Merge branch 'jc/fetch-pack-fsck-objects'
[git] / builtin / pack-redundant.c
1 /*
2 *
3 * Copyright 2005, Lukas Sandstrom <lukass@etek.chalmers.se>
4 *
5 * This file is licensed under the GPL v2.
6 *
7 */
8
9 #include "builtin.h"
10
11 #define BLKSIZE 512
12
13 static const char pack_redundant_usage[] =
14 "git pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
15
16 static int load_all_packs, verbose, alt_odb;
17
18 struct llist_item {
19         struct llist_item *next;
20         const unsigned char *sha1;
21 };
22 static struct llist {
23         struct llist_item *front;
24         struct llist_item *back;
25         size_t size;
26 } *all_objects; /* all objects which must be present in local packfiles */
27
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;
34
35 struct pll {
36         struct pll *next;
37         struct pack_list *pl;
38 };
39
40 static struct llist_item *free_nodes;
41
42 static inline void llist_item_put(struct llist_item *item)
43 {
44         item->next = free_nodes;
45         free_nodes = item;
46 }
47
48 static inline struct llist_item *llist_item_get(void)
49 {
50         struct llist_item *new;
51         if ( free_nodes ) {
52                 new = free_nodes;
53                 free_nodes = free_nodes->next;
54         } else {
55                 int i = 1;
56                 new = xmalloc(sizeof(struct llist_item) * BLKSIZE);
57                 for (; i < BLKSIZE; i++)
58                         llist_item_put(&new[i]);
59         }
60         return new;
61 }
62
63 static void llist_free(struct llist *list)
64 {
65         while ((list->back = list->front)) {
66                 list->front = list->front->next;
67                 llist_item_put(list->back);
68         }
69         free(list);
70 }
71
72 static inline void llist_init(struct llist **list)
73 {
74         *list = xmalloc(sizeof(struct llist));
75         (*list)->front = (*list)->back = NULL;
76         (*list)->size = 0;
77 }
78
79 static struct llist * llist_copy(struct llist *list)
80 {
81         struct llist *ret;
82         struct llist_item *new, *old, *prev;
83
84         llist_init(&ret);
85
86         if ((ret->size = list->size) == 0)
87                 return ret;
88
89         new = ret->front = llist_item_get();
90         new->sha1 = list->front->sha1;
91
92         old = list->front->next;
93         while (old) {
94                 prev = new;
95                 new = llist_item_get();
96                 prev->next = new;
97                 new->sha1 = old->sha1;
98                 old = old->next;
99         }
100         new->next = NULL;
101         ret->back = new;
102
103         return ret;
104 }
105
106 static inline struct llist_item *llist_insert(struct llist *list,
107                                               struct llist_item *after,
108                                                const unsigned char *sha1)
109 {
110         struct llist_item *new = llist_item_get();
111         new->sha1 = sha1;
112         new->next = NULL;
113
114         if (after != NULL) {
115                 new->next = after->next;
116                 after->next = new;
117                 if (after == list->back)
118                         list->back = new;
119         } else {/* insert in front */
120                 if (list->size == 0)
121                         list->back = new;
122                 else
123                         new->next = list->front;
124                 list->front = new;
125         }
126         list->size++;
127         return new;
128 }
129
130 static inline struct llist_item *llist_insert_back(struct llist *list,
131                                                    const unsigned char *sha1)
132 {
133         return llist_insert(list, list->back, sha1);
134 }
135
136 static inline struct llist_item *llist_insert_sorted_unique(struct llist *list,
137                         const unsigned char *sha1, struct llist_item *hint)
138 {
139         struct llist_item *prev = NULL, *l;
140
141         l = (hint == NULL) ? list->front : hint;
142         while (l) {
143                 int cmp = hashcmp(l->sha1, sha1);
144                 if (cmp > 0) { /* we insert before this entry */
145                         return llist_insert(list, prev, sha1);
146                 }
147                 if (!cmp) { /* already exists */
148                         return l;
149                 }
150                 prev = l;
151                 l = l->next;
152         }
153         /* insert at the end */
154         return llist_insert_back(list, sha1);
155 }
156
157 /* returns a pointer to an item in front of sha1 */
158 static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
159 {
160         struct llist_item *prev, *l;
161
162 redo_from_start:
