Merge http://www.kernel.org/pub/scm/gitk/gitk
[git] / 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 "cache.h"
10
11 static const char pack_redundant_usage[] =
12 "git-pack-redundant [ --verbose ] [ --alt-odb ] < --all | <.pack filename> ...>";
13
14 static int load_all_packs = 0, verbose = 0, alt_odb = 0;
15
16 struct llist_item {
17         struct llist_item *next;
18         unsigned char *sha1;
19 };
20 static struct llist {
21         struct llist_item *front;
22         struct llist_item *back;
23         size_t size;
24 } *all_objects; /* all objects which must be present in local packfiles */
25
26 static struct pack_list {
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;
32
33 struct pll {
34         struct pll *next;
35         struct pack_list *pl;
36         size_t pl_size;
37 };
38
39 static struct llist_item *free_nodes = NULL;
40
41 static inline struct llist_item *llist_item_get()
42 {
43         struct llist_item *new;
44         if ( free_nodes ) {
45                 new = free_nodes;
46                 free_nodes = free_nodes->next;
47         } else
48                 new = xmalloc(sizeof(struct llist_item));
49
50         return new;
51 }
52
53 static inline void llist_item_put(struct llist_item *item)
54 {
55         item->next = free_nodes;
56         free_nodes = item;
57 }
58
59 static void llist_free(struct llist *list)
60 {
61         while((list->back = list->front)) {
62                 list->front = list->front->next;
63                 llist_item_put(list->back);
64         }
65         free(list);
66 }
67
68 static inline void llist_init(struct llist **list)
69 {
70         *list = xmalloc(sizeof(struct llist));
71         (*list)->front = (*list)->back = NULL;
72         (*list)->size = 0;
73 }
74
75 static struct llist * llist_copy(struct llist *list)
76 {
77         struct llist *ret;
78         struct llist_item *new, *old, *prev;
79         
80         llist_init(&ret);
81
82         if ((ret->size = list->size) == 0)
83                 return ret;
84
85         new = ret->front = llist_item_get();
86         new->sha1 = list->front->sha1;
87
88         old = list->front->next;
89         while (old) {
90                 prev = new;
91                 new = llist_item_get();
92                 prev->next = new;
93                 new->sha1 = old->sha1;
94                 old = old->next;
95         }
96         new->next = NULL;
97         ret->back = new;
98         
99         return ret;
100 }
101
102 static inline struct llist_item * llist_insert(struct llist *list,
103                                                struct llist_item *after,
104                                                unsigned char *sha1)
105 {
106         struct llist_item *new = llist_item_get();
107         new->sha1 = sha1;
108         new->next = NULL;
109
110         if (after != NULL) {
111                 new->next = after->next;
112                 after->next = new;
113                 if (after == list->back)
114                         list->back = new;
115         } else {/* insert in front */
116                 if (list->size == 0)
117                         list->back = new;
118                 else
119                         new->next = list->front;
120                 list->front = new;
121         }
122         list->size++;
123         return new;
124 }
125
126 static inline struct llist_item *llist_insert_back(struct llist *list, unsigned char *sha1)
127 {
128         return llist_insert(list, list->back, sha1);
129 }
130
131 static inline struct llist_item *llist_insert_sorted_unique(struct llist *list, unsigned char *sha1, struct llist_item *hint)
132 {
133         struct llist_item *prev = NULL, *l;
134
135         l = (hint == NULL) ? list->front : hint;
136         while (l) {
137                 int cmp = memcmp(l->sha1, sha1, 20);
138                 if (cmp > 0) { /* we insert before this entry */
139                         return llist_insert(list, prev, sha1);
140                 }
141                 if(!cmp) { /* already exists */
142                         return l;
143                 }
144                 prev = l;
145                 l = l->next;
146         }
147         /* insert at the end */
148         return llist_insert_back(list, sha1);
149 }
150
151 /* returns a pointer to an item in front of sha1 */
152 static inline struct llist_item * llist_sorted_remove(struct llist *list, const unsigned char *sha1, struct llist_item *hint)
153 {
154         struct llist_item *prev, *l;
155
156 redo_from_start:
