gitweb: Do not print "log" and "shortlog" redundantly in commit view
[git/jrn.git] / builtin-pack-objects.c
blob96c069a81da643b7ee3515ca8c89734881fa3b77
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "blob.h"
5 #include "commit.h"
6 #include "tag.h"
7 #include "tree.h"
8 #include "delta.h"
9 #include "pack.h"
10 #include "csum-file.h"
11 #include "tree-walk.h"
12 #include "diff.h"
13 #include "revision.h"
14 #include "list-objects.h"
15 #include <sys/time.h>
16 #include <signal.h>
18 static const char pack_usage[] = "git-pack-objects [-q] [--no-reuse-delta] [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] [--revs [--unpacked | --all]*] [--stdout | base-name] <ref-list | <object-list]";
20 struct object_entry {
21 unsigned char sha1[20];
22 unsigned long size; /* uncompressed size */
23 unsigned long offset; /* offset into the final pack file;
24 * nonzero if already written.
26 unsigned int depth; /* delta depth */
27 unsigned int delta_limit; /* base adjustment for in-pack delta */
28 unsigned int hash; /* name hint hash */
29 enum object_type type;
30 enum object_type in_pack_type; /* could be delta */
31 unsigned long delta_size; /* delta data size (uncompressed) */
32 struct object_entry *delta; /* delta base object */
33 struct packed_git *in_pack; /* already in pack */
34 unsigned int in_pack_offset;
35 struct object_entry *delta_child; /* deltified objects who bases me */
36 struct object_entry *delta_sibling; /* other deltified objects who
37 * uses the same base as me
39 int preferred_base; /* we do not pack this, but is encouraged to
40 * be used as the base objectto delta huge
41 * objects against.
46 * Objects we are going to pack are collected in objects array (dynamically
47 * expanded). nr_objects & nr_alloc controls this array. They are stored
48 * in the order we see -- typically rev-list --objects order that gives us
49 * nice "minimum seek" order.
51 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
52 * elements in the objects array. The former is used to build the pack
53 * index (lists object names in the ascending order to help offset lookup),
54 * and the latter is used to group similar things together by try_delta()
55 * heuristics.
58 static unsigned char object_list_sha1[20];
59 static int non_empty;
60 static int no_reuse_delta;
61 static int local;
62 static int incremental;
63 static struct object_entry **sorted_by_sha, **sorted_by_type;
64 static struct object_entry *objects;
65 static int nr_objects, nr_alloc, nr_result;
66 static const char *base_name;
67 static unsigned char pack_file_sha1[20];
68 static int progress = 1;
69 static volatile sig_atomic_t progress_update;
70 static int window = 10;
71 static int pack_to_stdout;
72 static int num_preferred_base;
75 * The object names in objects array are hashed with this hashtable,
76 * to help looking up the entry by object name. Binary search from
77 * sorted_by_sha is also possible but this was easier to code and faster.
78 * This hashtable is built after all the objects are seen.
80 static int *object_ix;
81 static int object_ix_hashsz;
84 * Pack index for existing packs give us easy access to the offsets into
85 * corresponding pack file where each object's data starts, but the entries
86 * do not store the size of the compressed representation (uncompressed
87 * size is easily available by examining the pack entry header). We build
88 * a hashtable of existing packs (pack_revindex), and keep reverse index
89 * here -- pack index file is sorted by object name mapping to offset; this
90 * pack_revindex[].revindex array is an ordered list of offsets, so if you
91 * know the offset of an object, next offset is where its packed
92 * representation ends.
94 struct pack_revindex {
95 struct packed_git *p;
96 unsigned long *revindex;
97 } *pack_revindex = NULL;
98 static int pack_revindex_hashsz;
101 * stats
103 static int written;
104 static int written_delta;
105 static int reused;
106 static int reused_delta;
108 static int pack_revindex_ix(struct packed_git *p)
110 unsigned long ui = (unsigned long)p;
111 int i;
113 ui = ui ^ (ui >> 16); /* defeat structure alignment */
114 i = (int)(ui % pack_revindex_hashsz);
115 while (pack_revindex[i].p) {
116 if (pack_revindex[i].p == p)
117 return i;
118 if (++i == pack_revindex_hashsz)
119 i = 0;
121 return -1 - i;
124 static void prepare_pack_ix(void)
126 int num;
127 struct packed_git *p;
128 for (num = 0, p = packed_git; p; p = p->next)
129 num++;
130 if (!num)
131 return;
132 pack_revindex_hashsz = num * 11;
133 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
134 for (p = packed_git; p; p = p->next) {
135 num = pack_revindex_ix(p);
136 num = - 1 - num;
137 pack_revindex[num].p = p;
139 /* revindex elements are lazily initialized */
142 static int cmp_offset(const void *a_, const void *b_)
144 unsigned long a = *(unsigned long *) a_;
145 unsigned long b = *(unsigned long *) b_;
146 if (a < b)
147 return -1;
148 else if (a == b)
149 return 0;
150 else
151 return 1;
155 * Ordered list of offsets of objects in the pack.
157 static void prepare_pack_revindex(struct pack_revindex *rix)
159 struct packed_git *p = rix->p;
160 int num_ent = num_packed_objects(p);
161 int i;
162 void *index = p->index_base + 256;
164 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
165 for (i = 0; i < num_ent; i++) {
166 unsigned int hl = *((unsigned int *)((char *) index + 24*i));
167 rix->revindex[i] = ntohl(hl);
169 /* This knows the pack format -- the 20-byte trailer
170 * follows immediately after the last object data.
