10 #include "csum-file.h"
11 #include "tree-walk.h"
14 #include "list-objects.h"
17 static const char pack_usage[] = "\
18 git-pack-objects [{ -q | --progress | --all-progress }] \n\
19 [--local] [--incremental] [--window=N] [--depth=N] \n\
20 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
21 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
22 [--stdout | base-name] [<ref-list | <object-list]";
25 unsigned char sha1[20];
26 uint32_t crc32; /* crc of raw pack data for this object */
27 off_t offset; /* offset into the final pack file */
28 unsigned long size; /* uncompressed size */
29 unsigned int hash; /* name hint hash */
30 unsigned int depth; /* delta depth */
31 struct packed_git *in_pack; /* already in pack */
33 struct object_entry *delta; /* delta base object */
34 struct object_entry *delta_child; /* deltified objects who bases me */
35 struct object_entry *delta_sibling; /* other deltified objects who
36 * uses the same base as me
38 unsigned long delta_size; /* delta data size (uncompressed) */
39 enum object_type type;
40 enum object_type in_pack_type; /* could be delta */
41 unsigned char in_pack_header_size;
42 unsigned char preferred_base; /* we do not pack this, but is available
43 * to be used as the base objectto delta
49 * Objects we are going to pack are collected in objects array (dynamically
50 * expanded). nr_objects & nr_alloc controls this array. They are stored
51 * in the order we see -- typically rev-list --objects order that gives us
52 * nice "minimum seek" order.
54 static struct object_entry *objects;
55 static struct object_entry **written_list;
56 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
59 static int no_reuse_delta, no_reuse_object;
61 static int incremental;
62 static int allow_ofs_delta;
63 static const char *pack_tmp_name, *idx_tmp_name;
64 static char tmpname[PATH_MAX];
65 static const char *base_name;
66 static unsigned char pack_file_sha1[20];
67 static int progress = 1;
68 static int window = 10;
69 static uint32_t pack_size_limit;
70 static int depth = 50;
71 static int pack_to_stdout;
72 static int num_preferred_base;
73 static struct progress progress_state;
74 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
75 static int pack_compression_seen;
78 * The object names in objects array are hashed with this hashtable,
79 * to help looking up the entry by object name.
80 * This hashtable is built after all the objects are seen.
82 static int *object_ix;
83 static int object_ix_hashsz;
86 * Pack index for existing packs give us easy access to the offsets into
87 * corresponding pack file where each object's data starts, but the entries
88 * do not store the size of the compressed representation (uncompressed
89 * size is easily available by examining the pack entry header). It is
90 * also rather expensive to find the sha1 for an object given its offset.
92 * We build a hashtable of existing packs (pack_revindex), and keep reverse
93 * index here -- pack index file is sorted by object name mapping to offset;
94 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
95 * ordered by offset, so if you know the offset of an object, next offset
96 * is where its packed representation ends and the index_nr can be used to
97 * get the object sha1 from the main index.
99 struct revindex_entry {
103 struct pack_revindex {
104 struct packed_git *p;
105 struct revindex_entry *revindex;
107 static struct pack_revindex *pack_revindex;
108 static int pack_revindex_hashsz;
113 static uint32_t written, written_delta;
114 static uint32_t reused, reused_delta;
116 static int pack_revindex_ix(struct packed_git *p)
118 unsigned long ui = (unsigned long)p;
121 ui = ui ^ (ui >> 16); /* defeat structure alignment */
122 i = (int)(ui % pack_revindex_hashsz);
123 while (pack_revindex[i].p) {
124 if (pack_revindex[i].p == p)
126 if (++i == pack_revindex_hashsz)
132 static void prepare_pack_ix(void)
135 struct packed_git *p;
136 for (num = 0, p = packed_git; p; p = p->next)
140 pack_revindex_hashsz = num * 11;
141 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
142 for (p = packed_git; p; p = p->next) {
143 num = pack_revindex_ix(p);
145 pack_revindex[num].p = p;
147 /* revindex elements are lazily initialized */
150 static int cmp_offset(const void *a_, const void *b_)
152 const struct revindex_entry *a = a_;
153 const struct revindex_entry *b = b_;
154 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
158 * Ordered list of offsets of objects in the pack.
160 static void prepare_pack_revindex(struct pack_revindex *rix)
162 struct packed_git *p = rix->p;
163 int num_ent = p->num_objects;
165 const char *index = p->index_data;
167 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
170 if (p->index_version > 1) {
171 const uint32_t *off_32 =
172 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
173 const uint32_t *off_64 = off_32 + p->num_objects;
174 for (i = 0; i < num_ent; i++) {
175 uint32_t off = ntohl(*off_32++);
176 if (!(off & 0x80000000)) {
177 rix->revindex[i].offset = off;
179 rix->revindex[i].offset =
180 ((uint64_t)ntohl(*off_64++)) << 32;
181 rix->revindex[i].offset |=
184 rix->revindex[i].nr = i;
187 for (i = 0; i < num_ent; i++) {
188 uint32_t hl = *((uint32_t *)(index + 24 * i));
189 rix->revindex[i].offset = ntohl(hl);
190 rix->revindex[i].nr = i;
194 /* This knows the pack format -- the 20-byte trailer
195 * follows immediately after the last object data.
