8 static const char pack_usage[] = "git-pack-objects [--window=N] [--depth=N] {--stdout | base-name} < object-list";
11 unsigned char sha1[20];
16 enum object_type type;
17 unsigned long delta_size;
18 struct object_entry *delta;
21 static struct object_entry **sorted_by_sha, **sorted_by_type;
22 static struct object_entry *objects = NULL;
23 static int nr_objects = 0, nr_alloc = 0;
24 static const char *base_name;
25 static unsigned char pack_file_sha1[20];
27 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
29 unsigned long othersize, delta_size;
31 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
35 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
36 delta_buf = diff_delta(otherbuf, othersize,
37 buf, size, &delta_size, ~0UL);
38 if (!delta_buf || delta_size != entry->delta_size)
39 die("delta size changed");
46 * The per-object header is a pretty dense thing, which is
47 * - first byte: low four bits are "size", then three bits of "type",
48 * and the high bit is "size continues".
49 * - each byte afterwards: low seven bits are size continuation,
50 * with the high bit being "size continues"
52 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
57 if (type < OBJ_COMMIT || type > OBJ_DELTA)
58 die("bad type %d", type);
61 * Shift the size up by 7 bits at a time,
62 * until you get bits in the "high four".
63 * That will be our beginning. We'll have
64 * four size bits in 28..31, then groups
65 * of seven in 21..27, 14..20, 7..13 and
70 while (!(size & 0xfe000000)) {
75 c = (type << 4) | (size >> 28);
76 for (i = 1; i < n; i++) {
78 c = (size >> 21) & 0x7f;
85 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
89 void *buf = read_sha1_file(entry->sha1, type, &size);
90 unsigned char header[10];
91 unsigned hdrlen, datalen;
92 enum object_type obj_type;
95 die("unable to read %s", sha1_to_hex(entry->sha1));
96 if (size != entry->size)
97 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
100 * The object header is a byte of 'type' followed by zero or
101 * more bytes of length. For deltas, the 20 bytes of delta sha1
104 obj_type = entry->type;
106 buf = delta_against(buf, size, entry);
107 size = entry->delta_size;
108 obj_type = OBJ_DELTA;
110 hdrlen = encode_header(obj_type, size, header);
111 sha1write(f, header, hdrlen);
113 sha1write(f, entry->delta, 20);
116 datalen = sha1write_compressed(f, buf, size);
118 return hdrlen + datalen;
121 static void write_pack_file(void)
125 unsigned long offset;
127 struct pack_header hdr;
130 f = sha1fd(1, "<stdout>");
132 f = sha1create("%s.%s", base_name, "pack");
133 hdr.hdr_signature = htonl(PACK_SIGNATURE);
134 hdr.hdr_version = htonl(1);
135 hdr.hdr_entries = htonl(nr_objects);
136 sha1write(f, &hdr, sizeof(hdr));
137 offset = sizeof(hdr);
138 for (i = 0; i < nr_objects; i++) {
139 struct object_entry *entry = objects + i;
140 entry->offset = offset;
141 offset += write_object(f, entry);
143 sha1close(f, pack_file_sha1, 1);
148 static void write_index_file(void)
151 struct sha1file *f = sha1create("%s.%s", base_name, "idx");
152 struct object_entry **list = sorted_by_sha;
153 struct object_entry **last = list + nr_objects;
154 unsigned int array[256];
157 * Write the first-level table (the list is sorted,
158 * but we use a 256-entry lookup to be able to avoid
159 * having to do eight extra binary search iterations).
161 for (i = 0; i < 256; i++) {
162 struct object_entry **next = list;
163 while (next < last) {
164 struct object_entry *entry = *next;
165 if (entry->sha1[0] != i)
169 array[i] = htonl(next - sorted_by_sha);
172 sha1write(f, array, 256 * sizeof(int));
175 * Write the actual SHA1 entries..