172 rix->revindex[num_ent] = p->pack_size - 20;
173 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
176 static unsigned long find_packed_object_size(struct packed_git *p,
177 unsigned long ofs)
179 int num;
180 int lo, hi;
181 struct pack_revindex *rix;
182 unsigned long *revindex;
183 num = pack_revindex_ix(p);
184 if (num < 0)
185 die("internal error: pack revindex uninitialized");
186 rix = &pack_revindex[num];
187 if (!rix->revindex)
188 prepare_pack_revindex(rix);
189 revindex = rix->revindex;
190 lo = 0;
191 hi = num_packed_objects(p) + 1;
192 do {
193 int mi = (lo + hi) / 2;
194 if (revindex[mi] == ofs) {
195 return revindex[mi+1] - ofs;
197 else if (ofs < revindex[mi])
198 hi = mi;
199 else
200 lo = mi + 1;
201 } while (lo < hi);
202 die("internal error: pack revindex corrupt");
205 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
207 unsigned long othersize, delta_size;
208 char type[10];
209 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
210 void *delta_buf;
212 if (!otherbuf)
213 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
214 delta_buf = diff_delta(otherbuf, othersize,
215 buf, size, &delta_size, 0);
216 if (!delta_buf || delta_size != entry->delta_size)
217 die("delta size changed");
218 free(buf);
219 free(otherbuf);
220 return delta_buf;
224 * The per-object header is a pretty dense thing, which is
225 * - first byte: low four bits are "size", then three bits of "type",
226 * and the high bit is "size continues".
227 * - each byte afterwards: low seven bits are size continuation,
228 * with the high bit being "size continues"
230 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
232 int n = 1;
233 unsigned char c;
235 if (type < OBJ_COMMIT || type > OBJ_DELTA)
236 die("bad type %d", type);
238 c = (type << 4) | (size & 15);
239 size >>= 4;
240 while (size) {
241 *hdr++ = c | 0x80;
242 c = size & 0x7f;
243 size >>= 7;
244 n++;
246 *hdr = c;
247 return n;
250 static int check_inflate(unsigned char *data, unsigned long len, unsigned long expect)
252 z_stream stream;
253 unsigned char fakebuf[4096];
254 int st;
256 memset(&stream, 0, sizeof(stream));
257 stream.next_in = data;
258 stream.avail_in = len;
259 stream.next_out = fakebuf;
260 stream.avail_out = sizeof(fakebuf);
261 inflateInit(&stream);
263 while (1) {
264 st = inflate(&stream, Z_FINISH);
265 if (st == Z_STREAM_END || st == Z_OK) {
266 st = (stream.total_out == expect &&
267 stream.total_in == len) ? 0 : -1;
268 break;
270 if (st != Z_BUF_ERROR) {
271 st = -1;
272 break;
274 stream.next_out = fakebuf;
275 stream.avail_out = sizeof(fakebuf);
277 inflateEnd(&stream);
278 return st;
282 * we are going to reuse the existing pack entry data. make
283 * sure it is not corrupt.
285 static int revalidate_pack_entry(struct object_entry *entry, unsigned char *data, unsigned long len)
287 enum object_type type;
288 unsigned long size, used;
290 if (pack_to_stdout)
291 return 0;
293 /* the caller has already called use_packed_git() for us,
294 * so it is safe to access the pack data from mmapped location.
295 * make sure the entry inflates correctly.
297 used = unpack_object_header_gently(data, len, &type, &size);
298 if (!used)
299 return -1;
300 if (type == OBJ_DELTA)
301 used += 20; /* skip base object name */
302 data += used;
303 len -= used;
304 return check_inflate(data, len, entry->size);
307 static int revalidate_loose_object(struct object_entry *entry,
308 unsigned char *map,
309 unsigned long mapsize)
311 /* we already know this is a loose object with new type header. */
312 enum object_type type;
313 unsigned long size, used;
315 if (pack_to_stdout)
316 return 0;
318 used = unpack_object_header_gently(map, mapsize, &type, &size);
319 if (!used)
320 return -1;
321 map += used;
322 mapsize -= used;
323 return check_inflate(map, mapsize, size);
326 static unsigned long write_object(struct sha1file *f,
327 struct object_entry *entry)
329 unsigned long size;
330 char type[10];
331 void *buf;
332 unsigned char header[10];
333 unsigned hdrlen, datalen;
334 enum object_type obj_type;
335 int to_reuse = 0;
337 if (entry->preferred_base)
338 return 0;
340 obj_type = entry->type;
341 if (! entry->in_pack)
342 to_reuse = 0; /* can't reuse what we don't have */
343 else if (obj_type == OBJ_DELTA)
344 to_reuse = 1; /* check_object() decided it for us */
345 else if (obj_type != entry->in_pack_type)
346 to_reuse = 0; /* pack has delta which is unusable */
347 else if (entry->delta)
348 to_reuse = 0; /* we want to pack afresh */
349 else
350 to_reuse = 1; /* we have it in-pack undeltified,
351 * and we do not need to deltify it.
354 if (!entry->in_pack && !entry->delta) {
355 unsigned char *map;
356 unsigned long mapsize;
357 map = map_sha1_file(entry->sha1, &mapsize);
358 if (map && !legacy_loose_object(map)) {
359 /* We can copy straight into the pack file */
360 if (revalidate_loose_object(entry, map, mapsize))
361 die("corrupt loose object %s",
362 sha1_to_hex(entry->sha1));
363 sha1write(f, map, mapsize);
364 munmap(map, mapsize);
365 written++;
366 reused++;
367 return mapsize;
369 if (map)
370 munmap(map, mapsize);
373 if (!to_reuse) {
374 buf = read_sha1_file(entry->sha1, type, &size);
375 if (!buf)
376 die("unable to read %s", sha1_to_hex(entry->sha1));
377 if (size != entry->size)
378 die("object %s size inconsistency (%lu vs %lu)",
379 sha1_to_hex(entry->sha1), size, entry->size);
380 if (entry->delta) {
381 buf = delta_against(buf, size, entry);
382 size = entry->delta_size;
383 obj_type = OBJ_DELTA;
386 * The object header is a byte of 'type' followed by zero or
387 * more bytes of length. For deltas, the 20 bytes of delta
388 * sha1 follows that.
