upload-pack: prepare for sideband message support.
[git/git.git] / pack-objects.c
1 #include "cache.h"
2 #include "object.h"
3 #include "blob.h"
4 #include "commit.h"
5 #include "tag.h"
6 #include "tree.h"
7 #include "delta.h"
8 #include "pack.h"
9 #include "csum-file.h"
10 #include "tree-walk.h"
11 #include <sys/time.h>
12 #include <signal.h>
13
14 static const char pack_usage[] = "git-pack-objects [-q] [--no-reuse-delta] [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
15
16 struct object_entry {
17 unsigned char sha1[20];
18 unsigned long size; /* uncompressed size */
19 unsigned long offset; /* offset into the final pack file;
20 * nonzero if already written.
21 */
22 unsigned int depth; /* delta depth */
23 unsigned int delta_limit; /* base adjustment for in-pack delta */
24 unsigned int hash; /* name hint hash */
25 enum object_type type;
26 enum object_type in_pack_type; /* could be delta */
27 unsigned long delta_size; /* delta data size (uncompressed) */
28 struct object_entry *delta; /* delta base object */
29 struct packed_git *in_pack; /* already in pack */
30 unsigned int in_pack_offset;
31 struct object_entry *delta_child; /* delitified objects who bases me */
32 struct object_entry *delta_sibling; /* other deltified objects who
33 * uses the same base as me
34 */
35 int preferred_base; /* we do not pack this, but is encouraged to
36 * be used as the base objectto delta huge
37 * objects against.
38 */
39 };
40
41 /*
42 * Objects we are going to pack are colected in objects array (dynamically
43 * expanded). nr_objects & nr_alloc controls this array. They are stored
44 * in the order we see -- typically rev-list --objects order that gives us
45 * nice "minimum seek" order.
46 *
47 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
48 * elements in the objects array. The former is used to build the pack
49 * index (lists object names in the ascending order to help offset lookup),
50 * and the latter is used to group similar things together by try_delta()
51 * heuristics.
52 */
53
54 static unsigned char object_list_sha1[20];
55 static int non_empty = 0;
56 static int no_reuse_delta = 0;
57 static int local = 0;
58 static int incremental = 0;
59 static struct object_entry **sorted_by_sha, **sorted_by_type;
60 static struct object_entry *objects = NULL;
61 static int nr_objects = 0, nr_alloc = 0, nr_result = 0;
62 static const char *base_name;
63 static unsigned char pack_file_sha1[20];
64 static int progress = 1;
65 static volatile sig_atomic_t progress_update = 0;
66
67 /*
68 * The object names in objects array are hashed with this hashtable,
69 * to help looking up the entry by object name. Binary search from
70 * sorted_by_sha is also possible but this was easier to code and faster.
71 * This hashtable is built after all the objects are seen.
72 */
73 static int *object_ix = NULL;
74 static int object_ix_hashsz = 0;
75
76 /*
77 * Pack index for existing packs give us easy access to the offsets into
78 * corresponding pack file where each object's data starts, but the entries
79 * do not store the size of the compressed representation (uncompressed
80 * size is easily available by examining the pack entry header). We build
81 * a hashtable of existing packs (pack_revindex), and keep reverse index
82 * here -- pack index file is sorted by object name mapping to offset; this
83 * pack_revindex[].revindex array is an ordered list of offsets, so if you
84 * know the offset of an object, next offset is where its packed
85 * representation ends.
86 */
87 struct pack_revindex {
88 struct packed_git *p;
89 unsigned long *revindex;
90 } *pack_revindex = NULL;
91 static int pack_revindex_hashsz = 0;
92
93 /*
94 * stats
95 */
96 static int written = 0;
97 static int written_delta = 0;
98 static int reused = 0;
99 static int reused_delta = 0;
100
101 static int pack_revindex_ix(struct packed_git *p)
102 {
103 unsigned long ui = (unsigned long)p;
104 int i;
105
106 ui = ui ^ (ui >> 16); /* defeat structure alignment */
107 i = (int)(ui % pack_revindex_hashsz);
108 while (pack_revindex[i].p) {
109 if (pack_revindex[i].p == p)
110 return i;
111 if (++i == pack_revindex_hashsz)
112 i = 0;
113 }
114 return -1 - i;
115 }
116
117 static void prepare_pack_ix(void)
118 {
119 int num;
120 struct packed_git *p;
121 for (num = 0, p = packed_git; p; p = p->next)
122 num++;
123 if (!num)
124 return;
125 pack_revindex_hashsz = num * 11;
126 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
127 for (p = packed_git; p; p = p->next) {
128 num = pack_revindex_ix(p);
129 num = - 1 - num;
130 pack_revindex[num].p = p;
131 }
132 /* revindex elements are lazily initialized */
133 }
134
135 static int cmp_offset(const void *a_, const void *b_)
136 {
137 unsigned long a = *(unsigned long *) a_;
138 unsigned long b = *(unsigned long *) b_;
139 if (a < b)
140 return -1;
141 else if (a == b)
142 return 0;
143 else
144 return 1;
145 }
146
147 /*
148 * Ordered list of offsets of objects in the pack.