177 list = sorted_by_sha;
178 for (i = 0; i < nr_objects; i++) {
179 struct object_entry *entry = *list++;
180 unsigned int offset = htonl(entry->offset);
181 sha1write(f, &offset, 4);
182 sha1write(f, entry->sha1, 20);
184 sha1write(f, pack_file_sha1, 20);
185 sha1close(f, NULL, 1);
188 static void add_object_entry(unsigned char *sha1, unsigned int hash)
190 unsigned int idx = nr_objects;
191 struct object_entry *entry;
193 if (idx >= nr_alloc) {
194 unsigned int needed = (idx + 1024) * 3 / 2;
195 objects = xrealloc(objects, needed * sizeof(*entry));
198 entry = objects + idx;
199 memset(entry, 0, sizeof(*entry));
200 memcpy(entry->sha1, sha1, 20);
205 static void check_object(struct object_entry *entry)
209 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
210 if (!strcmp(type, "commit")) {
211 entry->type = OBJ_COMMIT;
212 } else if (!strcmp(type, "tree")) {
213 entry->type = OBJ_TREE;
214 } else if (!strcmp(type, "blob")) {
215 entry->type = OBJ_BLOB;
216 } else if (!strcmp(type, "tag")) {
217 entry->type = OBJ_TAG;
219 die("unable to pack object %s of type %s",
220 sha1_to_hex(entry->sha1), type);
223 die("unable to get type of object %s",
224 sha1_to_hex(entry->sha1));
227 static void get_object_details(void)
230 struct object_entry *entry = objects;
232 for (i = 0; i < nr_objects; i++)
233 check_object(entry++);
236 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
238 static entry_sort_t current_sort;
240 static int sort_comparator(const void *_a, const void *_b)
242 struct object_entry *a = *(struct object_entry **)_a;
243 struct object_entry *b = *(struct object_entry **)_b;
244 return current_sort(a,b);
247 static struct object_entry **create_sorted_list(entry_sort_t sort)
249 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
252 for (i = 0; i < nr_objects; i++)
253 list[i] = objects + i;
255 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
259 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
261 return memcmp(a->sha1, b->sha1, 20);
264 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
266 if (a->type < b->type)
268 if (a->type > b->type)
270 if (a->hash < b->hash)
272 if (a->hash > b->hash)
274 if (a->size < b->size)
276 if (a->size > b->size)
278 return a < b ? -1 : (a > b);
282 struct object_entry *entry;
287 * We search for deltas _backwards_ in a list sorted by type and
288 * by size, so that we see progressively smaller and smaller files.
289 * That's because we prefer deltas to be from the bigger file
290 * to the smaller - deletes are potentially cheaper, but perhaps
291 * more importantly, the bigger file is likely the more recent
294 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
296 struct object_entry *cur_entry = cur->entry;
297 struct object_entry *old_entry = old->entry;
298 unsigned long size, oldsize, delta_size, sizediff;
302 /* Don't bother doing diffs between different types */
303 if (cur_entry->type != old_entry->type)
306 size = cur_entry->size;
309 oldsize = old_entry->size;
310 sizediff = oldsize > size ? oldsize - size : size - oldsize;
311 if (sizediff > size / 8)
313 if (old_entry->depth >= max_depth)
319 * We always delta from the bigger to the smaller, since that's
320 * more space-efficient (deletes don't have to say _what_ they
323 max_size = size / 2 - 20;
324 if (cur_entry->delta)
325 max_size = cur_entry->delta_size-1;
326 if (sizediff >= max_size)
328 delta_buf = diff_delta(old->data, oldsize,
329 cur->data, size, &delta_size, max_size);
332 cur_entry->delta = old_entry;
333 cur_entry->delta_size = delta_size;
334 cur_entry->depth = old_entry->depth + 1;
339 static void find_deltas(struct object_entry **list, int window, int depth)
342 unsigned int array_size = window * sizeof(struct unpacked);
343 struct unpacked *array = xmalloc(array_size);
345 memset(array, 0, array_size);
349 struct object_entry *entry = list[i];
350 struct unpacked *n = array + idx;
357 n->data = read_sha1_file(entry->sha1, type, &size);
358 if (size != entry->size)
359 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
362 unsigned int other_idx = idx + j;
364 if (other_idx >= window)
366 m = array + other_idx;
369 if (try_delta(n, m, depth) < 0)
378 int main(int argc, char **argv)
380 char line[PATH_MAX + 20];
381 int window = 10, depth = 10, pack_to_stdout = 0;
384 for (i = 1; i < argc; i++) {
385 const char *arg = argv[i];
388 if (!strncmp("--window=", arg, 9)) {
390 window = strtoul(arg+9, &end, 0);
395 if (!strncmp("--depth=", arg, 8)) {
397 depth = strtoul(arg+8, &end, 0);
402 if (!strcmp("--stdout", arg)) {
413 if (pack_to_stdout != !base_name)
416 while (fgets(line, sizeof(line), stdin) != NULL) {
419 unsigned char sha1[20];
421 if (get_sha1_hex(line, sha1))
422 die("expected sha1, got garbage");
426 unsigned char c = *p++;
429 hash = hash * 11 + c;
431 add_object_entry(sha1, hash);
433 get_object_details();
435 fprintf(stderr, "Packing %d objects\n", nr_objects);
437 sorted_by_sha = create_sorted_list(sha1_sort);
438 sorted_by_type = create_sorted_list(type_size_sort);
440 find_deltas(sorted_by_type, window+1, depth);