390 hdrlen = encode_header(obj_type, size, header);
391 sha1write(f, header, hdrlen);
393 if (entry->delta) {
394 sha1write(f, entry->delta, 20);
395 hdrlen += 20;
397 datalen = sha1write_compressed(f, buf, size);
398 free(buf);
400 else {
401 struct packed_git *p = entry->in_pack;
402 use_packed_git(p);
404 datalen = find_packed_object_size(p, entry->in_pack_offset);
405 buf = (char *) p->pack_base + entry->in_pack_offset;
407 if (revalidate_pack_entry(entry, buf, datalen))
408 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
409 sha1write(f, buf, datalen);
410 unuse_packed_git(p);
411 hdrlen = 0; /* not really */
412 if (obj_type == OBJ_DELTA)
413 reused_delta++;
414 reused++;
416 if (obj_type == OBJ_DELTA)
417 written_delta++;
418 written++;
419 return hdrlen + datalen;
422 static unsigned long write_one(struct sha1file *f,
423 struct object_entry *e,
424 unsigned long offset)
426 if (e->offset)
427 /* offset starts from header size and cannot be zero
428 * if it is written already.
430 return offset;
431 e->offset = offset;
432 offset += write_object(f, e);
433 /* if we are deltified, write out its base object. */
434 if (e->delta)
435 offset = write_one(f, e->delta, offset);
436 return offset;
439 static void write_pack_file(void)
441 int i;
442 struct sha1file *f;
443 unsigned long offset;
444 struct pack_header hdr;
445 unsigned last_percent = 999;
446 int do_progress = 0;
448 if (!base_name)
449 f = sha1fd(1, "<stdout>");
450 else {
451 f = sha1create("%s-%s.%s", base_name,
452 sha1_to_hex(object_list_sha1), "pack");
453 do_progress = progress;
455 if (do_progress)
456 fprintf(stderr, "Writing %d objects.\n", nr_result);
458 hdr.hdr_signature = htonl(PACK_SIGNATURE);
459 hdr.hdr_version = htonl(PACK_VERSION);
460 hdr.hdr_entries = htonl(nr_result);
461 sha1write(f, &hdr, sizeof(hdr));
462 offset = sizeof(hdr);
463 if (!nr_result)
464 goto done;
465 for (i = 0; i < nr_objects; i++) {
466 offset = write_one(f, objects + i, offset);
467 if (do_progress) {
468 unsigned percent = written * 100 / nr_result;
469 if (progress_update || percent != last_percent) {
470 fprintf(stderr, "%4u%% (%u/%u) done\r",
471 percent, written, nr_result);
472 progress_update = 0;
473 last_percent = percent;
477 if (do_progress)
478 fputc('\n', stderr);
479 done:
480 sha1close(f, pack_file_sha1, 1);
483 static void write_index_file(void)
485 int i;
486 struct sha1file *f = sha1create("%s-%s.%s", base_name,
487 sha1_to_hex(object_list_sha1), "idx");
488 struct object_entry **list = sorted_by_sha;
489 struct object_entry **last = list + nr_result;
490 unsigned int array[256];
493 * Write the first-level table (the list is sorted,
494 * but we use a 256-entry lookup to be able to avoid
495 * having to do eight extra binary search iterations).
497 for (i = 0; i < 256; i++) {
498 struct object_entry **next = list;
499 while (next < last) {
500 struct object_entry *entry = *next;
501 if (entry->sha1[0] != i)
502 break;
503 next++;
505 array[i] = htonl(next - sorted_by_sha);
506 list = next;
508 sha1write(f, array, 256 * sizeof(int));
511 * Write the actual SHA1 entries..
513 list = sorted_by_sha;
514 for (i = 0; i < nr_result; i++) {
515 struct object_entry *entry = *list++;
516 unsigned int offset = htonl(entry->offset);
517 sha1write(f, &offset, 4);
518 sha1write(f, entry->sha1, 20);
520 sha1write(f, pack_file_sha1, 20);
521 sha1close(f, NULL, 1);
524 static int locate_object_entry_hash(const unsigned char *sha1)
526 int i;
527 unsigned int ui;
528 memcpy(&ui, sha1, sizeof(unsigned int));
529 i = ui % object_ix_hashsz;
530 while (0 < object_ix[i]) {
531 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
532 return i;
533 if (++i == object_ix_hashsz)
534 i = 0;
536 return -1 - i;
539 static struct object_entry *locate_object_entry(const unsigned char *sha1)
541 int i;
543 if (!object_ix_hashsz)
544 return NULL;
546 i = locate_object_entry_hash(sha1);
547 if (0 <= i)
548 return &objects[object_ix[i]-1];
549 return NULL;
552 static void rehash_objects(void)
554 int i;
555 struct object_entry *oe;
557 object_ix_hashsz = nr_objects * 3;
558 if (object_ix_hashsz < 1024)
559 object_ix_hashsz = 1024;
560 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
561 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
562 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
563 int ix = locate_object_entry_hash(oe->sha1);
564 if (0 <= ix)
565 continue;
566 ix = -1 - ix;
567 object_ix[ix] = i + 1;
571 static unsigned name_hash(const char *name)
573 unsigned char c;
574 unsigned hash = 0;
577 * This effectively just creates a sortable number from the
578 * last sixteen non-whitespace characters. Last characters
579 * count "most", so things that end in ".c" sort together.