157         l = (hint == NULL) ? list->front : hint;
158         prev = NULL;
159         while (l) {
160                 int cmp = memcmp(l->sha1, sha1, 20);
161                 if (cmp > 0) /* not in list, since sorted */
162                         return prev;
163                 if(!cmp) { /* found */
164                         if (prev == NULL) {
165                                 if (hint != NULL && hint != list->front) {
166                                         /* we don't know the previous element */
167                                         hint = NULL;
168                                         goto redo_from_start;
169                                 }
170                                 list->front = l->next;
171                         } else
172                                 prev->next = l->next;
173                         if (l == list->back)
174                                 list->back = prev;
175                         llist_item_put(l);
176                         list->size--;
177                         return prev;
178                 }
179                 prev = l;
180                 l = l->next;
181         }
182         return prev;
183 }
184
185 /* computes A\B */
186 static void llist_sorted_difference_inplace(struct llist *A,
187                                      struct llist *B)
188 {
189         struct llist_item *hint, *b;
190
191         hint = NULL;
192         b = B->front;
193
194         while (b) {
195                 hint = llist_sorted_remove(A, b->sha1, hint);
196                 b = b->next;
197         }
198 }
199
200 static inline struct pack_list * pack_list_insert(struct pack_list **pl,
201                                            struct pack_list *entry)
202 {
203         struct pack_list *p = xmalloc(sizeof(struct pack_list));
204         memcpy(p, entry, sizeof(struct pack_list));
205         p->next = *pl;
206         *pl = p;
207         return p;
208 }
209
210 static inline size_t pack_list_size(struct pack_list *pl)
211 {
212         size_t ret = 0;
213         while(pl) {
214                 ret++;
215                 pl = pl->next;
216         }
217         return ret;
218 }
219
220 static struct pack_list * pack_list_difference(const struct pack_list *A,
221                                                const struct pack_list *B)
222 {
223         struct pack_list *ret;
224         const struct pack_list *pl;
225
226         if (A == NULL)
227                 return NULL;
228
229         pl = B;
230         while (pl != NULL) {
231                 if (A->pack == pl->pack)
232                         return pack_list_difference(A->next, B);
233                 pl = pl->next;
234         }
235         ret = xmalloc(sizeof(struct pack_list));
236         memcpy(ret, A, sizeof(struct pack_list));
237         ret->next = pack_list_difference(A->next, B);
238         return ret;
239 }
240
241 static void cmp_two_packs(struct pack_list *p1, struct pack_list *p2)
242 {
243         int p1_off, p2_off;
244         void *p1_base, *p2_base;
245         struct llist_item *p1_hint = NULL, *p2_hint = NULL;
246         
247         p1_off = p2_off = 256 * 4 + 4;
248         p1_base = (void *)p1->pack->index_base;
249         p2_base = (void *)p2->pack->index_base;
250
251         while (p1_off <= p1->pack->index_size - 3 * 20 &&
252                p2_off <= p2->pack->index_size - 3 * 20)
253         {
254                 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
255                 /* cmp ~ p1 - p2 */
256                 if (cmp == 0) {
257                         p1_hint = llist_sorted_remove(p1->unique_objects,
258                                         p1_base + p1_off, p1_hint);
259                         p2_hint = llist_sorted_remove(p2->unique_objects,
260                                         p1_base + p1_off, p2_hint);
261                         p1_off+=24;
262                         p2_off+=24;
263                         continue;
264                 }
265                 if (cmp < 0) { /* p1 has the object, p2 doesn't */
266                         p1_off+=24;
267                 } else { /* p2 has the object, p1 doesn't */
268                         p2_off+=24;
269                 }
270         }
271 }
272
273 static void pll_insert(struct pll **pll, struct pll **hint_table)
274 {
275         struct pll *prev;
276         int i = (*pll)->pl_size - 1;
277
278         if (hint_table[i] == NULL) {
279                 hint_table[i--] = *pll;
280                 for (; i >= 0; --i) {
