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, verbose, alt_odb;
19 struct llist_item *next;
20 const unsigned char *sha1;
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;
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,
109 const unsigned char *sha1)
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,
132 const unsigned char *sha1)
134 return llist_insert(list, list->back, sha1);
137 static inline struct llist_item *llist_insert_sorted_unique(struct llist *list,
138 const unsigned char *sha1, struct llist_item *hint)
140 struct llist_item *prev = NULL, *l;
142 l = (hint == NULL) ? list->front : hint;
144 int cmp = hashcmp(l->sha1, sha1);
145 if (cmp > 0) { /* we insert before this entry */
146 return llist_insert(list, prev, sha1);
148 if(!cmp) { /* already exists */
154 /* insert at the end */
155 return llist_insert_back(list, sha1);
158 /* returns a pointer to an item in front of sha1 */
159 static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
161 struct llist_item *prev, *l;
164 l = (hint == NULL) ? list->front : hint;
167 int cmp = hashcmp(l->sha1, sha1);
168 if (cmp > 0) /* not in list, since sorted */
170 if(!cmp) { /* found */
172 if (hint != NULL && hint != list->front) {
173 /* we don't know the previous element */
175 goto redo_from_start;
177 list->front = l->next;
179 prev->next = l->next;
193 static void llist_sorted_difference_inplace(struct llist *A,
196 struct llist_item *hint, *b;
202 hint = llist_sorted_remove(A, b->sha1, hint);
207 static inline struct pack_list * pack_list_insert(struct pack_list **pl,
208 struct pack_list *entry)
210 struct pack_list *p = xmalloc(sizeof(struct pack_list));
211 memcpy(p, entry, sizeof(struct pack_list));
217 static inline size_t pack_list_size(struct pack_list *pl)
227 static struct pack_list * pack_list_difference(const struct pack_list *A,
228 const struct pack_list *B)
230 struct pack_list *ret;
231 const struct pack_list *pl;
238 if (A->pack == pl->pack)
239 return pack_list_difference(A->next, B);
242 ret = xmalloc(sizeof(struct pack_list));
243 memcpy(ret, A, sizeof(struct pack_list));
244 ret->next = pack_list_difference(A->next, B);
248 static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
251 const unsigned char *p1_base, *p2_base;
252 struct llist_item *p1_hint = NULL, *p2_hint = NULL;
254 p1_off = p2_off = 256 * 4 + 4;
255 p1_base = p1->pack->index_data;
256 p2_base = p2->pack->index_data;
258 while (p1_off <= p1->pack->index_size - 3 * 20 &&
259 p2_off <= p2->pack->index_size - 3 * 20)
261 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
264 p1_hint = llist_sorted_remove(p1->unique_objects,
265 p1_base + p1_off, p1_hint);
266 p2_hint = llist_sorted_remove(p2->unique_objects,
267 p1_base + p1_off, p2_hint);
272 if (cmp < 0) { /* p1 has the object, p2 doesn't */
274 } else { /* p2 has the object, p1 doesn't */
280 static void pll_free(struct pll *l)
283 struct pack_list *opl;
297 /* all the permutations have to be free()d at the same time,
298 * since they refer to each other
300 static struct pll * get_permutations(struct pack_list *list, int n)
302 struct pll *subset, *ret = NULL, *new_pll = NULL, *pll;
304 if (list == NULL || pack_list_size(list) < n || n == 0)
309 new_pll = xmalloc(sizeof(pll));
311 pack_list_insert(&new_pll->pl, list);
320 subset = get_permutations(list->next, n - 1);
322 new_pll = xmalloc(sizeof(pll));
323 new_pll->pl = subset->pl;
324 pack_list_insert(&new_pll->pl, list);
327 subset = subset->next;
334 static int is_superset(struct pack_list *pl, struct llist *list)
338 diff = llist_copy(list);
341 llist_sorted_difference_inplace(diff, pl->all_objects);
342 if (diff->size == 0) { /* we're done */
352 static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
356 const unsigned char *p1_base, *p2_base;
358 p1_off = p2_off = 256 * 4 + 4;
359 p1_base = p1->index_data;
360 p2_base = p2->index_data;
362 while (p1_off <= p1->index_size - 3 * 20 &&
363 p2_off <= p2->index_size - 3 * 20)
365 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
373 if (cmp < 0) { /* p1 has the object, p2 doesn't */
375 } else { /* p2 has the object, p1 doesn't */
382 /* another O(n^2) function ... */
383 static size_t get_pack_redundancy(struct pack_list *pl)
385 struct pack_list *subset;
391 while ((subset = pl->next)) {
393 ret += sizeof_union(pl->pack, subset->pack);
394 subset = subset->next;
401 static inline off_t pack_set_bytecount(struct pack_list *pl)
405 ret += pl->pack->pack_size;
406 ret += pl->pack->index_size;
412 static void minimize(struct pack_list **min)
