7 static const char pack_usage[] = "git-pack-objects [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
10 unsigned char sha1[20];
15 enum object_type type;
16 unsigned long delta_size;
17 struct object_entry *delta;
20 static unsigned char object_list_sha1[20];
21 static int non_empty = 0;
23 static int incremental = 0;
24 static struct object_entry **sorted_by_sha, **sorted_by_type;
25 static struct object_entry *objects = NULL;
26 static int nr_objects = 0, nr_alloc = 0;
27 static const char *base_name;
28 static unsigned char pack_file_sha1[20];
30 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
32 unsigned long othersize, delta_size;
34 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
38 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
39 delta_buf = diff_delta(otherbuf, othersize,
40 buf, size, &delta_size, 0);
41 if (!delta_buf || delta_size != entry->delta_size)
42 die("delta size changed");
49 * The per-object header is a pretty dense thing, which is
50 * - first byte: low four bits are "size", then three bits of "type",
51 * and the high bit is "size continues".
52 * - each byte afterwards: low seven bits are size continuation,
53 * with the high bit being "size continues"
55 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
60 if (type < OBJ_COMMIT || type > OBJ_DELTA)
61 die("bad type %d", type);
63 c = (type << 4) | (size & 15);
75 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
79 void *buf = read_sha1_file(entry->sha1, type, &size);
80 unsigned char header[10];
81 unsigned hdrlen, datalen;
82 enum object_type obj_type;
85 die("unable to read %s", sha1_to_hex(entry->sha1));
86 if (size != entry->size)
87 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
90 * The object header is a byte of 'type' followed by zero or
91 * more bytes of length. For deltas, the 20 bytes of delta sha1
94 obj_type = entry->type;
96 buf = delta_against(buf, size, entry);
97 size = entry->delta_size;
100 hdrlen = encode_header(obj_type, size, header);
101 sha1write(f, header, hdrlen);
103 sha1write(f, entry->delta, 20);
106 datalen = sha1write_compressed(f, buf, size);
108 return hdrlen + datalen;
111 static unsigned long write_one(struct sha1file *f,
112 struct object_entry *e,
113 unsigned long offset)
116 /* offset starts from header size and cannot be zero
117 * if it is written already.
121 offset += write_object(f, e);
122 /* if we are delitified, write out its base object. */
124 offset = write_one(f, e->delta, offset);
128 static void write_pack_file(void)
132 unsigned long offset;
134 struct pack_header hdr;
137 f = sha1fd(1, "<stdout>");
139 f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "pack");
140 hdr.hdr_signature = htonl(PACK_SIGNATURE);
141 hdr.hdr_version = htonl(PACK_VERSION);
142 hdr.hdr_entries = htonl(nr_objects);
143 sha1write(f, &hdr, sizeof(hdr));
144 offset = sizeof(hdr);
145 for (i = 0; i < nr_objects; i++)
146 offset = write_one(f, objects + i, offset);
148 sha1close(f, pack_file_sha1, 1);
153 static void write_index_file(void)
156 struct sha1file *f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "idx");
157 struct object_entry **list = sorted_by_sha;
158 struct object_entry **last = list + nr_objects;
159 unsigned int array[256];
162 * Write the first-level table (the list is sorted,
163 * but we use a 256-entry lookup to be able to avoid
164 * having to do eight extra binary search iterations).
166 for (i = 0; i < 256; i++) {
167 struct object_entry **next = list;
168 while (next < last) {
169 struct object_entry *entry = *next;
170 if (entry->sha1[0] != i)
174 array[i] = htonl(next - sorted_by_sha);
177 sha1write(f, array, 256 * sizeof(int));
180 * Write the actual SHA1 entries..
182 list = sorted_by_sha;
183 for (i = 0; i < nr_objects; i++) {
184 struct object_entry *entry = *list++;
185 unsigned int offset = htonl(entry->offset);
186 sha1write(f, &offset, 4);
187 sha1write(f, entry->sha1, 20);
189 sha1write(f, pack_file_sha1, 20);
190 sha1close(f, NULL, 1);
193 static int add_object_entry(unsigned char *sha1, unsigned int hash)
195 unsigned int idx = nr_objects;
196 struct object_entry *entry;
198 if (incremental || local) {
199 struct packed_git *p;
201 for (p = packed_git; p; p = p->next) {
204 if (find_pack_entry_one(sha1, &e, p)) {
207 if (local && !p->pack_local)
213 if (idx >= nr_alloc) {
214 unsigned int needed = (idx + 1024) * 3 / 2;
215 objects = xrealloc(objects, needed * sizeof(*entry));
218 entry = objects + idx;
219 memset(entry, 0, sizeof(*entry));
220 memcpy(entry->sha1, sha1, 20);
226 static void check_object(struct object_entry *entry)
230 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
231 if (!strcmp(type, "commit")) {
232 entry->type = OBJ_COMMIT;
233 } else if (!strcmp(type, "tree")) {
234 entry->type = OBJ_TREE;
235 } else if (!strcmp(type, "blob")) {
236 entry->type = OBJ_BLOB;
237 } else if (!strcmp(type, "tag")) {
238 entry->type = OBJ_TAG;
240 die("unable to pack object %s of type %s",
241 sha1_to_hex(entry->sha1), type);
244 die("unable to get type of object %s",
245 sha1_to_hex(entry->sha1));
248 static void get_object_details(void)
251 struct object_entry *entry = objects;
253 for (i = 0; i < nr_objects; i++)
254 check_object(entry++);
257 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
259 static entry_sort_t current_sort;
261 static int sort_comparator(const void *_a, const void *_b)
263 struct object_entry *a = *(struct object_entry **)_a;
264 struct object_entry *b = *(struct object_entry **)_b;
265 return current_sort(a,b);
268 static struct object_entry **create_sorted_list(entry_sort_t sort)
270 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
273 for (i = 0; i < nr_objects; i++)
274 list[i] = objects + i;
276 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
280 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
282 return memcmp(a->sha1, b->sha1, 20);
285 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
287 if (a->type < b->type)
289 if (a->type > b->type)
291 if (a->hash < b->hash)
293 if (a->hash > b->hash)
295 if (a->size < b->size)
297 if (a->size > b->size)
299 return a < b ? -1 : (a > b);
303 struct object_entry *entry;
308 * We search for deltas _backwards_ in a list sorted by type and
309 * by size, so that we see progressively smaller and smaller files.