149 */
150 static void prepare_pack_revindex(struct pack_revindex *rix)
151 {
152 struct packed_git *p = rix->p;
153 int num_ent = num_packed_objects(p);
154 int i;
155 void *index = p->index_base + 256;
156
157 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
158 for (i = 0; i < num_ent; i++) {
159 unsigned int hl = *((unsigned int *)(index + 24 * i));
160 rix->revindex[i] = ntohl(hl);
161 }
162 /* This knows the pack format -- the 20-byte trailer
163 * follows immediately after the last object data.
164 */
165 rix->revindex[num_ent] = p->pack_size - 20;
166 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
167 }
168
169 static unsigned long find_packed_object_size(struct packed_git *p,
170 unsigned long ofs)
171 {
172 int num;
173 int lo, hi;
174 struct pack_revindex *rix;
175 unsigned long *revindex;
176 num = pack_revindex_ix(p);
177 if (num < 0)
178 die("internal error: pack revindex uninitialized");
179 rix = &pack_revindex[num];
180 if (!rix->revindex)
181 prepare_pack_revindex(rix);
182 revindex = rix->revindex;
183 lo = 0;
184 hi = num_packed_objects(p) + 1;
185 do {
186 int mi = (lo + hi) / 2;
187 if (revindex[mi] == ofs) {
188 return revindex[mi+1] - ofs;
189 }
190 else if (ofs < revindex[mi])
191 hi = mi;
192 else
193 lo = mi + 1;
194 } while (lo < hi);
195 die("internal error: pack revindex corrupt");
196 }
197
198 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
199 {
200 unsigned long othersize, delta_size;
201 char type[10];
202 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
203 void *delta_buf;
204
205 if (!otherbuf)
206 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
207 delta_buf = diff_delta(otherbuf, othersize,
208 buf, size, &delta_size, 0);
209 if (!delta_buf || delta_size != entry->delta_size)
210 die("delta size changed");
211 free(buf);
212 free(otherbuf);
213 return delta_buf;
214 }
215
216 /*
217 * The per-object header is a pretty dense thing, which is
218 * - first byte: low four bits are "size", then three bits of "type",
219 * and the high bit is "size continues".
220 * - each byte afterwards: low seven bits are size continuation,
221 * with the high bit being "size continues"
222 */
223 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
224 {
225 int n = 1;
226 unsigned char c;
227
228 if (type < OBJ_COMMIT || type > OBJ_DELTA)
229 die("bad type %d", type);
230
231 c = (type << 4) | (size & 15);
232 size >>= 4;
233 while (size) {
234 *hdr++ = c | 0x80;
235 c = size & 0x7f;
236 size >>= 7;
237 n++;
238 }
239 *hdr = c;
240 return n;
241 }
242
243 static unsigned long write_object(struct sha1file *f,
244 struct object_entry *entry)
245 {
246 unsigned long size;
247 char type[10];
248 void *buf;
249 unsigned char header[10];
250 unsigned hdrlen, datalen;
251 enum object_type obj_type;
252 int to_reuse = 0;
253
254 if (entry->preferred_base)
255 return 0;
256
257 obj_type = entry->type;
258 if (! entry->in_pack)
259 to_reuse = 0; /* can't reuse what we don't have */
260 else if (obj_type == OBJ_DELTA)
261 to_reuse = 1; /* check_object() decided it for us */
262 else if (obj_type != entry->in_pack_type)
263 to_reuse = 0; /* pack has delta which is unusable */
264 else if (entry->delta)
265 to_reuse = 0; /* we want to pack afresh */
266 else
267 to_reuse = 1; /* we have it in-pack undeltified,
268 * and we do not need to deltify it.
269 */
270
271 if (! to_reuse) {
272 buf = read_sha1_file(entry->sha1, type, &size);
273 if (!buf)
274 die("unable to read %s", sha1_to_hex(entry->sha1));
275 if (size != entry->size)
276 die("object %s size inconsistency (%lu vs %lu)",
277 sha1_to_hex(entry->sha1), size, entry->size);
278 if (entry->delta) {
279 buf = delta_against(buf, size, entry);
280 size = entry->delta_size;
281 obj_type = OBJ_DELTA;
282 }
283 /*
284 * The object header is a byte of 'type' followed by zero or
285 * more bytes of length. For deltas, the 20 bytes of delta
286 * sha1 follows that.
287 */
288 hdrlen = encode_header(obj_type, size, header);
289 sha1write(f, header, hdrlen);
290
291 if (entry->delta) {
292 sha1write(f, entry->delta, 20);
293 hdrlen += 20;
294 }
295 datalen = sha1write_compressed(f, buf, size);
296 free(buf);
297 }
298 else {
299 struct packed_git *p = entry->in_pack;
300 use_packed_git(p);
301
302 datalen = find_packed_object_size(p, entry->in_pack_offset);
303 buf = p->pack_base + entry->in_pack_offset;
304 sha1write(f, buf, datalen);
305 unuse_packed_git(p);
306 hdrlen = 0; /* not really */
307 if (obj_type == OBJ_DELTA)
308 reused_delta++;
309 reused++;
310 }
311 if (obj_type == OBJ_DELTA)
312 written_delta++;
313 written++;
314 return hdrlen + datalen;
315 }
316
317 static unsigned long write_one(struct sha1file *f,
318 struct object_entry *e,
319 unsigned long offset)
320 {
321 if (e->offset)
322 /* offset starts from header size and cannot be zero
323 * if it is written already.