581 while ((c = *name++) != 0) {
582 if (isspace(c))
583 continue;
584 hash = (hash >> 2) + (c << 24);
586 return hash;
589 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
591 unsigned int idx = nr_objects;
592 struct object_entry *entry;
593 struct packed_git *p;
594 unsigned int found_offset = 0;
595 struct packed_git *found_pack = NULL;
596 int ix, status = 0;
598 if (!exclude) {
599 for (p = packed_git; p; p = p->next) {
600 unsigned long offset = find_pack_entry_one(sha1, p);
601 if (offset) {
602 if (incremental)
603 return 0;
604 if (local && !p->pack_local)
605 return 0;
606 if (!found_pack) {
607 found_offset = offset;
608 found_pack = p;
613 if ((entry = locate_object_entry(sha1)) != NULL)
614 goto already_added;
616 if (idx >= nr_alloc) {
617 unsigned int needed = (idx + 1024) * 3 / 2;
618 objects = xrealloc(objects, needed * sizeof(*entry));
619 nr_alloc = needed;
621 entry = objects + idx;
622 nr_objects = idx + 1;
623 memset(entry, 0, sizeof(*entry));
624 hashcpy(entry->sha1, sha1);
625 entry->hash = hash;
627 if (object_ix_hashsz * 3 <= nr_objects * 4)
628 rehash_objects();
629 else {
630 ix = locate_object_entry_hash(entry->sha1);
631 if (0 <= ix)
632 die("internal error in object hashing.");
633 object_ix[-1 - ix] = idx + 1;
635 status = 1;
637 already_added:
638 if (progress_update) {
639 fprintf(stderr, "Counting objects...%d\r", nr_objects);
640 progress_update = 0;
642 if (exclude)
643 entry->preferred_base = 1;
644 else {
645 if (found_pack) {
646 entry->in_pack = found_pack;
647 entry->in_pack_offset = found_offset;
650 return status;
653 struct pbase_tree_cache {
654 unsigned char sha1[20];
655 int ref;
656 int temporary;
657 void *tree_data;
658 unsigned long tree_size;
661 static struct pbase_tree_cache *(pbase_tree_cache[256]);
662 static int pbase_tree_cache_ix(const unsigned char *sha1)
664 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
666 static int pbase_tree_cache_ix_incr(int ix)
668 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
671 static struct pbase_tree {
672 struct pbase_tree *next;
673 /* This is a phony "cache" entry; we are not
674 * going to evict it nor find it through _get()
675 * mechanism -- this is for the toplevel node that
676 * would almost always change with any commit.
678 struct pbase_tree_cache pcache;
679 } *pbase_tree;
681 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
683 struct pbase_tree_cache *ent, *nent;
684 void *data;
685 unsigned long size;
686 char type[20];
687 int neigh;
688 int my_ix = pbase_tree_cache_ix(sha1);
689 int available_ix = -1;
691 /* pbase-tree-cache acts as a limited hashtable.
692 * your object will be found at your index or within a few
693 * slots after that slot if it is cached.
695 for (neigh = 0; neigh < 8; neigh++) {
696 ent = pbase_tree_cache[my_ix];
697 if (ent && !hashcmp(ent->sha1, sha1)) {
698 ent->ref++;
699 return ent;
701 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
702 ((0 <= available_ix) &&
703 (!ent && pbase_tree_cache[available_ix])))
704 available_ix = my_ix;
705 if (!ent)
706 break;
707 my_ix = pbase_tree_cache_ix_incr(my_ix);
710 /* Did not find one. Either we got a bogus request or
711 * we need to read and perhaps cache.
