5 static const char pack_usage[] = "git-pack-objects [--window=N] base-name < object-list";
12 OBJ_DELTA // NOTE! This is _not_ the same as a "delta" object in the filesystem
16 unsigned char sha1[20];
21 enum object_type type;
22 unsigned long delta_size;
23 struct object_entry *delta;
26 static struct object_entry **sorted_by_sha, **sorted_by_type;
27 static struct object_entry *objects = NULL;
28 static int nr_objects = 0, nr_alloc = 0;
29 static int delta_window = 10;
30 static const char *base_name;
35 unsigned char buffer[8192];
38 static FILE *create_file(const char *suffix)
40 static char filename[PATH_MAX];
43 len = snprintf(filename, PATH_MAX, "%s.%s", base_name, suffix);
45 die("you wascally wabbit, you");
46 return fopen(filename, "w");
49 static unsigned long fwrite_compressed(void *in, unsigned long size, FILE *f)
52 unsigned long maxsize;
55 memset(&stream, 0, sizeof(stream));
56 deflateInit(&stream, Z_DEFAULT_COMPRESSION);
57 maxsize = deflateBound(&stream, size);
58 out = xmalloc(maxsize);
62 stream.avail_in = size;
64 stream.next_out = out;
65 stream.avail_out = maxsize;
67 while (deflate(&stream, Z_FINISH) == Z_OK)
71 size = stream.total_out;
72 fwrite(out, size, 1, f);
77 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
79 unsigned long othersize, delta_size;
81 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
85 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
86 delta_buf = diff_delta(buf, size, otherbuf, othersize, &delta_size);
87 if (delta_size != entry->delta_size)
88 die("delta size changed");
94 static unsigned long write_object(FILE *f, struct object_entry *entry)
98 void *buf = read_sha1_file(entry->sha1, type, &size);
100 unsigned hdrlen, datalen;
103 die("unable to read %s", sha1_to_hex(entry->sha1));
104 if (size != entry->size)
105 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
108 * The object header is a byte of 'type' followed by four bytes of
109 * length, except for deltas that has the 20 bytes of delta sha
112 header[0] = ".CTB"[entry->type];
113 datalen = htonl(size);
114 memcpy(header+1, &datalen, 4);
117 memcpy(header+1, entry->delta, 20);
118 buf = delta_against(buf, size, entry);
119 size = entry->delta_size;
122 fwrite(header, hdrlen, 1, f);
123 datalen = fwrite_compressed(buf, size, f);
125 return hdrlen + datalen;
128 static void write_pack_file(void)
131 FILE *f = create_file("pack");
132 unsigned long offset = 0;
135 for (i = 0; i < nr_objects; i++) {
136 struct object_entry *entry = objects + i;
137 entry->offset = offset;
138 offset += write_object(f, entry);
145 static void write_index_file(void)
148 FILE *f = create_file("idx");
149 struct object_entry **list = sorted_by_sha;
150 struct object_entry **last = list + nr_objects;
151 unsigned int array[256];
154 * Write the first-level table (the list is sorted,
155 * but we use a 256-entry lookup to be able to avoid
156 * having to do eight extra binary search iterations)
158 for (i = 0; i < 256; i++) {
159 struct object_entry **next = list;
160 while (next < last) {
161 struct object_entry *entry = *next;
162 if (entry->sha1[0] != i)
166 array[i] = htonl(next - sorted_by_sha);
169 fwrite(array, 256, sizeof(int), f);
172 * Write the actual SHA1 entries..
