (A->pack == pl->pack)
237 return pack_list_difference(A->next, B);
240 ret = xmalloc(sizeof(struct pack_list));
241 memcpy(ret, A, sizeof(struct pack_list));
242 ret->next = pack_list_difference(A->next, B);
246 static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
249 void *p1_base, *p2_base;
250 struct llist_item *p1_hint = NULL, *p2_hint = NULL;
252 p1_off = p2_off = 256 * 4 + 4;
253 p1_base = (void *)p1->pack->index_base;
254 p2_base = (void *)p2->pack->index_base;
256 while (p1_off <= p1->pack->index_size - 3 * 20 &&
257 p2_off <= p2->pack->index_size - 3 * 20)
259 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
262 p1_hint = llist_sorted_remove(p1->unique_objects,
263 p1_base + p1_off, p1_hint);
264 p2_hint = llist_sorted_remove(p2->unique_objects,
265 p1_base + p1_off, p2_hint);
270 if (cmp < 0) { /* p1 has the object, p2 doesn't */
272 } else { /* p2 has the object, p1 doesn't */
278 static void pll_free(struct pll *l)
281 struct pack_list *opl;
295 /* all the permutations have to be free()d at the same time,
296 * since they refer to each other
298 static struct pll * get_permutations(struct pack_list *list, int n)
300 struct pll *subset, *ret = NULL, *new_pll = NULL, *pll;
302 if (list == NULL || pack_list_size(list) < n || n == 0)
307 new_pll = xmalloc(sizeof(pll));
309 pack_list_insert(&new_pll->pl, list);
318 subset = get_permutations(list->next, n - 1);
320 new_pll = xmalloc(sizeof(pll));
321 new_pll->pl = subset->pl;
322 pack_list_insert(&new_pll->pl, list);
325 subset = subset->next;
332 static int is_superset(struct pack_list *pl, struct llist *list)
336 diff = llist_copy(list);
339 llist_sorted_difference_inplace(diff, pl->all_objects);
340 if (diff->size == 0) { /* we're done */
350 static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
354 void *p1_base, *p2_base;
356 p1_off = p2_off = 256 * 4 + 4;
357 p1_base = (void *)p1->index_base;
358 p2_base = (void *)p2->index_base;
360 while (p1_off <= p1->index_size - 3 * 20 &&
361 p2_off <= p2->index_size - 3 * 20)
363 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
371 if (cmp < 0) { /* p1 has the object, p2 doesn't */
373 } else { /* p2 has the object, p1 doesn't */
380 /* another O(n^2) function ... */
381 static size_t get_pack_redundancy(struct pack_list *pl)
383 struct pack_list *subset;
389 while ((subset = pl->next)) {
391 ret += sizeof_union(pl->pack, subset->pack);
392 subset = subset->next;
399 static inline size_t pack_set_bytecount(struct pack_list *pl)
403 ret += pl->pack->pack_size;
404 ret += pl->pack->index_size;
410 static void minimize(struct pack_list **min)
412 struct pack_list *pl, *unique = NULL,
413 *non_unique = NULL, *min_perm = NULL;
414 struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
415 struct llist *missing;
416 size_t min_perm_size = (size_t)-1, perm_size;
421 if(pl->unique_objects->size)
422 pack_list_insert(&unique, pl);
424 pack_list_insert(&non_unique, pl);
427 /* find out which objects are missing from the set of unique packs */
428 missing = llist_copy(all_objects);
431 llist_sorted_difference_inplace(missing, pl->all_objects);
435 /* return if there are no objects missing from the unique set */
436 if (missing->size == 0) {
441 /* find the permutations which contain all missing objects */
442 for (n = 1; n <= pack_list_size(non_unique) && !perm_ok; n++) {
443 perm_all = perm = get_permutations(non_unique, n);
445 if (is_superset(perm->pl, missing)) {
446 new_perm = xmalloc(sizeof(struct pll));
447 memcpy(new_perm, perm, sizeof(struct pll));
448 new_perm->next = perm_ok;