310 * That's because we prefer deltas to be from the bigger file
311 * to the smaller - deletes are potentially cheaper, but perhaps
312 * more importantly, the bigger file is likely the more recent
315 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
317 struct object_entry *cur_entry = cur->entry;
318 struct object_entry *old_entry = old->entry;
319 unsigned long size, oldsize, delta_size, sizediff;
323 /* Don't bother doing diffs between different types */
324 if (cur_entry->type != old_entry->type)
327 size = cur_entry->size;
330 oldsize = old_entry->size;
331 sizediff = oldsize > size ? oldsize - size : size - oldsize;
332 if (sizediff > size / 8)
334 if (old_entry->depth >= max_depth)
340 * We always delta from the bigger to the smaller, since that's
341 * more space-efficient (deletes don't have to say _what_ they
344 max_size = size / 2 - 20;
345 if (cur_entry->delta)
346 max_size = cur_entry->delta_size-1;
347 if (sizediff >= max_size)
349 delta_buf = diff_delta(old->data, oldsize,
350 cur->data, size, &delta_size, max_size);
353 cur_entry->delta = old_entry;
354 cur_entry->delta_size = delta_size;
355 cur_entry->depth = old_entry->depth + 1;
360 static void find_deltas(struct object_entry **list, int window, int depth)
363 unsigned int array_size = window * sizeof(struct unpacked);
364 struct unpacked *array = xmalloc(array_size);
366 memset(array, 0, array_size);
370 struct object_entry *entry = list[i];
371 struct unpacked *n = array + idx;
378 n->data = read_sha1_file(entry->sha1, type, &size);
379 if (size != entry->size)
380 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
383 unsigned int other_idx = idx + j;
385 if (other_idx >= window)
387 m = array + other_idx;
390 if (try_delta(n, m, depth) < 0)
398 for (i = 0; i < window; ++i)
403 static void prepare_pack(int window, int depth)
405 get_object_details();
407 fprintf(stderr, "Packing %d objects\n", nr_objects);
409 sorted_by_type = create_sorted_list(type_size_sort);
411 find_deltas(sorted_by_type, window+1, depth);
415 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
417 static const char cache[] = "pack-cache/pack-%s.%s";
418 char *cached_pack, *cached_idx;
419 int ifd, ofd, ifd_ix = -1;
421 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
422 ifd = open(cached_pack, O_RDONLY);
426 if (!pack_to_stdout) {
427 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
428 ifd_ix = open(cached_idx, O_RDONLY);
435 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
438 if (pack_to_stdout) {
445 snprintf(name, sizeof(name),
446 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
447 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
449 die("unable to open %s (%s)", name, strerror(errno));
450 if (copy_fd(ifd, ofd))
454 snprintf(name, sizeof(name),
455 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
456 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
458 die("unable to open %s (%s)", name, strerror(errno));
459 if (copy_fd(ifd_ix, ofd))
462 puts(sha1_to_hex(sha1));
468 int main(int argc, char **argv)
471 char line[PATH_MAX + 20];
472 int window = 10, depth = 10, pack_to_stdout = 0;
473 struct object_entry **list;
476 setup_git_directory();
478 for (i = 1; i < argc; i++) {
479 const char *arg = argv[i];
482 if (!strcmp("--non-empty", arg)) {
486 if (!strcmp("--local", arg)) {
490 if (!strcmp("--incremental", arg)) {
494 if (!strncmp("--window=", arg, 9)) {
496 window = strtoul(arg+9, &end, 0);
501 if (!strncmp("--depth=", arg, 8)) {
503 depth = strtoul(arg+8, &end, 0);
508 if (!strcmp("--stdout", arg)) {
519 if (pack_to_stdout != !base_name)
522 prepare_packed_git();
523 while (fgets(line, sizeof(line), stdin) != NULL) {
526 unsigned char sha1[20];
528 if (get_sha1_hex(line, sha1))
529 die("expected sha1, got garbage:\n %s", line);
533 unsigned char c = *p++;
536 hash = hash * 11 + c;
538 add_object_entry(sha1, hash);
540 if (non_empty && !nr_objects)
543 sorted_by_sha = create_sorted_list(sha1_sort);
545 list = sorted_by_sha;
546 for (i = 0; i < nr_objects; i++) {
547 struct object_entry *entry = *list++;
548 SHA1_Update(&ctx, entry->sha1, 20);
550 SHA1_Final(object_list_sha1, &ctx);
552 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
555 prepare_pack(window, depth);
556 if (!pack_to_stdout) {
558 puts(sha1_to_hex(object_list_sha1));