163         l = (hint == NULL) ? list->front : hint;
164         prev = NULL;
165         while (l) {
166                 int cmp = hashcmp(l->sha1, sha1);
167                 if (cmp > 0) /* not in list, since sorted */
168                         return prev;
169                 if (!cmp) { /* found */
170                         if (prev == NULL) {
171                                 if (hint != NULL && hint != list->front) {
172                                         /* we don't know the previous element */
173                                         hint = NULL;
174                                         goto redo_from_start;
175                                 }
176                                 list->front = l->next;
177                         } else
178                                 prev->next = l->next;
179                         if (l == list->back)
180                                 list->back = prev;
181                         llist_item_put(l);
182                         list->size--;
183                         return prev;
184                 }
185                 prev = l;
186                 l = l->next;
187         }
188         return prev;
189 }
190
191 /* computes A\B */
192 static void llist_sorted_difference_inplace(struct llist *A,
193                                      struct llist *B)
194 {
195         struct llist_item *hint, *b;
196
197         hint = NULL;
198         b = B->front;
199
200         while (b) {
201                 hint = llist_sorted_remove(A, b->sha1, hint);
202                 b = b->next;
203         }
204 }
205
206 static inline struct pack_list * pack_list_insert(struct pack_list **pl,
207                                            struct pack_list *entry)
208 {
209         struct pack_list *p = xmalloc(sizeof(struct pack_list));
210         memcpy(p, entry, sizeof(struct pack_list));
211         p->next = *pl;
212         *pl = p;
213         return p;
214 }
215
216 static inline size_t pack_list_size(struct pack_list *pl)
217 {
218         size_t ret = 0;
219         while (pl) {
220                 ret++;
221                 pl = pl->next;
222         }
223         return ret;
224 }
225
226 static struct pack_list * pack_list_difference(const struct pack_list *A,
227                                                const struct pack_list *B)
228 {
229         struct pack_list *ret;
230         const struct pack_list *pl;
231
232         if (A == NULL)
233                 return NULL;
234
235         pl = B;
236         while (pl != NULL) {
237                 if (A->pack == pl->pack)
238                         return pack_list_difference(A->next, B);
239                 pl = pl->next;
240         }
241         ret = xmalloc(sizeof(struct pack_list));
242         memcpy(ret, A, sizeof(struct pack_list));
243         ret->next = pack_list_difference(A->next, B);
244         return ret;
245 }
246
247 static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
248 {
249         unsigned long p1_off = 0, p2_off = 0, p1_step, p2_step;
250         const unsigned char *p1_base, *p2_base;
251         struct llist_item *p1_hint = NULL, *p2_hint = NULL;
252
253         p1_base = p1->pack->index_data;
254         p2_base = p2->pack->index_data;
255         p1_base += 256 * 4 + ((p1->pack->index_version < 2) ? 4 : 8);
256         p2_base += 256 * 4 + ((p2->pack->index_version < 2) ? 4 : 8);
257         p1_step = (p1->pack->index_version < 2) ? 24 : 20;
258         p2_step = (p2->pack->index_version < 2) ? 24 : 20;
259
260         while (p1_off < p1->pack->num_objects * p1_step &&
261                p2_off < p2->pack->num_objects * p2_step)
262         {
263                 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
264                 /* cmp ~ p1 - p2 */
265                 if (cmp == 0) {
266                         p1_hint = llist_sorted_remove(p1->unique_objects,
267                                         p1_base + p1_off, p1_hint);
268                         p2_hint = llist_sorted_remove(p2->unique_objects,
269                                         p1_base + p1_off, p2_hint);
270                         p1_off += p1_step;
271                         p2_off += p2_step;
272                         continue;
273                 }
274                 if (cmp < 0) { /* p1 has the object, p2 doesn't */
275                         p1_off += p1_step;
276                 } else { /* p2 has the object, p1 doesn't */
277                         p2_off += p2_step;
278                 }
279         }
280 }
281
282 static void pll_free(struct pll *l)
283 {
284         struct pll *old;
285         struct pack_list *opl;
286
287         while (l) {
288                 old = l;
289                 while (l->pl) {
290                         opl = l->pl;
291                         l->pl = opl->next;