713 data = read_sha1_file(sha1, type, &size);
714 if (!data)
715 return NULL;
716 if (strcmp(type, tree_type)) {
717 free(data);
718 return NULL;
721 /* We need to either cache or return a throwaway copy */
723 if (available_ix < 0)
724 ent = NULL;
725 else {
726 ent = pbase_tree_cache[available_ix];
727 my_ix = available_ix;
730 if (!ent) {
731 nent = xmalloc(sizeof(*nent));
732 nent->temporary = (available_ix < 0);
734 else {
735 /* evict and reuse */
736 free(ent->tree_data);
737 nent = ent;
739 hashcpy(nent->sha1, sha1);
740 nent->tree_data = data;
741 nent->tree_size = size;
742 nent->ref = 1;
743 if (!nent->temporary)
744 pbase_tree_cache[my_ix] = nent;
745 return nent;
748 static void pbase_tree_put(struct pbase_tree_cache *cache)
750 if (!cache->temporary) {
751 cache->ref--;
752 return;
754 free(cache->tree_data);
755 free(cache);
758 static int name_cmp_len(const char *name)
760 int i;
761 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
763 return i;
766 static void add_pbase_object(struct tree_desc *tree,
767 const char *name,
768 int cmplen,
769 const char *fullname)
771 struct name_entry entry;
773 while (tree_entry(tree,&entry)) {
774 unsigned long size;
775 char type[20];
777 if (entry.pathlen != cmplen ||
778 memcmp(entry.path, name, cmplen) ||
779 !has_sha1_file(entry.sha1) ||
780 sha1_object_info(entry.sha1, type, &size))
781 continue;
782 if (name[cmplen] != '/') {
783 unsigned hash = name_hash(fullname);
784 add_object_entry(entry.sha1, hash, 1);
785 return;
787 if (!strcmp(type, tree_type)) {
788 struct tree_desc sub;
789 struct pbase_tree_cache *tree;
790 const char *down = name+cmplen+1;
791 int downlen = name_cmp_len(down);
793 tree = pbase_tree_get(entry.sha1);
794 if (!tree)
795 return;
796 sub.buf = tree->tree_data;
797 sub.size = tree->tree_size;
799 add_pbase_object(&sub, down, downlen, fullname);
800 pbase_tree_put(tree);
805 static unsigned *done_pbase_paths;
806 static int done_pbase_paths_num;
807 static int done_pbase_paths_alloc;
808 static int done_pbase_path_pos(unsigned hash)
810 int lo = 0;
811 int hi = done_pbase_paths_num;
812 while (lo < hi) {
813 int mi = (hi + lo) / 2;
814 if (done_pbase_paths[mi] == hash)
815 return mi;
816 if (done_pbase_paths[mi] < hash)
817 hi = mi;
818 else
819 lo = mi + 1;
821 return -lo-1;
824 static int check_pbase_path(unsigned hash)
826 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
827 if (0 <= pos)
828 return 1;
829 pos = -pos - 1;
830 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
831 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
832 done_pbase_paths = xrealloc(done_pbase_paths,
833 done_pbase_paths_alloc *
834 sizeof(unsigned));
836 done_pbase_paths_num++;
837 if (pos < done_pbase_paths_num)
838 memmove(done_pbase_paths + pos + 1,
839 done_pbase_paths + pos,
840 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
841 done_pbase_paths[pos] = hash;
842 return 0;
845 static void add_preferred_base_object(const char *name, unsigned hash)
847 struct pbase_tree *it;
848 int cmplen = name_cmp_len(name);
850 if (check_pbase_path(hash))
851 return;
853 for (it = pbase_tree; it; it = it->next) {
854 if (cmplen == 0) {
855 hash = name_hash("");
856 add_object_entry(it->pcache.sha1, hash, 1);
858 else {
859 struct tree_desc tree;
860 tree.buf = it->pcache.tree_data;
861 tree.size = it->pcache.tree_size;
862 add_pbase_object(&tree, name, cmplen, name);
867 static void add_preferred_base(unsigned char *sha1)
869 struct pbase_tree *it;
870 void *data;
871 unsigned long size;
872 unsigned char tree_sha1[20];
874 if (window <= num_preferred_base++)
875 return;
877 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
878 if (!data)
879 return;
881 for (it = pbase_tree; it; it = it->next) {
882 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
883 free(data);
884 return;
888 it = xcalloc(1, sizeof(*it));
889 it->next = pbase_tree;
890 pbase_tree = it;
892 hashcpy(it->pcache.sha1, tree_sha1);
893 it->pcache.tree_data = data;
894 it->pcache.tree_size = size;
897 static void check_object(struct object_entry *entry)
899 char type[20];
901 if (entry->in_pack && !entry->preferred_base) {
902 unsigned char base[20];
903 unsigned long size;
904 struct object_entry *base_entry;
906 /* We want in_pack_type even if we do not reuse delta.
907 * There is no point not reusing non-delta representations.
909 check_reuse_pack_delta(entry->in_pack,
910 entry->in_pack_offset,
911 base, &size,
912 &entry->in_pack_type);
914 /* Check if it is delta, and the base is also an object
915 * we are going to pack. If so we will reuse the existing
916 * delta.
918 if (!no_reuse_delta &&
919 entry->in_pack_type == OBJ_DELTA &&
920 (base_entry = locate_object_entry(base)) &&
921 (!base_entry->preferred_base)) {
923 /* Depth value does not matter - find_deltas()
924 * will never consider reused delta as the
925 * base object to deltify other objects
926 * against, in order to avoid circular deltas.
929 /* uncompressed size of the delta data */
930 entry->size = entry->delta_size = size;
931 entry->delta = base_entry;
932 entry->type = OBJ_DELTA;
934 entry->delta_sibling = base_entry->delta_child;
935 base_entry->delta_child = entry;
937 return;
939 /* Otherwise we would do the usual */
942 if (sha1_object_info(entry->sha1, type, &entry->size))
943 die("unable to get type of object %s",
944 sha1_to_hex(entry->sha1));
946 if (!strcmp(type, commit_type)) {
947 entry->type = OBJ_COMMIT;
948 } else if (!strcmp(type, tree_type)) {
949 entry->type = OBJ_TREE;
950 } else if (!strcmp(type, blob_type)) {
951 entry->type = OBJ_BLOB;
952 } else if (!strcmp(type, tag_type)) {
953 entry->type = OBJ_TAG;
954 } else
955 die("unable to pack object %s of type %s",
956 sha1_to_hex(entry->sha1), type);
959 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
961 struct object_entry *child = me->delta_child;
962 unsigned int m = n;
963 while (child) {
964 unsigned int c = check_delta_limit(child, n + 1);
965 if (m < c)
966 m = c;
967 child = child->delta_sibling;
969 return m;
972 static void get_object_details(void)
974 int i;
975 struct object_entry *entry;
977 prepare_pack_ix();
978 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
979 check_object(entry);
981 if (nr_objects == nr_result) {
983 * Depth of objects that depend on the entry -- this
984 * is subtracted from depth-max to break too deep
985 * delta chain because of delta data reusing.