197 rix->revindex[num_ent].offset = p->pack_size - 20;
198 rix->revindex[num_ent].nr = -1;
199 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
202 static struct revindex_entry * find_packed_object(struct packed_git *p,
207 struct pack_revindex *rix;
208 struct revindex_entry *revindex;
209 num = pack_revindex_ix(p);
211 die("internal error: pack revindex uninitialized");
212 rix = &pack_revindex[num];
214 prepare_pack_revindex(rix);
215 revindex = rix->revindex;
217 hi = p->num_objects + 1;
219 int mi = (lo + hi) / 2;
220 if (revindex[mi].offset == ofs) {
221 return revindex + mi;
223 else if (ofs < revindex[mi].offset)
228 die("internal error: pack revindex corrupt");
231 static const unsigned char *find_packed_object_name(struct packed_git *p,
234 struct revindex_entry *entry = find_packed_object(p, ofs);
235 return nth_packed_object_sha1(p, entry->nr);
238 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
240 unsigned long othersize, delta_size;
241 enum object_type type;
242 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
246 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
247 delta_buf = diff_delta(otherbuf, othersize,
248 buf, size, &delta_size, 0);
249 if (!delta_buf || delta_size != entry->delta_size)
250 die("delta size changed");
257 * The per-object header is a pretty dense thing, which is
258 * - first byte: low four bits are "size", then three bits of "type",
259 * and the high bit is "size continues".
260 * - each byte afterwards: low seven bits are size continuation,
261 * with the high bit being "size continues"
263 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
268 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
269 die("bad type %d", type);
271 c = (type << 4) | (size & 15);
284 * we are going to reuse the existing object data as is. make
285 * sure it is not corrupt.
287 static int check_pack_inflate(struct packed_git *p,
288 struct pack_window **w_curs,
291 unsigned long expect)
294 unsigned char fakebuf[4096], *in;
297 memset(&stream, 0, sizeof(stream));
298 inflateInit(&stream);
300 in = use_pack(p, w_curs, offset, &stream.avail_in);
302 stream.next_out = fakebuf;
303 stream.avail_out = sizeof(fakebuf);
304 st = inflate(&stream, Z_FINISH);
305 offset += stream.next_in - in;
306 } while (st == Z_OK || st == Z_BUF_ERROR);
308 return (st == Z_STREAM_END &&
309 stream.total_out == expect &&
310 stream.total_in == len) ? 0 : -1;
313 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
314 off_t offset, off_t len, unsigned int nr)
316 const uint32_t *index_crc;
317 uint32_t data_crc = crc32(0, Z_NULL, 0);
321 void *data = use_pack(p, w_curs, offset, &avail);
324 data_crc = crc32(data_crc, data, avail);
329 index_crc = p->index_data;
330 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
332 return data_crc != ntohl(*index_crc);
335 static void copy_pack_data(struct sha1file *f,
336 struct packed_git *p,
337 struct pack_window **w_curs,
345 in = use_pack(p, w_curs, offset, &avail);
347 avail = (unsigned int)len;
348 sha1write(f, in, avail);
354 static unsigned long write_object(struct sha1file *f,
355 struct object_entry *entry)
358 enum object_type type;
360 unsigned char header[10];
363 enum object_type obj_type;
369 obj_type = entry->type;
371 to_reuse = 0; /* explicit */
372 else if (!entry->in_pack)
373 to_reuse = 0; /* can't reuse what we don't have */
374 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
375 to_reuse = 1; /* check_object() decided it for us */
376 else if (obj_type != entry->in_pack_type)
377 to_reuse = 0; /* pack has delta which is unusable */
378 else if (entry->delta)
379 to_reuse = 0; /* we want to pack afresh */
381 to_reuse = 1; /* we have it in-pack undeltified,
382 * and we do not need to deltify it.
386 buf = read_sha1_file(entry->sha1, &type, &size);
388 die("unable to read %s", sha1_to_hex(entry->sha1));
389 if (size != entry->size)
390 die("object %s size inconsistency (%lu vs %lu)",
391 sha1_to_hex(entry->sha1), size, entry->size);
393 buf = delta_against(buf, size, entry);
394 size = entry->delta_size;
395 obj_type = (allow_ofs_delta && entry->delta->offset) ?
396 OBJ_OFS_DELTA : OBJ_REF_DELTA;
399 * The object header is a byte of 'type' followed by zero or
400 * more bytes of length.
402 hdrlen = encode_header(obj_type, size, header);
403 sha1write(f, header, hdrlen);
405 if (obj_type == OBJ_OFS_DELTA) {
407 * Deltas with relative base contain an additional
408 * encoding of the relative offset for the delta
409 * base from this object's position in the pack.