281                         if (hint_table[i] != NULL)
282                                 break;
283                 }
284                 if (hint_table[i] == NULL) /* no elements in list */
285                         die("Why did this happen?");
286         }
287
288         prev = hint_table[i];
289         while (prev->next && prev->next->pl_size < (*pll)->pl_size)
290                 prev = prev->next;
291
292         (*pll)->next = prev->next;
293         prev->next = *pll;
294 }
295
296 /* all the permutations have to be free()d at the same time,
297  * since they refer to each other
298  */
299 static struct pll * get_all_permutations(struct pack_list *list)
300 {
301         struct pll *subset, *pll, *new_pll = NULL; /*silence warning*/
302         static struct pll **hint = NULL;
303         if (hint == NULL)
304                 hint = xcalloc(pack_list_size(list), sizeof(struct pll *));
305                 
306         if (list == NULL)
307                 return NULL;
308
309         if (list->next == NULL) {
310                 new_pll = xmalloc(sizeof(struct pll));
311                 hint[0] = new_pll;
312                 new_pll->next = NULL;
313                 new_pll->pl = list;
314                 new_pll->pl_size = 1;
315                 return new_pll;
316         }
317
318         pll = subset = get_all_permutations(list->next);
319         while (pll) {
320                 if (pll->pl->pack == list->pack) {
321                         pll = pll->next;
322                         continue;
323                 }
324                 new_pll = xmalloc(sizeof(struct pll));
325
326                 new_pll->pl = xmalloc(sizeof(struct pack_list));
327                 memcpy(new_pll->pl, list, sizeof(struct pack_list));
328                 new_pll->pl->next = pll->pl;
329                 new_pll->pl_size = pll->pl_size + 1;
330                 
331                 pll_insert(&new_pll, hint);
332
333                 pll = pll->next;
334         }
335         /* add ourself */
336         new_pll = xmalloc(sizeof(struct pll));
337         new_pll->pl = xmalloc(sizeof(struct pack_list));
338         memcpy(new_pll->pl, list, sizeof(struct pack_list));
339         new_pll->pl->next = NULL;
340         new_pll->pl_size = 1;
341         pll_insert(&new_pll, hint);
342
343         return hint[0];
344 }
345
346 static int is_superset(struct pack_list *pl, struct llist *list)
347 {
348         struct llist *diff;
349
350         diff = llist_copy(list);
351
352         while (pl) {
353                 llist_sorted_difference_inplace(diff, pl->all_objects);
354                 if (diff->size == 0) { /* we're done */
355                         llist_free(diff);
356                         return 1;
357                 }
358                 pl = pl->next;
359         }
360         llist_free(diff);
361         return 0;
362 }
363
364 static size_t sizeof_union(struct packed_git *p1, struct packed_git *p2)
365 {
366         size_t ret = 0;
367         int p1_off, p2_off;
368         void *p1_base, *p2_base;
369
370         p1_off = p2_off = 256 * 4 + 4;
371         p1_base = (void *)p1->index_base;
372         p2_base = (void *)p2->index_base;
373
374         while (p1_off <= p1->index_size - 3 * 20 &&
375                p2_off <= p2->index_size - 3 * 20)
376         {
377                 int cmp = memcmp(p1_base + p1_off, p2_base + p2_off, 20);
378                 /* cmp ~ p1 - p2 */
379                 if (cmp == 0) {
380                         ret++;
381                         p1_off+=24;
382                         p2_off+=24;
383                         continue;
384                 }
385                 if (cmp < 0) { /* p1 has the object, p2 doesn't */
386                         p1_off+=24;
387                 } else { /* p2 has the object, p1 doesn't */
388                         p2_off+=24;
389                 }
390         }
391         return ret;
392 }
393
394 /* another O(n^2) function ... */
395 static size_t get_pack_redundancy(struct pack_list *pl)
396 {
397         struct pack_list *subset;
398         size_t ret = 0;
399
400         if (pl == NULL)
401                 return 0;
402
403         while ((subset = pl->next)) {
404                 while(subset) {
405                         ret += sizeof_union(pl->pack, subset->pack);
406                         subset = subset->next;
407                 }
408                 pl = pl->next;
409         }
410         return ret;
411 }
412
413 static inline size_t pack_set_bytecount(struct pack_list *pl)