986 * However, we loosen this restriction when we know we
987 * are creating a thin pack -- it will have to be
988 * expanded on the other end anyway, so do not
989 * artificially cut the delta chain and let it go as
990 * deep as it wants.
992 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
993 if (!entry->delta && entry->delta_child)
994 entry->delta_limit =
995 check_delta_limit(entry, 1);
999 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1001 static entry_sort_t current_sort;
1003 static int sort_comparator(const void *_a, const void *_b)
1005 struct object_entry *a = *(struct object_entry **)_a;
1006 struct object_entry *b = *(struct object_entry **)_b;
1007 return current_sort(a,b);
1010 static struct object_entry **create_sorted_list(entry_sort_t sort)
1012 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1013 int i;
1015 for (i = 0; i < nr_objects; i++)
1016 list[i] = objects + i;
1017 current_sort = sort;
1018 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1019 return list;
1022 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1024 return hashcmp(a->sha1, b->sha1);
1027 static struct object_entry **create_final_object_list(void)
1029 struct object_entry **list;
1030 int i, j;
1032 for (i = nr_result = 0; i < nr_objects; i++)
1033 if (!objects[i].preferred_base)
1034 nr_result++;
1035 list = xmalloc(nr_result * sizeof(struct object_entry *));
1036 for (i = j = 0; i < nr_objects; i++) {
1037 if (!objects[i].preferred_base)
1038 list[j++] = objects + i;
1040 current_sort = sha1_sort;
1041 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1042 return list;
1045 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1047 if (a->type < b->type)
1048 return -1;
1049 if (a->type > b->type)
1050 return 1;
1051 if (a->hash < b->hash)
1052 return -1;
1053 if (a->hash > b->hash)
1054 return 1;
1055 if (a->preferred_base < b->preferred_base)
1056 return -1;
1057 if (a->preferred_base > b->preferred_base)
1058 return 1;
1059 if (a->size < b->size)
1060 return -1;
1061 if (a->size > b->size)
1062 return 1;
1063 return a < b ? -1 : (a > b);
1066 struct unpacked {
1067 struct object_entry *entry;
1068 void *data;
1069 struct delta_index *index;
1073 * We search for deltas _backwards_ in a list sorted by type and
1074 * by size, so that we see progressively smaller and smaller files.
1075 * That's because we prefer deltas to be from the bigger file
1076 * to the smaller - deletes are potentially cheaper, but perhaps
1077 * more importantly, the bigger file is likely the more recent
1078 * one.
1080 static int try_delta(struct unpacked *trg, struct unpacked *src,
1081 unsigned max_depth)
1083 struct object_entry *trg_entry = trg->entry;
1084 struct object_entry *src_entry = src->entry;
1085 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1086 char type[10];
1087 void *delta_buf;
1089 /* Don't bother doing diffs between different types */
1090 if (trg_entry->type != src_entry->type)
1091 return -1;
1093 /* We do not compute delta to *create* objects we are not
1094 * going to pack.
1096 if (trg_entry->preferred_base)
1097 return -1;
1100 * We do not bother to try a delta that we discarded
1101 * on an earlier try, but only when reusing delta data.
1103 if (!no_reuse_delta && trg_entry->in_pack &&
1104 trg_entry->in_pack == src_entry->in_pack)
1105 return 0;
1108 * If the current object is at pack edge, take the depth the
1109 * objects that depend on the current object into account --
1110 * otherwise they would become too deep.
1112 if (trg_entry->delta_child) {
1113 if (max_depth <= trg_entry->delta_limit)
1114 return 0;
1115 max_depth -= trg_entry->delta_limit;
1117 if (src_entry->depth >= max_depth)
1118 return 0;
1120 /* Now some size filtering heuristics. */
1121 trg_size = trg_entry->size;
1122 max_size = trg_size/2 - 20;
1123 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1124 if (max_size == 0)
1125 return 0;
1126 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1127 max_size = trg_entry->delta_size-1;
1128 src_size = src_entry->size;
1129 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1130 if (sizediff >= max_size)
1131 return 0;
1133 /* Load data if not already done */
1134 if (!trg->data) {
1135 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1136 if (sz != trg_size)
1137 die("object %s inconsistent object length (%lu vs %lu)",
1138 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1140 if (!src->data) {
1141 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1142 if (sz != src_size)
1143 die("object %s inconsistent object length (%lu vs %lu)",
1144 sha1_to_hex(src_entry->sha1), sz, src_size);
1146 if (!src->index) {
1147 src->index = create_delta_index(src->data, src_size);
1148 if (!src->index)
1149 die("out of memory");
1152 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1153 if (!delta_buf)
1154 return 0;
1156 trg_entry->delta = src_entry;
1157 trg_entry->delta_size = delta_size;
1158 trg_entry->depth = src_entry->depth + 1;
1159 free(delta_buf);
1160 return 1;
1163 static void progress_interval(int signum)
1165 progress_update = 1;
1168 static void find_deltas(struct object_entry **list, int window, int depth)
1170 int i, idx;
1171 unsigned int array_size = window * sizeof(struct unpacked);
1172 struct unpacked *array = xmalloc(array_size);
1173 unsigned processed = 0;
1174 unsigned last_percent = 999;
1176 memset(array, 0, array_size);
1177 i = nr_objects;
1178 idx = 0;
1179 if (progress)
1180 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1182 while (--i >= 0) {
1183 struct object_entry *entry = list[i];
1184 struct unpacked *n = array + idx;
1185 int j;
1187 if (!entry->preferred_base)
1188 processed++;
1190 if (progress) {
1191 unsigned percent = processed * 100 / nr_result;
1192 if (percent != last_percent || progress_update) {
1193 fprintf(stderr, "%4u%% (%u/%u) done\r",
1194 percent, processed, nr_result);
1195 progress_update = 0;
1196 last_percent = percent;
1200 if (entry->delta)
1201 /* This happens if we decided to reuse existing
1202 * delta from a pack. "!no_reuse_delta &&" is implied.