324 */
325 return offset;
326 e->offset = offset;
327 offset += write_object(f, e);
328 /* if we are deltified, write out its base object. */
329 if (e->delta)
330 offset = write_one(f, e->delta, offset);
331 return offset;
332 }
333
334 static void write_pack_file(void)
335 {
336 int i;
337 struct sha1file *f;
338 unsigned long offset;
339 struct pack_header hdr;
340 unsigned last_percent = 999;
341 int do_progress = 0;
342
343 if (!base_name)
344 f = sha1fd(1, "<stdout>");
345 else {
346 f = sha1create("%s-%s.%s", base_name,
347 sha1_to_hex(object_list_sha1), "pack");
348 do_progress = progress;
349 }
350 if (do_progress)
351 fprintf(stderr, "Writing %d objects.\n", nr_result);
352
353 hdr.hdr_signature = htonl(PACK_SIGNATURE);
354 hdr.hdr_version = htonl(PACK_VERSION);
355 hdr.hdr_entries = htonl(nr_result);
356 sha1write(f, &hdr, sizeof(hdr));
357 offset = sizeof(hdr);
358 if (!nr_result)
359 goto done;
360 for (i = 0; i < nr_objects; i++) {
361 offset = write_one(f, objects + i, offset);
362 if (do_progress) {
363 unsigned percent = written * 100 / nr_result;
364 if (progress_update || percent != last_percent) {
365 fprintf(stderr, "%4u%% (%u/%u) done\r",
366 percent, written, nr_result);
367 progress_update = 0;
368 last_percent = percent;
369 }
370 }
371 }
372 if (do_progress)
373 fputc('\n', stderr);
374 done:
375 sha1close(f, pack_file_sha1, 1);
376 }
377
378 static void write_index_file(void)
379 {
380 int i;
381 struct sha1file *f = sha1create("%s-%s.%s", base_name,
382 sha1_to_hex(object_list_sha1), "idx");
383 struct object_entry **list = sorted_by_sha;
384 struct object_entry **last = list + nr_result;
385 unsigned int array[256];
386
387 /*
388 * Write the first-level table (the list is sorted,
389 * but we use a 256-entry lookup to be able to avoid
390 * having to do eight extra binary search iterations).
391 */
392 for (i = 0; i < 256; i++) {
393 struct object_entry **next = list;
394 while (next < last) {
395 struct object_entry *entry = *next;
396 if (entry->sha1[0] != i)
397 break;
398 next++;
399 }
400 array[i] = htonl(next - sorted_by_sha);
401 list = next;
402 }
403 sha1write(f, array, 256 * sizeof(int));
404
405 /*
406 * Write the actual SHA1 entries..
407 */
408 list = sorted_by_sha;
409 for (i = 0; i < nr_result; i++) {
410 struct object_entry *entry = *list++;
411 unsigned int offset = htonl(entry->offset);
412 sha1write(f, &offset, 4);
413 sha1write(f, entry->sha1, 20);
414 }
415 sha1write(f, pack_file_sha1, 20);
416 sha1close(f, NULL, 1);
417 }
418
419 static int locate_object_entry_hash(const unsigned char *sha1)
420 {
421 int i;
422 unsigned int ui;
423 memcpy(&ui, sha1, sizeof(unsigned int));
424 i = ui % object_ix_hashsz;
425 while (0 < object_ix[i]) {
426 if (!memcmp(sha1, objects[object_ix[i]-1].sha1, 20))
427 return i;
428 if (++i == object_ix_hashsz)
429 i = 0;
430 }
431 return -1 - i;
432 }
433
434 static struct object_entry *locate_object_entry(const unsigned char *sha1)
435 {
436 int i;
437
438 if (!object_ix_hashsz)
439 return NULL;
440
441 i = locate_object_entry_hash(sha1);
442 if (0 <= i)
443 return &objects[object_ix[i]-1];
444 return NULL;
445 }
446
447 static void rehash_objects(void)
448 {
449 int i;
450 struct object_entry *oe;
451
452 object_ix_hashsz = nr_objects * 3;
453 if (object_ix_hashsz < 1024)
454 object_ix_hashsz = 1024;
455 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
456 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
457 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
458 int ix = locate_object_entry_hash(oe->sha1);
459 if (0 <= ix)
460 continue;
461 ix = -1 - ix;
462 object_ix[ix] = i + 1;
463 }
464 }
465
466 static unsigned name_hash(const char *name)
467 {
468 unsigned char c;
469 unsigned hash = 0;
470
471 /*
472 * This effectively just creates a sortable number from the
473 * last sixteen non-whitespace characters. Last characters
474 * count "most", so things that end in ".c" sort together.