411 off_t ofs = entry->offset - entry->delta->offset;
412 unsigned pos = sizeof(header) - 1;
413 header[pos] = ofs & 127;
415 header[--pos] = 128 | (--ofs & 127);
416 sha1write(f, header + pos, sizeof(header) - pos);
417 hdrlen += sizeof(header) - pos;
418 } else if (obj_type == OBJ_REF_DELTA) {
420 * Deltas with a base reference contain
421 * an additional 20 bytes for the base sha1.
423 sha1write(f, entry->delta->sha1, 20);
426 datalen = sha1write_compressed(f, buf, size, pack_compression_level);
430 struct packed_git *p = entry->in_pack;
431 struct pack_window *w_curs = NULL;
432 struct revindex_entry *revidx;
436 obj_type = (allow_ofs_delta && entry->delta->offset) ?
437 OBJ_OFS_DELTA : OBJ_REF_DELTA;
440 hdrlen = encode_header(obj_type, entry->size, header);
441 sha1write(f, header, hdrlen);
442 if (obj_type == OBJ_OFS_DELTA) {
443 off_t ofs = entry->offset - entry->delta->offset;
444 unsigned pos = sizeof(header) - 1;
445 header[pos] = ofs & 127;
447 header[--pos] = 128 | (--ofs & 127);
448 sha1write(f, header + pos, sizeof(header) - pos);
449 hdrlen += sizeof(header) - pos;
450 } else if (obj_type == OBJ_REF_DELTA) {
451 sha1write(f, entry->delta->sha1, 20);
455 offset = entry->in_pack_offset;
456 revidx = find_packed_object(p, offset);
457 datalen = revidx[1].offset - offset;
458 if (!pack_to_stdout && p->index_version > 1 &&
459 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
460 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
461 offset += entry->in_pack_header_size;
462 datalen -= entry->in_pack_header_size;
463 if (!pack_to_stdout && p->index_version == 1 &&
464 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
465 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
466 copy_pack_data(f, p, &w_curs, offset, datalen);
474 entry->crc32 = crc32_end(f);
475 return hdrlen + datalen;
478 static off_t write_one(struct sha1file *f,
479 struct object_entry *e,
484 /* offset is non zero if object is written already. */
485 if (e->offset || e->preferred_base)
488 /* if we are deltified, write out base object first. */
490 offset = write_one(f, e->delta, offset);
493 size = write_object(f, e);
495 /* make sure off_t is sufficiently large not to wrap */
496 if (offset > offset + size)
497 die("pack too large for current definition of off_t");
498 return offset + size;
501 static int open_object_dir_tmp(const char *path)
503 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
504 return mkstemp(tmpname);
507 /* forward declarations for write_pack_file */
508 static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
509 static int adjust_perm(const char *path, mode_t mode);
511 static void write_pack_file(void)
515 off_t offset, last_obj_offset = 0;
516 struct pack_header hdr;
517 int do_progress = progress;
519 if (pack_to_stdout) {
520 f = sha1fd(1, "<stdout>");
523 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
525 die("unable to create %s: %s\n", tmpname, strerror(errno));
526 pack_tmp_name = xstrdup(tmpname);
527 f = sha1fd(fd, pack_tmp_name);
531 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
533 hdr.hdr_signature = htonl(PACK_SIGNATURE);
534 hdr.hdr_version = htonl(PACK_VERSION);
535 hdr.hdr_entries = htonl(nr_result);
536 sha1write(f, &hdr, sizeof(hdr));
537 offset = sizeof(hdr);
540 for (i = 0; i < nr_objects; i++) {
541 last_obj_offset = offset;
542 offset = write_one(f, objects + i, offset);
544 display_progress(&progress_state, written);
547 stop_progress(&progress_state);
549 if (written != nr_result)
550 die("wrote %u objects while expecting %u", written, nr_result);
551 sha1close(f, pack_file_sha1, 1);
553 if (!pack_to_stdout) {
554 unsigned char object_list_sha1[20];
555 mode_t mode = umask(0);
560 write_index_file(last_obj_offset, object_list_sha1);
561 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
562 base_name, sha1_to_hex(object_list_sha1));
563 if (adjust_perm(pack_tmp_name, mode))
564 die("unable to make temporary pack file readable: %s",
566 if (rename(pack_tmp_name, tmpname))
567 die("unable to rename temporary pack file: %s",
569 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
570 base_name, sha1_to_hex(object_list_sha1));
571 if (adjust_perm(idx_tmp_name, mode))
572 die("unable to make temporary index file readable: %s",
574 if (rename(idx_tmp_name, tmpname))
575 die("unable to rename temporary index file: %s",
577 puts(sha1_to_hex(object_list_sha1));
581 static int sha1_sort(const void *_a, const void *_b)
583 const struct object_entry *a = *(struct object_entry **)_a;
584 const struct object_entry *b = *(struct object_entry **)_b;
585 return hashcmp(a->sha1, b->sha1);
588 static uint32_t index_default_version = 1;
589 static uint32_t index_off32_limit = 0x7fffffff;
591 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
594 struct object_entry **sorted_by_sha, **list, **last;
596 uint32_t i, index_version;
599 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
601 die("unable to create %s: %s\n", tmpname, strerror(errno));
602 idx_tmp_name = xstrdup(tmpname);
603 f = sha1fd(fd, idx_tmp_name);
608 xcalloc(nr_result, sizeof(struct object_entry *));
609 for (i = 0; i < nr_objects; i++)
610 if (!objects[i].preferred_base)
611 sorted_by_sha[j++] = objects + i;
613 die("listed %u objects while expecting %u", j, nr_result);
614 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
615 list = sorted_by_sha;
616 last = sorted_by_sha + nr_result;
618 sorted_by_sha = list = last = NULL;
620 /* if last object's offset is >= 2^31 we should use index V2 */
621 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
623 /* index versions 2 and above need a header */
624 if (index_version >= 2) {
625 struct pack_idx_header hdr;
626 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
627 hdr.idx_version = htonl(index_version);
628 sha1write(f, &hdr, sizeof(hdr));
632 * Write the first-level table (the list is sorted,
633 * but we use a 256-entry lookup to be able to avoid
634 * having to do eight extra binary search iterations).