1204 continue;
1206 if (entry->size < 50)
1207 continue;
1208 free_delta_index(n->index);
1209 n->index = NULL;
1210 free(n->data);
1211 n->data = NULL;
1212 n->entry = entry;
1214 j = window;
1215 while (--j > 0) {
1216 unsigned int other_idx = idx + j;
1217 struct unpacked *m;
1218 if (other_idx >= window)
1219 other_idx -= window;
1220 m = array + other_idx;
1221 if (!m->entry)
1222 break;
1223 if (try_delta(n, m, depth) < 0)
1224 break;
1226 /* if we made n a delta, and if n is already at max
1227 * depth, leaving it in the window is pointless. we
1228 * should evict it first.
1230 if (entry->delta && depth <= entry->depth)
1231 continue;
1233 idx++;
1234 if (idx >= window)
1235 idx = 0;
1238 if (progress)
1239 fputc('\n', stderr);
1241 for (i = 0; i < window; ++i) {
1242 free_delta_index(array[i].index);
1243 free(array[i].data);
1245 free(array);
1248 static void prepare_pack(int window, int depth)
1250 get_object_details();
1251 sorted_by_type = create_sorted_list(type_size_sort);
1252 if (window && depth)
1253 find_deltas(sorted_by_type, window+1, depth);
1256 static int reuse_cached_pack(unsigned char *sha1)
1258 static const char cache[] = "pack-cache/pack-%s.%s";
1259 char *cached_pack, *cached_idx;
1260 int ifd, ofd, ifd_ix = -1;
1262 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1263 ifd = open(cached_pack, O_RDONLY);
1264 if (ifd < 0)
1265 return 0;
1267 if (!pack_to_stdout) {
1268 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1269 ifd_ix = open(cached_idx, O_RDONLY);
1270 if (ifd_ix < 0) {
1271 close(ifd);
1272 return 0;
1276 if (progress)
1277 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1278 sha1_to_hex(sha1));
1280 if (pack_to_stdout) {
1281 if (copy_fd(ifd, 1))
1282 exit(1);
1283 close(ifd);
1285 else {
1286 char name[PATH_MAX];
1287 snprintf(name, sizeof(name),
1288 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1289 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1290 if (ofd < 0)
1291 die("unable to open %s (%s)", name, strerror(errno));
1292 if (copy_fd(ifd, ofd))
1293 exit(1);
1294 close(ifd);
1296 snprintf(name, sizeof(name),
1297 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1298 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1299 if (ofd < 0)
1300 die("unable to open %s (%s)", name, strerror(errno));
1301 if (copy_fd(ifd_ix, ofd))
1302 exit(1);
1303 close(ifd_ix);
1304 puts(sha1_to_hex(sha1));
1307 return 1;
1310 static void setup_progress_signal(void)
1312 struct sigaction sa;
1313 struct itimerval v;
1315 memset(&sa, 0, sizeof(sa));
1316 sa.sa_handler = progress_interval;
1317 sigemptyset(&sa.sa_mask);
1318 sa.sa_flags = SA_RESTART;
1319 sigaction(SIGALRM, &sa, NULL);
1321 v.it_interval.tv_sec = 1;
1322 v.it_interval.tv_usec = 0;
1323 v.it_value = v.it_interval;
1324 setitimer(ITIMER_REAL, &v, NULL);
1327 static int git_pack_config(const char *k, const char *v)
1329 if(!strcmp(k, "pack.window")) {
1330 window = git_config_int(k, v);
1331 return 0;
1333 return git_default_config(k, v);
1336 static void read_object_list_from_stdin(void)
1338 char line[40 + 1 + PATH_MAX + 2];
1339 unsigned char sha1[20];
1340 unsigned hash;
1342 for (;;) {
1343 if (!fgets(line, sizeof(line), stdin)) {
1344 if (feof(stdin))
1345 break;
1346 if (!ferror(stdin))
1347 die("fgets returned NULL, not EOF, not error!");
1348 if (errno != EINTR)
1349 die("fgets: %s", strerror(errno));
1350 clearerr(stdin);
1351 continue;
1353 if (line[0] == '-') {
1354 if (get_sha1_hex(line+1, sha1))
1355 die("expected edge sha1, got garbage:\n %s",
1356 line);
1357 add_preferred_base(sha1);
1358 continue;
1360 if (get_sha1_hex(line, sha1))
1361 die("expected sha1, got garbage:\n %s", line);
1363 hash = name_hash(line+41);
1364 add_preferred_base_object(line+41, hash);
1365 add_object_entry(sha1, hash, 0);
1369 static void show_commit(struct commit *commit)
1371 unsigned hash = name_hash("");
1372 add_preferred_base_object("", hash);
1373 add_object_entry(commit->object.sha1, hash, 0);
1376 static void show_object(struct object_array_entry *p)
1378 unsigned hash = name_hash(p->name);
1379 add_preferred_base_object(p->name, hash);
1380 add_object_entry(p->item->sha1, hash, 0);
1383 static void show_edge(struct commit *commit)
1385 add_preferred_base(commit->object.sha1);
1388 static void get_object_list(int ac, const char **av)
1390 struct rev_info revs;
1391 char line[1000];
1392 int flags = 0;
1394 init_revisions(&revs, NULL);
1395 save_commit_buffer = 0;
1396 track_object_refs = 0;
1397 setup_revisions(ac, av, &revs, NULL);