414 struct pack_list *pl, *unique = NULL,
415 *non_unique = NULL, *min_perm = NULL;
416 struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
417 struct llist *missing;
418 off_t min_perm_size = 0, perm_size;
423 if(pl->unique_objects->size)
424 pack_list_insert(&unique, pl);
426 pack_list_insert(&non_unique, pl);
429 /* find out which objects are missing from the set of unique packs */
430 missing = llist_copy(all_objects);
433 llist_sorted_difference_inplace(missing, pl->all_objects);
437 /* return if there are no objects missing from the unique set */
438 if (missing->size == 0) {
443 /* find the permutations which contain all missing objects */
444 for (n = 1; n <= pack_list_size(non_unique) && !perm_ok; n++) {
445 perm_all = perm = get_permutations(non_unique, n);
447 if (is_superset(perm->pl, missing)) {
448 new_perm = xmalloc(sizeof(struct pll));
449 memcpy(new_perm, perm, sizeof(struct pll));
450 new_perm->next = perm_ok;
460 die("Internal error: No complete sets found!\n");
462 /* find the permutation with the smallest size */
465 perm_size = pack_set_bytecount(perm->pl);
466 if (!min_perm_size || min_perm_size > perm_size) {
467 min_perm_size = perm_size;
473 /* add the unique packs to the list */
476 pack_list_insert(min, pl);
481 static void load_all_objects(void)
483 struct pack_list *pl = local_packs;
484 struct llist_item *hint, *l;
486 llist_init(&all_objects);
490 l = pl->all_objects->front;
492 hint = llist_insert_sorted_unique(all_objects,
498 /* remove objects present in remote packs */
501 llist_sorted_difference_inplace(all_objects, pl->all_objects);
506 /* this scales like O(n^2) */
507 static void cmp_local_packs(void)
509 struct pack_list *subset, *pl = local_packs;
511 while ((subset = pl)) {
512 while((subset = subset->next))
513 cmp_two_packs(pl, subset);
518 static void scan_alt_odb_packs(void)
520 struct pack_list *local, *alt;
526 llist_sorted_difference_inplace(local->unique_objects,
530 llist_sorted_difference_inplace(all_objects, alt->all_objects);
535 static struct pack_list * add_pack(struct packed_git *p)
539 const unsigned char *base;
541 if (!p->pack_local && !(alt_odb || verbose))
545 llist_init(&l.all_objects);
548 base = p->index_data;
549 while (off <= p->index_size - 3 * 20) {
550 llist_insert_back(l.all_objects, base + off);
553 /* this list will be pruned in cmp_two_packs later */
554 l.unique_objects = llist_copy(l.all_objects);
556 return pack_list_insert(&local_packs, &l);
558 return pack_list_insert(&altodb_packs, &l);
561 static struct pack_list * add_pack_file(char *filename)
563 struct packed_git *p = packed_git;
565 if (strlen(filename) < 40)
566 die("Bad pack filename: %s\n", filename);
569 if (strstr(p->pack_name, filename))
573 die("Filename %s not found in packed_git\n", filename);
576 static void load_all(void)
578 struct packed_git *p = packed_git;
586 int main(int argc, char **argv)
589 struct pack_list *min, *red, *pl;
590 struct llist *ignore;
592 char buf[42]; /* 40 byte sha1 + \n + \0 */
594 setup_git_directory();
596 for (i = 1; i < argc; i++) {
597 const char *arg = argv[i];
598 if(!strcmp(arg, "--")) {
602 if(!strcmp(arg, "--all")) {
606 if(!strcmp(arg, "--verbose")) {
610 if(!strcmp(arg, "--alt-odb")) {
615 usage(pack_redundant_usage);
620 prepare_packed_git();
625 while (*(argv + i) != NULL)
626 add_pack_file(*(argv + i++));
628 if (local_packs == NULL)
629 die("Zero packs found!\n");
635 scan_alt_odb_packs();
637 /* ignore objects given on stdin */
640 while (fgets(buf, sizeof(buf), stdin)) {
642 if (get_sha1_hex(buf, sha1))
643 die("Bad sha1 on stdin: %s", buf);
644 llist_insert_sorted_unique(ignore, sha1, NULL);
647 llist_sorted_difference_inplace(all_objects, ignore);
650 llist_sorted_difference_inplace(pl->unique_objects, ignore);
657 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
658 (unsigned long)pack_list_size(altodb_packs));
659 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
662 fprintf(stderr, "\t%s\n", pl->pack->pack_name);
665 fprintf(stderr, "containing %lu duplicate objects "
666 "with a total size of %lukb.\n",
667 (unsigned long)get_pack_redundancy(min),
668 (unsigned long)pack_set_bytecount(min)/1024);
669 fprintf(stderr, "A total of %lu unique objects were considered.\n",
670 (unsigned long)all_objects->size);
671 fprintf(stderr, "Redundant packs (with indexes):\n");
673 pl = red = pack_list_difference(local_packs, min);
676 sha1_pack_index_name(pl->pack->sha1),
677 pl->pack->pack_name);
681 fprintf(stderr, "%luMB of redundant packs in total.\n",
682 (unsigned long)pack_set_bytecount(red)/(1024*1024));