636 for (i = 0; i < 256; i++) {
637 struct object_entry **next = list;
638 while (next < last) {
639 struct object_entry *entry = *next;
640 if (entry->sha1[0] != i)
644 array[i] = htonl(next - sorted_by_sha);
647 sha1write(f, array, 256 * 4);
649 /* Compute the SHA1 hash of sorted object names. */
652 /* Write the actual SHA1 entries. */
653 list = sorted_by_sha;
654 for (i = 0; i < nr_result; i++) {
655 struct object_entry *entry = *list++;
656 if (index_version < 2) {
657 uint32_t offset = htonl(entry->offset);
658 sha1write(f, &offset, 4);
660 sha1write(f, entry->sha1, 20);
661 SHA1_Update(&ctx, entry->sha1, 20);
664 if (index_version >= 2) {
665 unsigned int nr_large_offset = 0;
667 /* write the crc32 table */
668 list = sorted_by_sha;
669 for (i = 0; i < nr_objects; i++) {
670 struct object_entry *entry = *list++;
671 uint32_t crc32_val = htonl(entry->crc32);
672 sha1write(f, &crc32_val, 4);
675 /* write the 32-bit offset table */
676 list = sorted_by_sha;
677 for (i = 0; i < nr_objects; i++) {
678 struct object_entry *entry = *list++;
679 uint32_t offset = (entry->offset <= index_off32_limit) ?
680 entry->offset : (0x80000000 | nr_large_offset++);
681 offset = htonl(offset);
682 sha1write(f, &offset, 4);
685 /* write the large offset table */
686 list = sorted_by_sha;
687 while (nr_large_offset) {
688 struct object_entry *entry = *list++;
689 uint64_t offset = entry->offset;
690 if (offset > index_off32_limit) {
692 split[0] = htonl(offset >> 32);
693 split[1] = htonl(offset & 0xffffffff);
694 sha1write(f, split, 8);
700 sha1write(f, pack_file_sha1, 20);
701 sha1close(f, NULL, 1);
703 SHA1_Final(sha1, &ctx);
706 static int locate_object_entry_hash(const unsigned char *sha1)
710 memcpy(&ui, sha1, sizeof(unsigned int));
711 i = ui % object_ix_hashsz;
712 while (0 < object_ix[i]) {
713 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
715 if (++i == object_ix_hashsz)
721 static struct object_entry *locate_object_entry(const unsigned char *sha1)
725 if (!object_ix_hashsz)
728 i = locate_object_entry_hash(sha1);
730 return &objects[object_ix[i]-1];
734 static void rehash_objects(void)
737 struct object_entry *oe;
739 object_ix_hashsz = nr_objects * 3;
740 if (object_ix_hashsz < 1024)
741 object_ix_hashsz = 1024;
742 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
743 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
744 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
745 int ix = locate_object_entry_hash(oe->sha1);
749 object_ix[ix] = i + 1;
753 static unsigned name_hash(const char *name)
759 * This effectively just creates a sortable number from the
760 * last sixteen non-whitespace characters. Last characters
761 * count "most", so things that end in ".c" sort together.
763 while ((c = *name++) != 0) {
766 hash = (hash >> 2) + (c << 24);
771 static int add_object_entry(const unsigned char *sha1, enum object_type type,
772 unsigned hash, int exclude)
774 struct object_entry *entry;
775 struct packed_git *p, *found_pack = NULL;
776 off_t found_offset = 0;
779 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
782 entry = objects + object_ix[ix] - 1;
783 if (!entry->preferred_base)
785 entry->preferred_base = 1;
790 for (p = packed_git; p; p = p->next) {
791 off_t offset = find_pack_entry_one(sha1, p);
794 found_offset = offset;
801 if (local && !p->pack_local)
806 if (nr_objects >= nr_alloc) {
807 nr_alloc = (nr_alloc + 1024) * 3 / 2;
808 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
811 entry = objects + nr_objects++;
812 memset(entry, 0, sizeof(*entry));
813 hashcpy(entry->sha1, sha1);
818 entry->preferred_base = 1;
822 entry->in_pack = found_pack;
823 entry->in_pack_offset = found_offset;
826 if (object_ix_hashsz * 3 <= nr_objects * 4)
829 object_ix[-1 - ix] = nr_objects;
832 display_progress(&progress_state, nr_objects);
837 struct pbase_tree_cache {
838 unsigned char sha1[20];
842 unsigned long tree_size;
845 static struct pbase_tree_cache *(pbase_tree_cache[256]);
846 static int pbase_tree_cache_ix(const unsigned char *sha1)
848 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
850 static int pbase_tree_cache_ix_incr(int ix)
852 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
855 static struct pbase_tree {
856 struct pbase_tree *next;
857 /* This is a phony "cache" entry; we are not
858 * going to evict it nor find it through _get()
859 * mechanism -- this is for the toplevel node that
860 * would almost always change with any commit.