475 */
476 while ((c = *name++) != 0) {
477 if (isspace(c))
478 continue;
479 hash = (hash >> 2) + (c << 24);
480 }
481 return hash;
482 }
483
484 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
485 {
486 unsigned int idx = nr_objects;
487 struct object_entry *entry;
488 struct packed_git *p;
489 unsigned int found_offset = 0;
490 struct packed_git *found_pack = NULL;
491 int ix, status = 0;
492
493 if (!exclude) {
494 for (p = packed_git; p; p = p->next) {
495 struct pack_entry e;
496 if (find_pack_entry_one(sha1, &e, p)) {
497 if (incremental)
498 return 0;
499 if (local && !p->pack_local)
500 return 0;
501 if (!found_pack) {
502 found_offset = e.offset;
503 found_pack = e.p;
504 }
505 }
506 }
507 }
508 if ((entry = locate_object_entry(sha1)) != NULL)
509 goto already_added;
510
511 if (idx >= nr_alloc) {
512 unsigned int needed = (idx + 1024) * 3 / 2;
513 objects = xrealloc(objects, needed * sizeof(*entry));
514 nr_alloc = needed;
515 }
516 entry = objects + idx;
517 nr_objects = idx + 1;
518 memset(entry, 0, sizeof(*entry));
519 memcpy(entry->sha1, sha1, 20);
520 entry->hash = hash;
521
522 if (object_ix_hashsz * 3 <= nr_objects * 4)
523 rehash_objects();
524 else {
525 ix = locate_object_entry_hash(entry->sha1);
526 if (0 <= ix)
527 die("internal error in object hashing.");
528 object_ix[-1 - ix] = idx + 1;
529 }
530 status = 1;
531
532 already_added:
533 if (progress_update) {
534 fprintf(stderr, "Counting objects...%d\r", nr_objects);
535 progress_update = 0;
536 }
537 if (exclude)
538 entry->preferred_base = 1;
539 else {
540 if (found_pack) {
541 entry->in_pack = found_pack;
542 entry->in_pack_offset = found_offset;
543 }
544 }
545 return status;
546 }
547
548 struct pbase_tree_cache {
549 unsigned char sha1[20];
550 int ref;
551 int temporary;
552 void *tree_data;
553 unsigned long tree_size;
554 };
555
556 static struct pbase_tree_cache *(pbase_tree_cache[256]);
557 static int pbase_tree_cache_ix(const unsigned char *sha1)
558 {
559 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
560 }
561 static int pbase_tree_cache_ix_incr(int ix)
562 {
563 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
564 }
565
566 static struct pbase_tree {
567 struct pbase_tree *next;
568 /* This is a phony "cache" entry; we are not
569 * going to evict it nor find it through _get()
570 * mechanism -- this is for the toplevel node that
571 * would almost always change with any commit.
572 */
573 struct pbase_tree_cache pcache;
574 } *pbase_tree;
575
576 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
577 {
578 struct pbase_tree_cache *ent, *nent;
579 void *data;
580 unsigned long size;
581 char type[20];
582 int neigh;
583 int my_ix = pbase_tree_cache_ix(sha1);
584 int available_ix = -1;
585
586 /* pbase-tree-cache acts as a limited hashtable.
587 * your object will be found at your index or within a few
588 * slots after that slot if it is cached.
589 */
590 for (neigh = 0; neigh < 8; neigh++) {
591 ent = pbase_tree_cache[my_ix];
592 if (ent && !memcmp(ent->sha1, sha1, 20)) {
593 ent->ref++;
594 return ent;
595 }
596 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
597 ((0 <= available_ix) &&
598 (!ent && pbase_tree_cache[available_ix])))
599 available_ix = my_ix;
600 if (!ent)
601 break;
602 my_ix = pbase_tree_cache_ix_incr(my_ix);
603 }
604
605 /* Did not find one. Either we got a bogus request or
606 * we need to read and perhaps cache.