414 {
415         size_t ret = 0;
416         while (pl) {
417                 ret += pl->pack->pack_size;
418                 ret += pl->pack->index_size;
419                 pl = pl->next;
420         }
421         return ret;
422 }
423
424 static void minimize(struct pack_list **min)
425 {
426         struct pack_list *pl, *unique = NULL,
427                 *non_unique = NULL, *min_perm = NULL;
428         struct pll *perm, *perm_all, *perm_ok = NULL, *new_perm;
429         struct llist *missing;
430         size_t min_perm_size = (size_t)-1, perm_size;
431
432         pl = local_packs;
433         while (pl) {
434                 if(pl->unique_objects->size)
435                         pack_list_insert(&unique, pl);
436                 else
437                         pack_list_insert(&non_unique, pl);
438                 pl = pl->next;
439         }
440         /* find out which objects are missing from the set of unique packs */
441         missing = llist_copy(all_objects);
442         pl = unique;
443         while (pl) {
444                 llist_sorted_difference_inplace(missing,
445                                                 pl->all_objects);
446                 pl = pl->next;
447         }
448
449         /* return if there are no objects missing from the unique set */
450         if (missing->size == 0) {
451                 *min = unique;
452                 return;
453         }
454
455         /* find the permutations which contain all missing objects */
456         perm_all = perm = get_all_permutations(non_unique);
457         while (perm) {
458                 if (perm_ok && perm->pl_size > perm_ok->pl_size)
459                         break; /* ignore all larger permutations */
460                 if (is_superset(perm->pl, missing)) {
461                         new_perm = xmalloc(sizeof(struct pll));
462                         memcpy(new_perm, perm, sizeof(struct pll));
463                         new_perm->next = perm_ok;
464                         perm_ok = new_perm;
465                 }
466                 perm = perm->next;
467         }
468         
469         if (perm_ok == NULL)
470                 die("Internal error: No complete sets found!\n");
471
472         /* find the permutation with the smallest size */
473         perm = perm_ok;
474         while (perm) {
475                 perm_size = pack_set_bytecount(perm->pl);
476                 if (min_perm_size > perm_size) {
477                         min_perm_size = perm_size;
478                         min_perm = perm->pl;
479                 }
480                 perm = perm->next;
481         }
482         *min = min_perm;
483         /* add the unique packs to the list */
484         pl = unique;
485         while(pl) {
486                 pack_list_insert(min, pl);
487                 pl = pl->next;
488         }
489 }
490
491 static void load_all_objects(void)
492 {
493         struct pack_list *pl = local_packs;
494         struct llist_item *hint, *l;
495
496         llist_init(&all_objects);
497
498         while (pl) {
499                 hint = NULL;
500                 l = pl->all_objects->front;
501                 while (l) {
502                         hint = llist_insert_sorted_unique(all_objects,
503                                                           l->sha1, hint);
504                         l = l->next;
505                 }
506                 pl = pl->next;
507         }
508         /* remove objects present in remote packs */
509         pl = altodb_packs;
510         while (pl) {
511                 llist_sorted_difference_inplace(all_objects, pl->all_objects);
512                 pl = pl->next;
513         }
514 }
515
516 /* this scales like O(n^2) */
517 static void cmp_local_packs(void)
518 {
519         struct pack_list *subset, *pl = local_packs;
520
521         while ((subset = pl)) {
522                 while((subset = subset->next))
523                         cmp_two_packs(pl, subset);
524                 pl = pl->next;
525         }
526 }
527
528 static void scan_alt_odb_packs(void)
529 {
530         struct pack_list *local, *alt;
531
532         alt = altodb_packs;
533         while (alt) {
534                 local = local_packs;
535                 while (local) {
536                         llist_sorted_difference_inplace(local->unique_objects,
537                                                         alt->all_objects);
538                         local = local->next;
539                 }