862 struct pbase_tree_cache pcache;
865 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
867 struct pbase_tree_cache *ent, *nent;
870 enum object_type type;
872 int my_ix = pbase_tree_cache_ix(sha1);
873 int available_ix = -1;
875 /* pbase-tree-cache acts as a limited hashtable.
876 * your object will be found at your index or within a few
877 * slots after that slot if it is cached.
879 for (neigh = 0; neigh < 8; neigh++) {
880 ent = pbase_tree_cache[my_ix];
881 if (ent && !hashcmp(ent->sha1, sha1)) {
885 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
886 ((0 <= available_ix) &&
887 (!ent && pbase_tree_cache[available_ix])))
888 available_ix = my_ix;
891 my_ix = pbase_tree_cache_ix_incr(my_ix);
894 /* Did not find one. Either we got a bogus request or
895 * we need to read and perhaps cache.
897 data = read_sha1_file(sha1, &type, &size);
900 if (type != OBJ_TREE) {
905 /* We need to either cache or return a throwaway copy */
907 if (available_ix < 0)
910 ent = pbase_tree_cache[available_ix];
911 my_ix = available_ix;
915 nent = xmalloc(sizeof(*nent));
916 nent->temporary = (available_ix < 0);
919 /* evict and reuse */
920 free(ent->tree_data);
923 hashcpy(nent->sha1, sha1);
924 nent->tree_data = data;
925 nent->tree_size = size;
927 if (!nent->temporary)
928 pbase_tree_cache[my_ix] = nent;
932 static void pbase_tree_put(struct pbase_tree_cache *cache)
934 if (!cache->temporary) {
938 free(cache->tree_data);
942 static int name_cmp_len(const char *name)
945 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
950 static void add_pbase_object(struct tree_desc *tree,
953 const char *fullname)
955 struct name_entry entry;
958 while (tree_entry(tree,&entry)) {
959 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
960 memcmp(name, entry.path, cmplen);
965 if (name[cmplen] != '/') {
966 unsigned hash = name_hash(fullname);
967 add_object_entry(entry.sha1,
968 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
972 if (S_ISDIR(entry.mode)) {
973 struct tree_desc sub;
974 struct pbase_tree_cache *tree;
975 const char *down = name+cmplen+1;
976 int downlen = name_cmp_len(down);
978 tree = pbase_tree_get(entry.sha1);
981 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
983 add_pbase_object(&sub, down, downlen, fullname);
984 pbase_tree_put(tree);
989 static unsigned *done_pbase_paths;
990 static int done_pbase_paths_num;
991 static int done_pbase_paths_alloc;
992 static int done_pbase_path_pos(unsigned hash)
995 int hi = done_pbase_paths_num;
997 int mi = (hi + lo) / 2;
998 if (done_pbase_paths[mi] == hash)
1000 if (done_pbase_paths[mi] < hash)
1008 static int check_pbase_path(unsigned hash)
1010 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1014 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1015 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1016 done_pbase_paths = xrealloc(done_pbase_paths,
1017 done_pbase_paths_alloc *
1020 done_pbase_paths_num++;
1021 if (pos < done_pbase_paths_num)
1022 memmove(done_pbase_paths + pos + 1,
1023 done_pbase_paths + pos,
1024 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1025 done_pbase_paths[pos] = hash;
1029 static void add_preferred_base_object(const char *name, unsigned hash)
1031 struct pbase_tree *it;
1034 if (!num_preferred_base || check_pbase_path(hash))
1037 cmplen = name_cmp_len(name);
1038 for (it = pbase_tree; it; it = it->next) {
1040 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
1043 struct tree_desc tree;
1044 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1045 add_pbase_object(&tree, name, cmplen, name);
1050 static void add_preferred_base(unsigned char *sha1)
1052 struct pbase_tree *it;
1055 unsigned char tree_sha1[20];
1057 if (window <= num_preferred_base++)
1060 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1064 for (it = pbase_tree; it; it = it->next) {
1065 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1071 it = xcalloc(1, sizeof(*it));
1072 it->next = pbase_tree;
1075 hashcpy(it->pcache.sha1, tree_sha1);
1076 it->pcache.tree_data = data;
1077 it->pcache.tree_size = size;
1080 static void check_object(struct object_entry *entry)
1082 if (entry->in_pack) {
1083 struct packed_git *p = entry->in_pack;
1084 struct pack_window *w_curs = NULL;
1085 const unsigned char *base_ref = NULL;
1086 struct object_entry *base_entry;
1087 unsigned long used, used_0;
1090 unsigned char *buf, c;
1092 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1095 * We want in_pack_type even if we do not reuse delta
1096 * since non-delta representations could still be reused.