607 */
608 data = read_sha1_file(sha1, type, &size);
609 if (!data)
610 return NULL;
611 if (strcmp(type, tree_type)) {
612 free(data);
613 return NULL;
614 }
615
616 /* We need to either cache or return a throwaway copy */
617
618 if (available_ix < 0)
619 ent = NULL;
620 else {
621 ent = pbase_tree_cache[available_ix];
622 my_ix = available_ix;
623 }
624
625 if (!ent) {
626 nent = xmalloc(sizeof(*nent));
627 nent->temporary = (available_ix < 0);
628 }
629 else {
630 /* evict and reuse */
631 free(ent->tree_data);
632 nent = ent;
633 }
634 memcpy(nent->sha1, sha1, 20);
635 nent->tree_data = data;
636 nent->tree_size = size;
637 nent->ref = 1;
638 if (!nent->temporary)
639 pbase_tree_cache[my_ix] = nent;
640 return nent;
641 }
642
643 static void pbase_tree_put(struct pbase_tree_cache *cache)
644 {
645 if (!cache->temporary) {
646 cache->ref--;
647 return;
648 }
649 free(cache->tree_data);
650 free(cache);
651 }
652
653 static int name_cmp_len(const char *name)
654 {
655 int i;
656 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
657 ;
658 return i;
659 }
660
661 static void add_pbase_object(struct tree_desc *tree,
662 const char *name,
663 int cmplen,
664 const char *fullname)
665 {
666 struct name_entry entry;
667
668 while (tree_entry(tree,&entry)) {
669 unsigned long size;
670 char type[20];
671
672 if (entry.pathlen != cmplen ||
673 memcmp(entry.path, name, cmplen) ||
674 !has_sha1_file(entry.sha1) ||
675 sha1_object_info(entry.sha1, type, &size))
676 continue;
677 if (name[cmplen] != '/') {
678 unsigned hash = name_hash(fullname);
679 add_object_entry(entry.sha1, hash, 1);
680 return;
681 }
682 if (!strcmp(type, tree_type)) {
683 struct tree_desc sub;
684 struct pbase_tree_cache *tree;
685 const char *down = name+cmplen+1;
686 int downlen = name_cmp_len(down);
687
688 tree = pbase_tree_get(entry.sha1);
689 if (!tree)
690 return;
691 sub.buf = tree->tree_data;
692 sub.size = tree->tree_size;
693
694 add_pbase_object(&sub, down, downlen, fullname);
695 pbase_tree_put(tree);
696 }
697 }
698 }
699
700 static unsigned *done_pbase_paths;
701 static int done_pbase_paths_num;
702 static int done_pbase_paths_alloc;
703 static int done_pbase_path_pos(unsigned hash)
704 {
705 int lo = 0;
706 int hi = done_pbase_paths_num;
707 while (lo < hi) {
708 int mi = (hi + lo) / 2;
709 if (done_pbase_paths[mi] == hash)
710 return mi;
711 if (done_pbase_paths[mi] < hash)
712 hi = mi;
713 else
714 lo = mi + 1;
715 }
716 return -lo-1;
717 }
718
719 static int check_pbase_path(unsigned hash)
720 {
721 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
722 if (0 <= pos)
723 return 1;
724 pos = -pos - 1;
725 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
726 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
727 done_pbase_paths = xrealloc(done_pbase_paths,
728 done_pbase_paths_alloc *
729 sizeof(unsigned));
730 }
731 done_pbase_paths_num++;
732 if (pos < done_pbase_paths_num)
733 memmove(done_pbase_paths + pos + 1,
734 done_pbase_paths + pos,
735 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
736 done_pbase_paths[pos] = hash;
737 return 0;
738 }
739
740 static void add_preferred_base_object(char *name, unsigned hash)
741 {
742 struct pbase_tree *it;
743 int cmplen = name_cmp_len(name);
744
745 if (check_pbase_path(hash))
746 return;
747
748 for (it = pbase_tree; it; it = it->next) {
749 if (cmplen == 0) {
750 hash = name_hash("");
751 add_object_entry(it->pcache.sha1, hash, 1);
752 }
753 else {
754 struct tree_desc tree;
755 tree.buf = it->pcache.tree_data;
756 tree.size = it->pcache.tree_size;
757 add_pbase_object(&tree, name, cmplen, name);
758 }
759 }
760 }
761
762 static void add_preferred_base(unsigned char *sha1)
763 {
764 struct pbase_tree *it;
765 void *data;
766 unsigned long size;
767 unsigned char tree_sha1[20];
768
769 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
770 if (!data)
771 return;
772
773 for (it = pbase_tree; it; it = it->next) {
774 if (!memcmp(it->pcache.sha1, tree_sha1, 20)) {
775 free(data);
776 return;
777 }
778 }
779
780 it = xcalloc(1, sizeof(*it));
781 it->next = pbase_tree;
782 pbase_tree = it;
783
784 memcpy(it->pcache.sha1, tree_sha1, 20);
785 it->pcache.tree_data = data;
786 it->pcache.tree_size = size;
787 }
788
789 static void check_object(struct object_entry *entry)
790 {
791 char type[20];
792
793 if (entry->in_pack && !entry->preferred_base) {
794 unsigned char base[20];
795 unsigned long size;
796 struct object_entry *base_entry;
797
798 /* We want in_pack_type even if we do not reuse delta.
799 * There is no point not reusing non-delta representations.
800 */
801 check_reuse_pack_delta(entry->in_pack,
802 entry->in_pack_offset,
803 base, &size,
804 &entry->in_pack_type);
805
806 /* Check if it is delta, and the base is also an object
807 * we are going to pack. If so we will reuse the existing
808 * delta.
809 */
810 if (!no_reuse_delta &&
811 entry->in_pack_type == OBJ_DELTA &&
812 (base_entry = locate_object_entry(base)) &&
813 (!base_entry->preferred_base)) {
814
815 /* Depth value does not matter - find_deltas()
816 * will never consider reused delta as the
817 * base object to deltify other objects
818 * against, in order to avoid circular deltas.