458 die("Internal error: No complete sets found!\n");
460 /* find the permutation with the smallest size */
463 perm_size = pack_set_bytecount(perm->pl);
464 if (min_perm_size > perm_size) {
465 min_perm_size = perm_size;
471 /* add the unique packs to the list */
474 pack_list_insert(min, pl);
479 static void load_all_objects(void)
481 struct pack_list *pl = local_packs;
482 struct llist_item *hint, *l;
484 llist_init(&all_objects);
488 l = pl->all_objects->front;
490 hint = llist_insert_sorted_unique(all_objects,
496 /* remove objects present in remote packs */
499 llist_sorted_difference_inplace(all_objects, pl->all_objects);
504 /* this scales like O(n^2) */
505 static void cmp_local_packs(void)
507 struct pack_list *subset, *pl = local_packs;
509 while ((subset = pl)) {
510 while((subset = subset->next))
511 cmp_two_packs(pl, subset);
516 static void scan_alt_odb_packs(void)
518 struct pack_list *local, *alt;
524 llist_sorted_difference_inplace(local->unique_objects,
528 llist_sorted_difference_inplace(all_objects, alt->all_objects);
533 static struct pack_list * add_pack(struct packed_git *p)
539 if (!p->pack_local && !(alt_odb || verbose))
543 llist_init(&l.all_objects);
546 base = (void *)p->index_base;
547 while (off <= p->index_size - 3 * 20) {
548 llist_insert_back(l.all_objects, base + off);
551 /* this list will be pruned in cmp_two_packs later */
552 l.unique_objects = llist_copy(l.all_objects);
554 return pack_list_insert(&local_packs, &l);
556 return pack_list_insert(&altodb_packs, &l);
559 static struct pack_list * add_pack_file(char *filename)
561 struct packed_git *p = packed_git;
563 if (strlen(filename) < 40)
564 die("Bad pack filename: %s\n", filename);
567 if (strstr(p->pack_name, filename))
571 die("Filename %s not found in packed_git\n", filename);
574 static void load_all(void)
576 struct packed_git *p = packed_git;
584 int main(int argc, char **argv)
587 struct pack_list *min, *red, *pl;
588 struct llist *ignore;
590 char buf[42]; /* 40 byte sha1 + \n + \0 */
592 setup_git_directory();
594 for (i = 1; i < argc; i++) {
595 const char *arg = argv[i];
596 if(!strcmp(arg, "--")) {
600 if(!strcmp(arg, "--all")) {
604 if(!strcmp(arg, "--verbose")) {
608 if(!strcmp(arg, "--alt-odb")) {
613 usage(pack_redundant_usage);
618 prepare_packed_git();
623 while (*(argv + i) != NULL)
624 add_pack_file(*(argv + i++));
626 if (local_packs == NULL)
627 die("Zero packs found!\n");
633 scan_alt_odb_packs();
635 /* ignore objects given on stdin */
638 while (fgets(buf, sizeof(buf), stdin)) {
640 if (get_sha1_hex(buf, sha1))
641 die("Bad sha1 on stdin: %s", buf);
642 llist_insert_sorted_unique(ignore, sha1, NULL);
645 llist_sorted_difference_inplace(all_objects, ignore);
648 llist_sorted_difference_inplace(pl->unique_objects, ignore);
655 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
656 (unsigned long)pack_list_size(altodb_packs));
657 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
660 fprintf(stderr, "\t%s\n", pl->pack->pack_name);
663 fprintf(stderr, "containing %lu duplicate objects "
664 "with a total size of %lukb.\n",
665 (unsigned long)get_pack_redundancy(min),
666 (unsigned long)pack_set_bytecount(min)/1024);
667 fprintf(stderr, "A total of %lu unique objects were considered.\n",
668 (unsigned long)all_objects->size);
669 fprintf(stderr, "Redundant packs (with indexes):\n");
671 pl = red = pack_list_difference(local_packs, min);
674 sha1_pack_index_name(pl->pack->sha1),
675 pl->pack->pack_name);
679 fprintf(stderr, "%luMB of redundant packs in total.\n",
680 (unsigned long)pack_set_bytecount(red)/(1024*1024));