1098 used = unpack_object_header_gently(buf, avail,
1099 &entry->in_pack_type,
1103 * Determine if this is a delta and if so whether we can
1104 * reuse it or not. Otherwise let's find out as cheaply as
1105 * possible what the actual type and size for this object is.
1107 switch (entry->in_pack_type) {
1109 /* Not a delta hence we've already got all we need. */
1110 entry->type = entry->in_pack_type;
1111 entry->in_pack_header_size = used;
1112 unuse_pack(&w_curs);
1115 if (!no_reuse_delta && !entry->preferred_base)
1116 base_ref = use_pack(p, &w_curs,
1117 entry->in_pack_offset + used, NULL);
1118 entry->in_pack_header_size = used + 20;
1121 buf = use_pack(p, &w_curs,
1122 entry->in_pack_offset + used, NULL);
1128 if (!ofs || MSB(ofs, 7))
1129 die("delta base offset overflow in pack for %s",
1130 sha1_to_hex(entry->sha1));
1132 ofs = (ofs << 7) + (c & 127);
1134 if (ofs >= entry->in_pack_offset)
1135 die("delta base offset out of bound for %s",
1136 sha1_to_hex(entry->sha1));
1137 ofs = entry->in_pack_offset - ofs;
1138 if (!no_reuse_delta && !entry->preferred_base)
1139 base_ref = find_packed_object_name(p, ofs);
1140 entry->in_pack_header_size = used + used_0;
1144 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1146 * If base_ref was set above that means we wish to
1147 * reuse delta data, and we even found that base
1148 * in the list of objects we want to pack. Goodie!
1150 * Depth value does not matter - find_deltas() will
1151 * never consider reused delta as the base object to
1152 * deltify other objects against, in order to avoid
1155 entry->type = entry->in_pack_type;
1156 entry->delta = base_entry;
1157 entry->delta_sibling = base_entry->delta_child;
1158 base_entry->delta_child = entry;
1159 unuse_pack(&w_curs);
1165 * This must be a delta and we already know what the
1166 * final object type is. Let's extract the actual
1167 * object size from the delta header.
1169 entry->size = get_size_from_delta(p, &w_curs,
1170 entry->in_pack_offset + entry->in_pack_header_size);
1171 unuse_pack(&w_curs);
1176 * No choice but to fall back to the recursive delta walk
1177 * with sha1_object_info() to find about the object type
1180 unuse_pack(&w_curs);
1183 entry->type = sha1_object_info(entry->sha1, &entry->size);
1184 if (entry->type < 0)
1185 die("unable to get type of object %s",
1186 sha1_to_hex(entry->sha1));
1189 static int pack_offset_sort(const void *_a, const void *_b)
1191 const struct object_entry *a = *(struct object_entry **)_a;
1192 const struct object_entry *b = *(struct object_entry **)_b;
1194 /* avoid filesystem trashing with loose objects */
1195 if (!a->in_pack && !b->in_pack)
1196 return hashcmp(a->sha1, b->sha1);
1198 if (a->in_pack < b->in_pack)
1200 if (a->in_pack > b->in_pack)
1202 return a->in_pack_offset < b->in_pack_offset ? -1 :
1203 (a->in_pack_offset > b->in_pack_offset);
1206 static void get_object_details(void)
1209 struct object_entry **sorted_by_offset;
1211 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1212 for (i = 0; i < nr_objects; i++)
1213 sorted_by_offset[i] = objects + i;
1214 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1217 for (i = 0; i < nr_objects; i++)
1218 check_object(sorted_by_offset[i]);
1219 free(sorted_by_offset);
1222 static int type_size_sort(const void *_a, const void *_b)
1224 const struct object_entry *a = *(struct object_entry **)_a;
1225 const struct object_entry *b = *(struct object_entry **)_b;
1227 if (a->type < b->type)
1229 if (a->type > b->type)
1231 if (a->hash < b->hash)
1233 if (a->hash > b->hash)
1235 if (a->preferred_base < b->preferred_base)
1237 if (a->preferred_base > b->preferred_base)
1239 if (a->size < b->size)
1241 if (a->size > b->size)
1243 return a > b ? -1 : (a < b); /* newest last */
1247 struct object_entry *entry;
1249 struct delta_index *index;
1253 * We search for deltas _backwards_ in a list sorted by type and
1254 * by size, so that we see progressively smaller and smaller files.
1255 * That's because we prefer deltas to be from the bigger file
1256 * to the smaller - deletes are potentially cheaper, but perhaps
1257 * more importantly, the bigger file is likely the more recent
1260 static int try_delta(struct unpacked *trg, struct unpacked *src,
1263 struct object_entry *trg_entry = trg->entry;
1264 struct object_entry *src_entry = src->entry;
1265 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1266 enum object_type type;
1269 /* Don't bother doing diffs between different types */
1270 if (trg_entry->type != src_entry->type)
1273 /* We do not compute delta to *create* objects we are not
1276 if (trg_entry->preferred_base)
1280 * We do not bother to try a delta that we discarded
1281 * on an earlier try, but only when reusing delta data.