540                 alt = alt->next;
541         }
542 }
543
544 static struct pack_list * add_pack(struct packed_git *p)
545 {
546         struct pack_list l;
547         size_t off;
548         void *base;
549
550         if (!p->pack_local && !(alt_odb || verbose))
551                 return NULL;
552
553         l.pack = p;
554         llist_init(&l.all_objects);
555
556         off = 256 * 4 + 4;
557         base = (void *)p->index_base;
558         while (off <= p->index_size - 3 * 20) {
559                 llist_insert_back(l.all_objects, base + off);
560                 off += 24;
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(char *filename)
571 {
572         struct packed_git *p = packed_git;
573
574         if (strlen(filename) < 40)
575                 die("Bad pack filename: %s\n", 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\n", 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 main(int argc, char **argv)
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         for (i = 1; i < argc; i++) {
604                 const char *arg = argv[i];
605                 if(!strcmp(arg, "--")) {
606                         i++;
607                         break;
608                 }
609                 if(!strcmp(arg, "--all")) {
610                         load_all_packs = 1;
611                         continue;
612                 }
613                 if(!strcmp(arg, "--verbose")) {
614                         verbose = 1;
615                         continue;
616                 }
617                 if(!strcmp(arg, "--alt-odb")) {
618                         alt_odb = 1;
619                         continue;
620                 }
621                 if(*arg == '-')
622                         usage(pack_redundant_usage);
623                 else
624                         break;
625         }
626
627         prepare_packed_git();
628
629         if (load_all_packs)
630                 load_all();
631         else
632                 while (*(argv + i) != NULL)
633                         add_pack_file(*(argv + i++));
634
635         if (local_packs == NULL)
636                 die("Zero packs found!\n");
637
638         load_all_objects();
639
640         cmp_local_packs();
641         if (alt_odb)
642                 scan_alt_odb_packs();
643
644         /* ignore objects given on stdin */
645         llist_init(&ignore);
646         if (!isatty(0)) {
647                 while (fgets(buf, sizeof(buf), stdin)) {
648                         sha1 = xmalloc(20);
649                         if (get_sha1_hex(buf, sha1))
650                                 die("Bad sha1 on stdin: %s", buf);
651                         llist_insert_sorted_unique(ignore, sha1, NULL);
652                 }
653         }
654         llist_sorted_difference_inplace(all_objects, ignore);
655         pl = local_packs;
656         while (pl) {
657                 llist_sorted_difference_inplace(pl->unique_objects, ignore);
658                 pl = pl->next;
659         }
660
661         minimize(&min);
662
663         if (verbose) {
664                 fprintf(stderr, "There are %lu packs available in alt-odbs.\n",
665                         (unsigned long)pack_list_size(altodb_packs));
666                 fprintf(stderr, "The smallest (bytewise) set of packs is:\n");
667                 pl = min;
668                 while (pl) {
669                         fprintf(stderr, "\t%s\n", pl->pack->pack_name);
670                         pl = pl->next;
671                 }
672                 fprintf(stderr, "containing %lu duplicate objects "
673                                 "with a total size of %lukb.\n",
674                         (unsigned long)get_pack_redundancy(min),
675                         (unsigned long)pack_set_bytecount(min)/1024);
676                 fprintf(stderr, "A total of %lu unique objects were considered.\n",
677                         (unsigned long)all_objects->size);
678                 fprintf(stderr, "Redundant packs (with indexes):\n");
679         }
680         pl = red = pack_list_difference(local_packs, min);
681         while (pl) {
682                 printf("%s\n%s\n",
683                        sha1_pack_index_name(pl->pack->sha1),
684                        pl->pack->pack_name);
685                 pl = pl->next;
686         }
687         if (verbose)
688                 fprintf(stderr, "%luMB of redundant packs in total.\n",
689                         (unsigned long)pack_set_bytecount(red)/(1024*1024));
690
691         return 0;
692 }