1399 while (fgets(line, sizeof(line), stdin) != NULL) {
1400 int len = strlen(line);
1401 if (line[len - 1] == '\n')
1402 line[--len] = 0;
1403 if (!len)
1404 break;
1405 if (*line == '-') {
1406 if (!strcmp(line, "--not")) {
1407 flags ^= UNINTERESTING;
1408 continue;
1410 die("not a rev '%s'", line);
1412 if (handle_revision_arg(line, &revs, flags, 1))
1413 die("bad revision '%s'", line);
1416 prepare_revision_walk(&revs);
1417 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1418 traverse_commit_list(&revs, show_commit, show_object);
1421 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1423 SHA_CTX ctx;
1424 int depth = 10;
1425 struct object_entry **list;
1426 int use_internal_rev_list = 0;
1427 int thin = 0;
1428 int i;
1429 const char *rp_av[64];
1430 int rp_ac;
1432 rp_av[0] = "pack-objects";
1433 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1434 rp_ac = 2;
1436 git_config(git_pack_config);
1438 progress = isatty(2);
1439 for (i = 1; i < argc; i++) {
1440 const char *arg = argv[i];
1442 if (*arg != '-')
1443 break;
1445 if (!strcmp("--non-empty", arg)) {
1446 non_empty = 1;
1447 continue;
1449 if (!strcmp("--local", arg)) {
1450 local = 1;
1451 continue;
1453 if (!strcmp("--progress", arg)) {
1454 progress = 1;
1455 continue;
1457 if (!strcmp("--incremental", arg)) {
1458 incremental = 1;
1459 continue;
1461 if (!strncmp("--window=", arg, 9)) {
1462 char *end;
1463 window = strtoul(arg+9, &end, 0);
1464 if (!arg[9] || *end)
1465 usage(pack_usage);
1466 continue;
1468 if (!strncmp("--depth=", arg, 8)) {
1469 char *end;
1470 depth = strtoul(arg+8, &end, 0);
1471 if (!arg[8] || *end)
1472 usage(pack_usage);
1473 continue;
1475 if (!strcmp("--progress", arg)) {
1476 progress = 1;
1477 continue;
1479 if (!strcmp("-q", arg)) {
1480 progress = 0;
1481 continue;
1483 if (!strcmp("--no-reuse-delta", arg)) {
1484 no_reuse_delta = 1;
1485 continue;
1487 if (!strcmp("--stdout", arg)) {
1488 pack_to_stdout = 1;
1489 continue;
1491 if (!strcmp("--revs", arg)) {
1492 use_internal_rev_list = 1;
1493 continue;
1495 if (!strcmp("--unpacked", arg) ||
1496 !strncmp("--unpacked=", arg, 11) ||
1497 !strcmp("--all", arg)) {
1498 use_internal_rev_list = 1;
1499 if (ARRAY_SIZE(rp_av) - 1 <= rp_ac)
1500 die("too many internal rev-list options");
1501 rp_av[rp_ac++] = arg;
1502 continue;
1504 if (!strcmp("--thin", arg)) {
1505 use_internal_rev_list = 1;
1506 thin = 1;
1507 rp_av[1] = "--objects-edge";
1508 continue;
1510 usage(pack_usage);
1513 /* Traditionally "pack-objects [options] base extra" failed;
1514 * we would however want to take refs parameter that would
1515 * have been given to upstream rev-list ourselves, which means
1516 * we somehow want to say what the base name is. So the
1517 * syntax would be:
1519 * pack-objects [options] base <refs...>
1521 * in other words, we would treat the first non-option as the
1522 * base_name and send everything else to the internal revision
1523 * walker.
1526 if (!pack_to_stdout)
1527 base_name = argv[i++];
1529 if (pack_to_stdout != !base_name)
1530 usage(pack_usage);
1532 if (!pack_to_stdout && thin)
1533 die("--thin cannot be used to build an indexable pack.");
1535 prepare_packed_git();
1537 if (progress) {
1538 fprintf(stderr, "Generating pack...\n");
1539 setup_progress_signal();
1542 if (!use_internal_rev_list)
1543 read_object_list_from_stdin();
1544 else {
1545 rp_av[rp_ac] = NULL;
1546 get_object_list(rp_ac, rp_av);
1549 if (progress)
1550 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1551 sorted_by_sha = create_final_object_list();
1552 if (non_empty && !nr_result)
1553 return 0;
1555 SHA1_Init(&ctx);
1556 list = sorted_by_sha;
1557 for (i = 0; i < nr_result; i++) {
1558 struct object_entry *entry = *list++;
1559 SHA1_Update(&ctx, entry->sha1, 20);
1561 SHA1_Final(object_list_sha1, &ctx);
1562 if (progress && (nr_objects != nr_result))
1563 fprintf(stderr, "Result has %d objects.\n", nr_result);
1565 if (reuse_cached_pack(object_list_sha1))
1567 else {
1568 if (nr_result)
1569 prepare_pack(window, depth);
1570 if (progress && pack_to_stdout) {
1571 /* the other end usually displays progress itself */
1572 struct itimerval v = {{0,},};
1573 setitimer(ITIMER_REAL, &v, NULL);
1574 signal(SIGALRM, SIG_IGN );
1575 progress_update = 0;
1577 write_pack_file();
1578 if (!pack_to_stdout) {
1579 write_index_file();
1580 puts(sha1_to_hex(object_list_sha1));
1583 if (progress)
1584 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1585 nr_result, written, written_delta, reused, reused_delta);
1586 return 0;