819 */
820
821 /* uncompressed size of the delta data */
822 entry->size = entry->delta_size = size;
823 entry->delta = base_entry;
824 entry->type = OBJ_DELTA;
825
826 entry->delta_sibling = base_entry->delta_child;
827 base_entry->delta_child = entry;
828
829 return;
830 }
831 /* Otherwise we would do the usual */
832 }
833
834 if (sha1_object_info(entry->sha1, type, &entry->size))
835 die("unable to get type of object %s",
836 sha1_to_hex(entry->sha1));
837
838 if (!strcmp(type, commit_type)) {
839 entry->type = OBJ_COMMIT;
840 } else if (!strcmp(type, tree_type)) {
841 entry->type = OBJ_TREE;
842 } else if (!strcmp(type, blob_type)) {
843 entry->type = OBJ_BLOB;
844 } else if (!strcmp(type, tag_type)) {
845 entry->type = OBJ_TAG;
846 } else
847 die("unable to pack object %s of type %s",
848 sha1_to_hex(entry->sha1), type);
849 }
850
851 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
852 {
853 struct object_entry *child = me->delta_child;
854 unsigned int m = n;
855 while (child) {
856 unsigned int c = check_delta_limit(child, n + 1);
857 if (m < c)
858 m = c;
859 child = child->delta_sibling;
860 }
861 return m;
862 }
863
864 static void get_object_details(void)
865 {
866 int i;
867 struct object_entry *entry;
868
869 prepare_pack_ix();
870 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
871 check_object(entry);
872
873 if (nr_objects == nr_result) {
874 /*
875 * Depth of objects that depend on the entry -- this
876 * is subtracted from depth-max to break too deep
877 * delta chain because of delta data reusing.
878 * However, we loosen this restriction when we know we
879 * are creating a thin pack -- it will have to be
880 * expanded on the other end anyway, so do not
881 * artificially cut the delta chain and let it go as
882 * deep as it wants.
883 */
884 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
885 if (!entry->delta && entry->delta_child)
886 entry->delta_limit =
887 check_delta_limit(entry, 1);
888 }
889 }
890
891 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
892
893 static entry_sort_t current_sort;
894
895 static int sort_comparator(const void *_a, const void *_b)
896 {
897 struct object_entry *a = *(struct object_entry **)_a;
898 struct object_entry *b = *(struct object_entry **)_b;
899 return current_sort(a,b);
900 }
901
902 static struct object_entry **create_sorted_list(entry_sort_t sort)
903 {
904 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
905 int i;
906
907 for (i = 0; i < nr_objects; i++)
908 list[i] = objects + i;
909 current_sort = sort;
910 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
911 return list;
912 }
913
914 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
915 {
916 return memcmp(a->sha1, b->sha1, 20);
917 }
918
919 static struct object_entry **create_final_object_list(void)
920 {
921 struct object_entry **list;
922 int i, j;
923
924 for (i = nr_result = 0; i < nr_objects; i++)
925 if (!objects[i].preferred_base)
926 nr_result++;
927 list = xmalloc(nr_result * sizeof(struct object_entry *));
928 for (i = j = 0; i < nr_objects; i++) {
929 if (!objects[i].preferred_base)
930 list[j++] = objects + i;
931 }
932 current_sort = sha1_sort;
933 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
934 return list;
935 }
936
937 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
938 {
939 if (a->type < b->type)
940 return -1;
941 if (a->type > b->type)
942 return 1;
943 if (a->hash < b->hash)
944 return -1;
945 if (a->hash > b->hash)
946 return 1;
947 if (a->preferred_base < b->preferred_base)
948 return -1;
949 if (a->preferred_base > b->preferred_base)
950 return 1;
951 if (a->size < b->size)
952 return -1;
953 if (a->size > b->size)
954 return 1;
955 return a < b ? -1 : (a > b);
956 }
957
958 struct unpacked {
959 struct object_entry *entry;
960 void *data;
961 struct delta_index *index;
962 };
963
964 /*
965 * We search for deltas _backwards_ in a list sorted by type and
966 * by size, so that we see progressively smaller and smaller files.
967 * That's because we prefer deltas to be from the bigger file
968 * to the smaller - deletes are potentially cheaper, but perhaps
969 * more importantly, the bigger file is likely the more recent
970 * one.
971 */
972 static int try_delta(struct unpacked *trg, struct unpacked *src,
973 struct delta_index *src_index, unsigned max_depth)
974 {
975 struct object_entry *trg_entry = trg->entry;
976 struct object_entry *src_entry = src->entry;
977 unsigned long size, src_size, delta_size, sizediff, max_size;
978 void *delta_buf;
979
980 /* Don't bother doing diffs between different types */
981 if (trg_entry->type != src_entry->type)
982 return -1;
983
984 /* We do not compute delta to *create* objects we are not
985 * going to pack.
986 */
987 if (trg_entry->preferred_base)
988 return -1;
989
990 /*
991 * If the current object is at pack edge, take the depth the
992 * objects that depend on the current object into account --
993 * otherwise they would become too deep.