1283 if (!no_reuse_delta && trg_entry->in_pack &&
1284 trg_entry->in_pack == src_entry->in_pack &&
1285 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1286 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1289 /* Let's not bust the allowed depth. */
1290 if (src_entry->depth >= max_depth)
1293 /* Now some size filtering heuristics. */
1294 trg_size = trg_entry->size;
1295 max_size = trg_size/2 - 20;
1296 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1299 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1300 max_size = trg_entry->delta_size-1;
1301 src_size = src_entry->size;
1302 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1303 if (sizediff >= max_size)
1306 /* Load data if not already done */
1308 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1310 die("object %s inconsistent object length (%lu vs %lu)",
1311 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1314 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1316 die("object %s inconsistent object length (%lu vs %lu)",
1317 sha1_to_hex(src_entry->sha1), sz, src_size);
1320 src->index = create_delta_index(src->data, src_size);
1322 die("out of memory");
1325 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1329 trg_entry->delta = src_entry;
1330 trg_entry->delta_size = delta_size;
1331 trg_entry->depth = src_entry->depth + 1;
1336 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1338 struct object_entry *child = me->delta_child;
1341 unsigned int c = check_delta_limit(child, n + 1);
1344 child = child->delta_sibling;
1349 static void find_deltas(struct object_entry **list, int window, int depth)
1351 uint32_t i = nr_objects, idx = 0, processed = 0;
1352 unsigned int array_size = window * sizeof(struct unpacked);
1353 struct unpacked *array;
1358 array = xmalloc(array_size);
1359 memset(array, 0, array_size);
1361 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1364 struct object_entry *entry = list[--i];
1365 struct unpacked *n = array + idx;
1368 if (!entry->preferred_base)
1372 display_progress(&progress_state, processed);
1375 /* This happens if we decided to reuse existing
1376 * delta from a pack. "!no_reuse_delta &&" is implied.
1380 if (entry->size < 50)
1382 free_delta_index(n->index);
1389 * If the current object is at pack edge, take the depth the
1390 * objects that depend on the current object into account
1391 * otherwise they would become too deep.
1394 if (entry->delta_child) {
1395 max_depth -= check_delta_limit(entry, 0);
1402 uint32_t other_idx = idx + j;
1404 if (other_idx >= window)
1405 other_idx -= window;
1406 m = array + other_idx;
1409 if (try_delta(n, m, max_depth) < 0)
1413 /* if we made n a delta, and if n is already at max
1414 * depth, leaving it in the window is pointless. we
1415 * should evict it first.
1417 if (entry->delta && depth <= entry->depth)
1427 stop_progress(&progress_state);
1429 for (i = 0; i < window; ++i) {
1430 free_delta_index(array[i].index);
1431 free(array[i].data);
1436 static void prepare_pack(int window, int depth)
1438 struct object_entry **delta_list;
1441 get_object_details();
1443 if (!window || !depth)
1446 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1447 for (i = 0; i < nr_objects; i++)
1448 delta_list[i] = objects + i;
1449 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1450 find_deltas(delta_list, window+1, depth);
1454 static int git_pack_config(const char *k, const char *v)
1456 if(!strcmp(k, "pack.window")) {
1457 window = git_config_int(k, v);
1460 if(!strcmp(k, "pack.depth")) {
1461 depth = git_config_int(k, v);
1464 if (!strcmp(k, "pack.compression")) {
1465 int level = git_config_int(k, v);
1467 level = Z_DEFAULT_COMPRESSION;
1468 else if (level < 0 || level > Z_BEST_COMPRESSION)
1469 die("bad pack compression level %d", level);
1470 pack_compression_level = level;
1471 pack_compression_seen = 1;
1474 return git_default_config(k, v);
1477 static void read_object_list_from_stdin(void)
1479 char line[40 + 1 + PATH_MAX + 2];
1480 unsigned char sha1[20];
1484 if (!fgets(line, sizeof(line), stdin)) {
1488 die("fgets returned NULL, not EOF, not error!");
1490 die("fgets: %s", strerror(errno));
1494 if (line[0] == '-') {
1495 if (get_sha1_hex(line+1, sha1))
1496 die("expected edge sha1, got garbage:\n %s",
1498 add_preferred_base(sha1);
1501 if (get_sha1_hex(line, sha1))
1502 die("expected sha1, got garbage:\n %s", line);
1504 hash = name_hash(line+41);
1505 add_preferred_base_object(line+41, hash);
1506 add_object_entry(sha1, 0, hash, 0);
1510 static void show_commit(struct commit *commit)
1512 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