174 list = sorted_by_sha;
175 for (i = 0; i < nr_objects; i++) {
176 struct object_entry *entry = *list++;
177 unsigned int offset = htonl(entry->offset);
178 fwrite(&offset, 4, 1, f);
179 fwrite(entry->sha1, 20, 1, f);
184 static void add_object_entry(unsigned char *sha1)
186 unsigned int idx = nr_objects;
187 struct object_entry *entry;
189 if (idx >= nr_alloc) {
190 unsigned int needed = (idx + 1024) * 3 / 2;
191 objects = xrealloc(objects, needed * sizeof(*entry));
194 entry = objects + idx;
195 memset(entry, 0, sizeof(*entry));
196 memcpy(entry->sha1, sha1, 20);
200 static void check_object(struct object_entry *entry)
204 unsigned long mapsize;
208 map = map_sha1_file(entry->sha1, &mapsize);
210 die("unable to map %s", sha1_to_hex(entry->sha1));
211 if (unpack_sha1_header(&stream, map, mapsize, buffer, sizeof(buffer)) < 0)
212 die("unable to unpack %s header", sha1_to_hex(entry->sha1));
213 munmap(map, mapsize);
214 if (parse_sha1_header(buffer, type, &entry->size) < 0)
215 die("unable to parse %s header", sha1_to_hex(entry->sha1));
216 if (!strcmp(type, "commit")) {
217 entry->type = OBJ_COMMIT;
218 } else if (!strcmp(type, "tree")) {
219 entry->type = OBJ_TREE;
220 } else if (!strcmp(type, "blob")) {
221 entry->type = OBJ_BLOB;
223 die("unable to pack object %s of type %s", sha1_to_hex(entry->sha1), type);
226 static void get_object_details(void)
229 struct object_entry *entry = objects;
231 for (i = 0; i < nr_objects; i++)
232 check_object(entry++);
235 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
237 static entry_sort_t current_sort;
239 static int sort_comparator(const void *_a, const void *_b)
241 struct object_entry *a = *(struct object_entry **)_a;
242 struct object_entry *b = *(struct object_entry **)_b;
243 return current_sort(a,b);
246 static struct object_entry **create_sorted_list(entry_sort_t sort)
248 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
251 for (i = 0; i < nr_objects; i++)
252 list[i] = objects + i;
254 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
258 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
260 return memcmp(a->sha1, b->sha1, 20);
263 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
265 if (a->type < b->type)
267 if (a->type > b->type)
269 if (a->size < b->size)
271 if (a->size > b->size)
273 return a < b ? -1 : (a > b);
277 struct object_entry *entry;
282 * We search for deltas in a list sorted by type and by size, and
283 * walk the "old" chain backwards in the list, so if the type
284 * has changed or the size difference is too big, there's no point
285 * in even continuing the walk, since the other old objects are
286 * going to be even smaller or of a different type. So return -1
287 * once we determine that there's no point even trying.
289 static int try_delta(struct unpacked *cur, struct unpacked *old)
291 struct object_entry *cur_entry = cur->entry;
292 struct object_entry *old_entry = old->entry;
293 unsigned long size, oldsize, delta_size;
296 /* Don't bother doing diffs between different types */
297 if (cur_entry->type != old_entry->type)
300 /* Size is guaranteed to be larger than or equal to oldsize */
301 size = cur_entry->size;
302 oldsize = old_entry->size;
303 if (size - oldsize > oldsize / 4)
309 * We always delta from the bigger to the smaller, since that's
310 * more space-efficient (deletes don't have to say _what_ they
313 delta_buf = diff_delta(cur->data, size, old->data, oldsize, &delta_size);
315 die("unable to create delta");
316 if (delta_size + 20 < size / 2) {
317 if (!cur_entry->delta || cur_entry->delta_size > delta_size) {
318 cur_entry->delta = old_entry;
319 cur_entry->delta_size = delta_size;
325 static void find_deltas(struct object_entry **list, int window)
328 unsigned int array_size = window * sizeof(struct unpacked);
329 struct unpacked *array = xmalloc(array_size);
331 memset(array, 0, array_size);
332 for (i = 0; i < nr_objects; i++) {
333 unsigned int idx = i % window;
334 struct object_entry *entry = list[i];
335 struct unpacked *n = array + idx;
342 n->data = read_sha1_file(entry->sha1, type, &size);
343 if (size != entry->size)
344 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
345 for (j = 0; j < window; j++) {
346 unsigned int other_idx = idx - 1 - j;
350 m = array + other_idx;
353 if (try_delta(n, m) < 0)
359 int main(int argc, char **argv)
364 for (i = 1; i < argc; i++) {
365 const char *arg = argv[i];
368 if (!strncmp("--window=", arg, 9)) {
370 delta_window = strtoul(arg+9, &end, 0);
385 while (fgets(line, sizeof(line), stdin) != NULL) {
386 unsigned char sha1[20];
387 if (get_sha1_hex(line, sha1))
388 die("expected sha1, got garbage");
389 add_object_entry(sha1);
391 get_object_details();
393 printf("Packing %d objects\n", nr_objects);
395 sorted_by_sha = create_sorted_list(sha1_sort);
396 sorted_by_type = create_sorted_list(type_size_sort);
398 find_deltas(sorted_by_type, delta_window);