994 */
995 if (trg_entry->delta_child) {
996 if (max_depth <= trg_entry->delta_limit)
997 return 0;
998 max_depth -= trg_entry->delta_limit;
999 }
1000 if (src_entry->depth >= max_depth)
1001 return 0;
1002
1003 /* Now some size filtering heuristics. */
1004 size = trg_entry->size;
1005 max_size = size/2 - 20;
1006 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1007 if (max_size == 0)
1008 return 0;
1009 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1010 max_size = trg_entry->delta_size-1;
1011 src_size = src_entry->size;
1012 sizediff = src_size < size ? size - src_size : 0;
1013 if (sizediff >= max_size)
1014 return 0;
1015
1016 delta_buf = create_delta(src_index, trg->data, size, &delta_size, max_size);
1017 if (!delta_buf)
1018 return 0;
1019
1020 trg_entry->delta = src_entry;
1021 trg_entry->delta_size = delta_size;
1022 trg_entry->depth = src_entry->depth + 1;
1023 free(delta_buf);
1024 return 1;
1025 }
1026
1027 static void progress_interval(int signum)
1028 {
1029 progress_update = 1;
1030 }
1031
1032 static void find_deltas(struct object_entry **list, int window, int depth)
1033 {
1034 int i, idx;
1035 unsigned int array_size = window * sizeof(struct unpacked);
1036 struct unpacked *array = xmalloc(array_size);
1037 unsigned processed = 0;
1038 unsigned last_percent = 999;
1039
1040 memset(array, 0, array_size);
1041 i = nr_objects;
1042 idx = 0;
1043 if (progress)
1044 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1045
1046 while (--i >= 0) {
1047 struct object_entry *entry = list[i];
1048 struct unpacked *n = array + idx;
1049 unsigned long size;
1050 char type[10];
1051 int j;
1052
1053 if (!entry->preferred_base)
1054 processed++;
1055
1056 if (progress) {
1057 unsigned percent = processed * 100 / nr_result;
1058 if (percent != last_percent || progress_update) {
1059 fprintf(stderr, "%4u%% (%u/%u) done\r",
1060 percent, processed, nr_result);
1061 progress_update = 0;
1062 last_percent = percent;
1063 }
1064 }
1065
1066 if (entry->delta)
1067 /* This happens if we decided to reuse existing
1068 * delta from a pack. "!no_reuse_delta &&" is implied.
1069 */
1070 continue;
1071
1072 if (entry->size < 50)
1073 continue;
1074 free_delta_index(n->index);
1075 n->index = NULL;
1076 free(n->data);
1077 n->entry = entry;
1078 n->data = read_sha1_file(entry->sha1, type, &size);
1079 if (size != entry->size)
1080 die("object %s inconsistent object length (%lu vs %lu)",
1081 sha1_to_hex(entry->sha1), size, entry->size);
1082
1083 j = window;
1084 while (--j > 0) {
1085 unsigned int other_idx = idx + j;
1086 struct unpacked *m;
1087 if (other_idx >= window)
1088 other_idx -= window;
1089 m = array + other_idx;
1090 if (!m->entry)
1091 break;
1092 if (try_delta(n, m, m->index, depth) < 0)
1093 break;
1094 }
1095 /* if we made n a delta, and if n is already at max
1096 * depth, leaving it in the window is pointless. we
1097 * should evict it first.
1098 */
1099 if (entry->delta && depth <= entry->depth)
1100 continue;
1101
1102 n->index = create_delta_index(n->data, size);
1103 if (!n->index)
1104 die("out of memory");
1105
1106 idx++;
1107 if (idx >= window)
1108 idx = 0;
1109 }
1110
1111 if (progress)
1112 fputc('\n', stderr);
1113
1114 for (i = 0; i < window; ++i) {
1115 free_delta_index(array[i].index);
1116 free(array[i].data);
1117 }
1118 free(array);
1119 }
1120
1121 static void prepare_pack(int window, int depth)
1122 {
1123 get_object_details();
1124 sorted_by_type = create_sorted_list(type_size_sort);
1125 if (window && depth)
1126 find_deltas(sorted_by_type, window+1, depth);
1127 }
1128
1129 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
1130 {
1131 static const char cache[] = "pack-cache/pack-%s.%s";
1132 char *cached_pack, *cached_idx;
1133 int ifd, ofd, ifd_ix = -1;
1134
1135 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1136 ifd = open(cached_pack, O_RDONLY);
1137 if (ifd < 0)
1138 return 0;
1139
1140 if (!pack_to_stdout) {
1141 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1142 ifd_ix = open(cached_idx, O_RDONLY);
1143 if (ifd_ix < 0) {
1144 close(ifd);
1145 return 0;
1146 }
1147 }
1148
1149 if (progress)
1150 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1151 sha1_to_hex(sha1));
1152
1153 if (pack_to_stdout) {
1154 if (copy_fd(ifd, 1))
1155 exit(1);
1156 close(ifd);
1157 }
1158 else {
1159 char name[PATH_MAX];
1160 snprintf(name, sizeof(name),
1161 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1162 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1163 if (ofd < 0)