1515 static void show_object(struct object_array_entry *p)
1517 unsigned hash = name_hash(p->name);
1518 add_preferred_base_object(p->name, hash);
1519 add_object_entry(p->item->sha1, p->item->type, hash, 0);
1522 static void show_edge(struct commit *commit)
1524 add_preferred_base(commit->object.sha1);
1527 static void get_object_list(int ac, const char **av)
1529 struct rev_info revs;
1533 init_revisions(&revs, NULL);
1534 save_commit_buffer = 0;
1535 track_object_refs = 0;
1536 setup_revisions(ac, av, &revs, NULL);
1538 while (fgets(line, sizeof(line), stdin) != NULL) {
1539 int len = strlen(line);
1540 if (line[len - 1] == '\n')
1545 if (!strcmp(line, "--not")) {
1546 flags ^= UNINTERESTING;
1549 die("not a rev '%s'", line);
1551 if (handle_revision_arg(line, &revs, flags, 1))
1552 die("bad revision '%s'", line);
1555 prepare_revision_walk(&revs);
1556 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1557 traverse_commit_list(&revs, show_commit, show_object);
1560 static int adjust_perm(const char *path, mode_t mode)
1562 if (chmod(path, mode))
1564 return adjust_shared_perm(path);
1567 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1569 int use_internal_rev_list = 0;
1573 int rp_ac_alloc = 64;
1576 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1578 rp_av[0] = "pack-objects";
1579 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1582 git_config(git_pack_config);
1583 if (!pack_compression_seen && core_compression_seen)
1584 pack_compression_level = core_compression_level;
1586 progress = isatty(2);
1587 for (i = 1; i < argc; i++) {
1588 const char *arg = argv[i];
1593 if (!strcmp("--non-empty", arg)) {
1597 if (!strcmp("--local", arg)) {
1601 if (!strcmp("--incremental", arg)) {
1605 if (!prefixcmp(arg, "--compression=")) {
1607 int level = strtoul(arg+14, &end, 0);
1608 if (!arg[14] || *end)
1611 level = Z_DEFAULT_COMPRESSION;
1612 else if (level < 0 || level > Z_BEST_COMPRESSION)
1613 die("bad pack compression level %d", level);
1614 pack_compression_level = level;
1617 if (!prefixcmp(arg, "--window=")) {
1619 window = strtoul(arg+9, &end, 0);
1620 if (!arg[9] || *end)
1624 if (!prefixcmp(arg, "--depth=")) {
1626 depth = strtoul(arg+8, &end, 0);
1627 if (!arg[8] || *end)
1631 if (!strcmp("--progress", arg)) {
1635 if (!strcmp("--all-progress", arg)) {
1639 if (!strcmp("-q", arg)) {
1643 if (!strcmp("--no-reuse-delta", arg)) {
1647 if (!strcmp("--no-reuse-object", arg)) {
1648 no_reuse_object = no_reuse_delta = 1;
1651 if (!strcmp("--delta-base-offset", arg)) {
1652 allow_ofs_delta = 1;
1655 if (!strcmp("--stdout", arg)) {
1659 if (!strcmp("--revs", arg)) {
1660 use_internal_rev_list = 1;
1663 if (!strcmp("--unpacked", arg) ||
1664 !prefixcmp(arg, "--unpacked=") ||
1665 !strcmp("--reflog", arg) ||
1666 !strcmp("--all", arg)) {
1667 use_internal_rev_list = 1;
1668 if (rp_ac >= rp_ac_alloc - 1) {
1669 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1670 rp_av = xrealloc(rp_av,
1671 rp_ac_alloc * sizeof(*rp_av));
1673 rp_av[rp_ac++] = arg;
1676 if (!strcmp("--thin", arg)) {
1677 use_internal_rev_list = 1;
1679 rp_av[1] = "--objects-edge";
1682 if (!prefixcmp(arg, "--index-version=")) {
1684 index_default_version = strtoul(arg + 16, &c, 10);
1685 if (index_default_version > 2)
1688 index_off32_limit = strtoul(c+1, &c, 0);
1689 if (*c || index_off32_limit & 0x80000000)
1696 /* Traditionally "pack-objects [options] base extra" failed;
1697 * we would however want to take refs parameter that would
1698 * have been given to upstream rev-list ourselves, which means
1699 * we somehow want to say what the base name is. So the
1702 * pack-objects [options] base <refs...>
1704 * in other words, we would treat the first non-option as the
1705 * base_name and send everything else to the internal revision
1709 if (!pack_to_stdout)
1710 base_name = argv[i++];
1712 if (pack_to_stdout != !base_name)
1715 if (!pack_to_stdout && thin)
1716 die("--thin cannot be used to build an indexable pack.");
1718 prepare_packed_git();
1721 start_progress(&progress_state, "Generating pack...",
1722 "Counting objects: ", 0);
1723 if (!use_internal_rev_list)
1724 read_object_list_from_stdin();
1726 rp_av[rp_ac] = NULL;
1727 get_object_list(rp_ac, rp_av);
1730 stop_progress(&progress_state);
1731 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1734 if (non_empty && !nr_result)
1736 if (progress && (nr_objects != nr_result))
1737 fprintf(stderr, "Result has %u objects.\n", nr_result);
1739 prepare_pack(window, depth);
1742 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1743 written, written_delta, reused, reused_delta);