292                         free(opl);
293                 }
294                 l = l->next;
295                 free(old);
296         }
297 }
298
299 /* all the permutations have to be free()d at the same time,
300  * since they refer to each other
301  */
302 static struct pll * get_permutations(struct pack_list *list, int n)
303 {
304         struct pll *subset, *ret = NULL, *new_pll = NULL, *pll;
305
306         if (list == NULL || pack_list_size(list) < n || n == 0)
307                 return NULL;
308
309         if (n == 1) {
310                 while (list) {
311                         new_pll = xmalloc(sizeof(pll));
312                         new_pll->pl = NULL;
313                         pack_list_insert(&new_pll->pl, list);
314                         new_pll->next = ret;
315                         ret = new_pll;
316                         list = list->next;
317                 }
318                 return ret;
319         }
320
321         while (list->next) {
322                 subset = get_permutations(list->next, n - 1);
323                 while (subset) {
324                         new_pll = xmalloc(sizeof(pll));
325                         new_pll->pl = subset->pl;
326                         pack_list_insert(&new_pll->pl, list);
327                         new_pll->next = ret;
328                         ret = new_pll;
329                         subset = subset->next;
330                 }
331                 list = list->next;
332         }
333         return ret;
334 }
335
336 static int is_superset(struct pack_list *pl, struct llist *list)
337 {
338         struct llist *diff;
339
340         diff = llist_copy(list);
341
342         while (pl) {
343                 llist_sorted_difference_inplace(diff, pl->all_objects);
344                 if (diff->size == 0) { /* we're done */
345                         llist_free(diff);
346                         return 1;
347                 }
348                 pl = pl->next;
349         }
350         llist_free(diff);
351         return 0;
352 }
353
354 static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
355 {
356         size_t ret = 0;
357         unsigned long p1_off = 0, p2_off = 0, p1_step, p2_step;
358         const unsigned char *p1_base, *p2_base;
359
360         p1_base = p1->index_data;
361         p2_base = p2->index_data;
362         p1_base += 256 * 4 + ((p1->index_version < 2) ? 4 : 8);
363         p2_base += 256 * 4 + ((p2->index_version < 2) ? 4 : 8);
364         p1_step = (p1->index_version < 2) ? 24 : 20;
365         p2_step = (p2->index_version < 2) ? 24 : 20;
366
367         while (p1_off < p1->num_objects * p1_step &&
368                p2_off < p2->num_objects * p2_step)
369         {
370                 int cmp = hashcmp(p1_base + p1_off, p2_base + p2_off);
371                 /* cmp ~ p1 - p2 */
372                 if (cmp == 0) {
373                         ret++;
374                         p1_off += p1_step;
375                         p2_off += p2_step;
376                         continue;
377                 }
378                 if (cmp < 0) { /* p1 has the object, p2 doesn't */
379                         p1_off += p1_step;
380                 } else { /* p2 has the object, p1 doesn't */
381                         p2_off += p2_step;
382                 }
383         }
384         return ret;
385 }
386
387 /* another O(n^2) function ... */
388 static size_t get_pack_redundancy(struct pack_list *pl)
389 {
390         struct pack_list *subset;
391         size_t ret = 0;
392
393         if (pl == NULL)
394                 return 0;
395
396         while ((subset = pl->next)) {
397                 while (subset) {
398                         ret += sizeof_union(pl->pack, subset->pack);
399                         subset = subset->next;
400                 }
401                 pl = pl->next;
402         }
403         return ret;
404 }
405
406 static inline off_t pack_set_bytecount(struct pack_list *pl)
407 {
408         off_t ret = 0;
409         while (pl) {
410                 ret += pl->pack->pack_size;
411                 ret += pl->pack->index_size;
412                 pl = pl->next;
413         }
414         return ret;
415 }
416
417 static void minimize(struct pack_list **min)
418 {
419         struct pack_list *pl, *unique = NULL,
420                 *non_unique = NULL, *min_perm = NULL;
421         struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
422         struct llist *missing;
423         off_t min_perm_size = 0, perm_size;
424         int n;
425
426         pl = local_packs;
427         while (pl) {
428                 if (pl->unique_objects->size)
429                         pack_list_insert(&unique, pl);
430                 else