1164 die("unable to open %s (%s)", name, strerror(errno));
1165 if (copy_fd(ifd, ofd))
1166 exit(1);
1167 close(ifd);
1168
1169 snprintf(name, sizeof(name),
1170 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1171 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1172 if (ofd < 0)
1173 die("unable to open %s (%s)", name, strerror(errno));
1174 if (copy_fd(ifd_ix, ofd))
1175 exit(1);
1176 close(ifd_ix);
1177 puts(sha1_to_hex(sha1));
1178 }
1179
1180 return 1;
1181 }
1182
1183 static void setup_progress_signal(void)
1184 {
1185 struct sigaction sa;
1186 struct itimerval v;
1187
1188 memset(&sa, 0, sizeof(sa));
1189 sa.sa_handler = progress_interval;
1190 sigemptyset(&sa.sa_mask);
1191 sa.sa_flags = SA_RESTART;
1192 sigaction(SIGALRM, &sa, NULL);
1193
1194 v.it_interval.tv_sec = 1;
1195 v.it_interval.tv_usec = 0;
1196 v.it_value = v.it_interval;
1197 setitimer(ITIMER_REAL, &v, NULL);
1198 }
1199
1200 int main(int argc, char **argv)
1201 {
1202 SHA_CTX ctx;
1203 char line[40 + 1 + PATH_MAX + 2];
1204 int window = 10, depth = 10, pack_to_stdout = 0;
1205 struct object_entry **list;
1206 int num_preferred_base = 0;
1207 int i;
1208
1209 setup_git_directory();
1210
1211 progress = isatty(2);
1212 for (i = 1; i < argc; i++) {
1213 const char *arg = argv[i];
1214
1215 if (*arg == '-') {
1216 if (!strcmp("--non-empty", arg)) {
1217 non_empty = 1;
1218 continue;
1219 }
1220 if (!strcmp("--local", arg)) {
1221 local = 1;
1222 continue;
1223 }
1224 if (!strcmp("--progress", arg)) {
1225 progress = 1;
1226 continue;
1227 }
1228 if (!strcmp("--incremental", arg)) {
1229 incremental = 1;
1230 continue;
1231 }
1232 if (!strncmp("--window=", arg, 9)) {
1233 char *end;
1234 window = strtoul(arg+9, &end, 0);
1235 if (!arg[9] || *end)
1236 usage(pack_usage);
1237 continue;
1238 }
1239 if (!strncmp("--depth=", arg, 8)) {
1240 char *end;
1241 depth = strtoul(arg+8, &end, 0);
1242 if (!arg[8] || *end)
1243 usage(pack_usage);
1244 continue;
1245 }
1246 if (!strcmp("--progress", arg)) {
1247 progress = 1;
1248 continue;
1249 }
1250 if (!strcmp("-q", arg)) {
1251 progress = 0;
1252 continue;
1253 }
1254 if (!strcmp("--no-reuse-delta", arg)) {
1255 no_reuse_delta = 1;
1256 continue;
1257 }
1258 if (!strcmp("--stdout", arg)) {
1259 pack_to_stdout = 1;
1260 continue;
1261 }
1262 usage(pack_usage);
1263 }
1264 if (base_name)
1265 usage(pack_usage);
1266 base_name = arg;
1267 }
1268
1269 if (pack_to_stdout != !base_name)
1270 usage(pack_usage);
1271
1272 prepare_packed_git();
1273
1274 if (progress) {
1275 fprintf(stderr, "Generating pack...\n");
1276 setup_progress_signal();
1277 }
1278
1279 for (;;) {
1280 unsigned char sha1[20];
1281 unsigned hash;
1282
1283 if (!fgets(line, sizeof(line), stdin)) {
1284 if (feof(stdin))
1285 break;
1286 if (!ferror(stdin))
1287 die("fgets returned NULL, not EOF, not error!");
1288 if (errno != EINTR)
1289 die("fgets: %s", strerror(errno));
1290 clearerr(stdin);
1291 continue;
1292 }
1293
1294 if (line[0] == '-') {
1295 if (get_sha1_hex(line+1, sha1))
1296 die("expected edge sha1, got garbage:\n %s",
1297 line+1);
1298 if (num_preferred_base++ < window)
1299 add_preferred_base(sha1);
1300 continue;
1301 }
1302 if (get_sha1_hex(line, sha1))
1303 die("expected sha1, got garbage:\n %s", line);
1304 hash = name_hash(line+41);
1305 add_preferred_base_object(line+41, hash);
1306 add_object_entry(sha1, hash, 0);
1307 }
1308 if (progress)
1309 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1310 sorted_by_sha = create_final_object_list();
1311 if (non_empty && !nr_result)
1312 return 0;
1313
1314 SHA1_Init(&ctx);
1315 list = sorted_by_sha;
1316 for (i = 0; i < nr_result; i++) {
1317 struct object_entry *entry = *list++;
1318 SHA1_Update(&ctx, entry->sha1, 20);
1319 }
1320 SHA1_Final(object_list_sha1, &ctx);
1321 if (progress && (nr_objects != nr_result))
1322 fprintf(stderr, "Result has %d objects.\n", nr_result);
1323
1324 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1325 ;
1326 else {
1327 if (nr_result)
1328 prepare_pack(window, depth);
1329 if (progress && pack_to_stdout) {
1330 /* the other end usually displays progress itself */
1331 struct itimerval v = {{0,},};
1332 setitimer(ITIMER_REAL, &v, NULL);
1333 signal(SIGALRM, SIG_IGN );
1334 progress_update = 0;
1335 }
1336 write_pack_file();
1337 if (!pack_to_stdout) {
1338 write_index_file();
1339 puts(sha1_to_hex(object_list_sha1));
1340 }
1341 }
1342 if (progress)
1343 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1344 nr_result, written, written_delta, reused, reused_delta);
1345 return 0;
1346 }