431                         pack_list_insert(&non_unique, pl);
432                 pl = pl->next;
433         }
434         /* find out which objects are missing from the set of unique packs */
435         missing = llist_copy(all_objects);
436         pl = unique;
437         while (pl) {
438                 llist_sorted_difference_inplace(missing, pl->all_objects);
439                 pl = pl->next;
440         }
441
442         /* return if there are no objects missing from the unique set */
443         if (missing->size == 0) {
444                 *min = unique;
445                 return;
446         }
447
448         /* find the permutations which contain all missing objects */
449         for (n = 1; n <= pack_list_size(non_unique) && !perm_ok; n++) {
450                 perm_all = perm = get_permutations(non_unique, n);
451                 while (perm) {
452                         if (is_superset(perm->pl, missing)) {
453                                 new_perm = xmalloc(sizeof(struct pll));
454                                 memcpy(new_perm, perm, sizeof(struct pll));
455                                 new_perm->next = perm_ok;
456                                 perm_ok = new_perm;
457                         }
458                         perm = perm->next;
459                 }
460                 if (perm_ok)
461                         break;
462                 pll_free(perm_all);
463         }
464         if (perm_ok == NULL)
465                 die("Internal error: No complete sets found!");
466
467         /* find the permutation with the smallest size */
468         perm = perm_ok;
469         while (perm) {
470                 perm_size = pack_set_bytecount(perm->pl);
471                 if (!min_perm_size || min_perm_size > perm_size) {
472                         min_perm_size = perm_size;
473                         min_perm = perm->pl;
474                 }
475                 perm = perm->next;
476         }
477         *min = min_perm;
478         /* add the unique packs to the list */
479         pl = unique;
480         while (pl) {
481                 pack_list_insert(min, pl);
482                 pl = pl->next;
483         }
484 }
485
486 static void load_all_objects(void)
487 {
488         struct pack_list *pl = local_packs;
489         struct llist_item *hint, *l;
490
491         llist_init(&all_objects);
492
493         while (pl) {
494                 hint = NULL;
495                 l = pl->all_objects->front;
496                 while (l) {
497                         hint = llist_insert_sorted_unique(all_objects,
498                                                           l->sha1, hint);
499                         l = l->next;
500                 }
501                 pl = pl->next;
502         }
503         /* remove objects present in remote packs */
504         pl = altodb_packs;
505         while (pl) {
506                 llist_sorted_difference_inplace(all_objects, pl->all_objects);
507                 pl = pl->next;
508         }
509 }
510
511 /* this scales like O(n^2) */
512 static void cmp_local_packs(void)
513 {
514         struct pack_list *subset, *pl = local_packs;
515
516         while ((subset = pl)) {
517                 while ((subset = subset->next))
518                         cmp_two_packs(pl, subset);
519                 pl = pl->next;
520         }
521 }
522
523 static void scan_alt_odb_packs(void)
524 {
525         struct pack_list *local, *alt;
526
527         alt = altodb_packs;
528         while (alt) {
529                 local = local_packs;
530                 while (local) {
531                         llist_sorted_difference_inplace(local->unique_objects,
532                                                         alt->all_objects);
533                         local = local->next;
534                 }
535                 llist_sorted_difference_inplace(all_objects, alt->all_objects);
536                 alt = alt->next;
537         }
538 }
539
540 static struct pack_list * add_pack(struct packed_git *p)
541 {
542         struct pack_list l;
543         unsigned long off = 0, step;
544         const unsigned char *base;
545
546         if (!p->pack_local && !(alt_odb || verbose))
547                 return NULL;
548
549         l.pack = p;
550         llist_init(&l.all_objects);
551
552         if (open_pack_index(p))
553                 return NULL;
554
555         base = p->index_data;
556         base += 256 * 4 + ((p->index_version < 2) ? 4 : 8);
557         step = (p->index_version < 2) ? 24 : 20;
558         while (off < p->num_objects * step) {
559                 llist_insert_back(l.all_objects, base + off);
560                 off += step;
561         }
562         /* this list will be pruned in cmp_two_packs later */
563         l.unique_objects = llist_copy(l.all_objects);
564         if (p->pack_local)
565                 return pack_list_insert(&local_packs, &l);
566         else
567                 return pack_list_insert(&altodb_packs, &l);
568 }
569
570 static struct pack_list * add_pack_file(const char *filename)
571 {
572         struct packed_git *p = packed_git;
573
574         if (strlen(filename) < 40)
575                 die("Bad pack filename: %s", filename);
576
577         while (p) {
578                 if (strstr(p->pack_name, filename))
579                         return add_pack(p);
580                 p = p->next;
581         }
582         die("Filename %s not found in packed_git", filename);
583 }
584
585 static void load_all(void)
586 {
587         struct packed_git *p = packed_git;
588
589         while (p) {
590                 add_pack(p);
591                 p = p->next;
592         }
593 }
594
595 int cmd_pack_redundant(int argc, const char **argv, const char *prefix)
596 {
597         int i;
598         struct pack_list *min, *red, *pl;
599         struct llist *ignore;
600         unsigned char *sha1;
601         char buf[42]; /* 40 byte sha1 + \n + \0 */
602
603         if (argc == 2 && !strcmp(argv[1], "-h"))
604                 usage(pack_redundant_usage);
605
606         for (i = 1; i < argc; i++) {
607                 const char *arg = argv[i];
608                 if (!strcmp(arg, "--")) {
609                         i++;
610                         break;
611                 }
612                 if (!strcmp(arg, "--all")) {
613                         load_all_packs = 1;
614                         continue;
615                 }
616                 if (!strcmp(arg, "--verbose")) {
617                         verbose = 1;
618                         continue;
619                 }
620                 if (!strcmp(arg, "--alt-odb")) {
621                         alt_odb = 1;
622                         continue;
623                 }
624                 if (*arg == '-')
625                         usage(pack_redundant_usage);
626                 else
627                         break;
628         }
629
630         prepare_packed_git();
631
632         if (load_all_packs)
633                 load_all();
634         else
635                 while (*(argv + i) != NULL)
636                         add_pack_file(*(argv + i++));
637
638         if (local_packs == NULL)
639                 die("Zero packs found!");
640
641         load_all_objects();
642
643         cmp_local_packs();
644         if (alt_odb)
645                 scan_alt_odb_packs();
646
647         /* ignore objects given on stdin */
648         llist_init(&ignore);
649         if (!isatty(0)) {
650                 while (fgets(buf, sizeof(buf), stdin)) {
651                         sha1 = xmalloc(20);
652                         if (get_sha1_hex(buf, sha1))
653                                 die("Bad sha1 on stdin: %s", buf);
654                         llist_insert_sorted_unique(ignore, sha1, NULL);
655                 }
656         }
657         llist_sorted_difference_inplace(all_objects, ignore);
658         pl = local_packs;
659         while (pl) {
660                 llist_sorted_difference_inplace(pl->unique_objects, ignore);
661                 pl = pl->next;
662         }
663
664         minimize(&min);
665
666         if (verbose) {
667                 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
668                         (unsigned long)pack_list_size(altodb_packs));
669                 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
670                 pl = min;
671                 while (pl) {
672                         fprintf(stderr, "\t%s\n", pl->pack->pack_name);
673                         pl = pl->next;
674                 }
675                 fprintf(stderr, "containing %lu duplicate objects "
676                                 "with a total size of %lukb.\n",
677                         (unsigned long)get_pack_redundancy(min),
678                         (unsigned long)pack_set_bytecount(min)/1024);
679                 fprintf(stderr, "A total of %lu unique objects were considered.\n",
680                         (unsigned long)all_objects->size);
681                 fprintf(stderr, "Redundant packs (with indexes):\n");
682         }
683         pl = red = pack_list_difference(local_packs, min);
684         while (pl) {
685                 printf("%s\n%s\n",
686                        sha1_pack_index_name(pl->pack->sha1),
687                        pl->pack->pack_name);
688                 pl = pl->next;
689         }
690         if (verbose)
691                 fprintf(stderr, "%luMB of redundant packs in total.\n",
692                         (unsigned long)pack_set_bytecount(red)/(1024*1024));
693
694         return 0;
695 }