Teach packing about "tag" objects
[git.git] / pack-objects.c
blob90b27c60c3a889d7534e17af51b3e8b79b1d2b24
1 #include <ctype.h>
2 #include "cache.h"
3 #include "object.h"
4 #include "delta.h"
5 #include "csum-file.h"
7 static const char pack_usage[] = "git-pack-objects [--window=N] [--depth=N] base-name < object-list";
9 /*
10 * The object type is a single-character shorthand:
11 * - 'C' for "Commit"
12 * - 'T' for "Tree"
13 * - 'B' for "Blob"
14 * - 'G' for "taG"
15 * - 'D' for "Delta"
17 struct object_entry {
18 unsigned char sha1[20];
19 unsigned long size;
20 unsigned long offset;
21 unsigned int depth;
22 unsigned int hash;
23 unsigned char type;
24 unsigned long delta_size;
25 struct object_entry *delta;
28 static struct object_entry **sorted_by_sha, **sorted_by_type;
29 static struct object_entry *objects = NULL;
30 static int nr_objects = 0, nr_alloc = 0;
31 static const char *base_name;
32 static unsigned char pack_file_sha1[20];
34 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
36 unsigned long othersize, delta_size;
37 char type[10];
38 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
39 void *delta_buf;
41 if (!otherbuf)
42 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
43 delta_buf = diff_delta(otherbuf, othersize,
44 buf, size, &delta_size, ~0UL);
45 if (!delta_buf || delta_size != entry->delta_size)
46 die("delta size changed");
47 free(buf);
48 free(otherbuf);
49 return delta_buf;
52 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
54 unsigned long size;
55 char type[10];
56 void *buf = read_sha1_file(entry->sha1, type, &size);
57 char header[25];
58 unsigned hdrlen, datalen;
60 if (!buf)
61 die("unable to read %s", sha1_to_hex(entry->sha1));
62 if (size != entry->size)
63 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
66 * The object header is a byte of 'type' followed by four bytes of
67 * length, except for deltas that has the 20 bytes of delta sha
68 * instead.
70 header[0] = entry->type;
71 hdrlen = 5;
72 if (entry->delta) {
73 header[0] = 'D';
74 memcpy(header+5, entry->delta, 20);
75 buf = delta_against(buf, size, entry);
76 size = entry->delta_size;
77 hdrlen = 25;
79 datalen = htonl(size);
80 memcpy(header+1, &datalen, 4);
81 sha1write(f, header, hdrlen);
82 datalen = sha1write_compressed(f, buf, size);
83 free(buf);
84 return hdrlen + datalen;
87 static void write_pack_file(void)
89 int i;
90 struct sha1file *f = sha1create("%s.%s", base_name, "pack");
91 unsigned long offset = 0;
92 unsigned long mb;
94 for (i = 0; i < nr_objects; i++) {
95 struct object_entry *entry = objects + i;
96 entry->offset = offset;
97 offset += write_object(f, entry);
99 sha1close(f, pack_file_sha1, 1);
100 mb = offset >> 20;
101 offset &= 0xfffff;
104 static void write_index_file(void)
106 int i;
107 struct sha1file *f = sha1create("%s.%s", base_name, "idx");
108 struct object_entry **list = sorted_by_sha;
109 struct object_entry **last = list + nr_objects;
110 unsigned int array[256];
113 * Write the first-level table (the list is sorted,
114 * but we use a 256-entry lookup to be able to avoid
115 * having to do eight extra binary search iterations).
117 for (i = 0; i < 256; i++) {
118 struct object_entry **next = list;
119 while (next < last) {
120 struct object_entry *entry = *next;
121 if (entry->sha1[0] != i)
122 break;
123 next++;
125 array[i] = htonl(next - sorted_by_sha);
126 list = next;
128 sha1write(f, array, 256 * sizeof(int));
131 * Write the actual SHA1 entries..
133 list = sorted_by_sha;
134 for (i = 0; i < nr_objects; i++) {
135 struct object_entry *entry = *list++;
136 unsigned int offset = htonl(entry->offset);
137 sha1write(f, &offset, 4);
138 sha1write(f, entry->sha1, 20);
140 sha1write(f, pack_file_sha1, 20);
141 sha1close(f, NULL, 1);
144 static void add_object_entry(unsigned char *sha1, unsigned int hash)
146 unsigned int idx = nr_objects;
147 struct object_entry *entry;
149 if (idx >= nr_alloc) {
150 unsigned int needed = (idx + 1024) * 3 / 2;
151 objects = xrealloc(objects, needed * sizeof(*entry));
152 nr_alloc = needed;
154 entry = objects + idx;
155 memset(entry, 0, sizeof(*entry));
156 memcpy(entry->sha1, sha1, 20);
157 entry->hash = hash;
158 nr_objects = idx+1;
161 static void check_object(struct object_entry *entry)
163 char type[20];
165 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
166 if (!strcmp(type, "commit")) {
167 entry->type = 'C';
168 } else if (!strcmp(type, "tree")) {
169 entry->type = 'T';
170 } else if (!strcmp(type, "blob")) {
171 entry->type = 'B';
172 } else if (!strcmp(type, "tag")) {
173 entry->type = 'G';
174 } else
175 die("unable to pack object %s of type %s",
176 sha1_to_hex(entry->sha1), type);
178 else
179 die("unable to get type of object %s",
180 sha1_to_hex(entry->sha1));
183 static void get_object_details(void)
185 int i;
186 struct object_entry *entry = objects;
188 for (i = 0; i < nr_objects; i++)
189 check_object(entry++);
192 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
194 static entry_sort_t current_sort;
196 static int sort_comparator(const void *_a, const void *_b)
198 struct object_entry *a = *(struct object_entry **)_a;
199 struct object_entry *b = *(struct object_entry **)_b;
200 return current_sort(a,b);
203 static struct object_entry **create_sorted_list(entry_sort_t sort)
205 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
206 int i;
208 for (i = 0; i < nr_objects; i++)
209 list[i] = objects + i;
210 current_sort = sort;
211 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
212 return list;
215 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
217 return memcmp(a->sha1, b->sha1, 20);
220 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
222 if (a->type < b->type)
223 return -1;
224 if (a->type > b->type)
225 return 1;
226 if (a->hash < b->hash)
227 return -1;
228 if (a->hash > b->hash)
229 return 1;
230 if (a->size < b->size)
231 return -1;
232 if (a->size > b->size)
233 return 1;
234 return a < b ? -1 : (a > b);
237 struct unpacked {
238 struct object_entry *entry;
239 void *data;
243 * We search for deltas _backwards_ in a list sorted by type and
244 * by size, so that we see progressively smaller and smaller files.
245 * That's because we prefer deltas to be from the bigger file
246 * to the smaller - deletes are potentially cheaper, but perhaps
247 * more importantly, the bigger file is likely the more recent
248 * one.
250 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
252 struct object_entry *cur_entry = cur->entry;
253 struct object_entry *old_entry = old->entry;
254 unsigned long size, oldsize, delta_size, sizediff;
255 long max_size;
256 void *delta_buf;
258 /* Don't bother doing diffs between different types */
259 if (cur_entry->type != old_entry->type)
260 return -1;
262 size = cur_entry->size;
263 if (size < 50)
264 return -1;
265 oldsize = old_entry->size;
266 sizediff = oldsize > size ? oldsize - size : size - oldsize;
267 if (sizediff > size / 8)
268 return -1;
269 if (old_entry->depth >= max_depth)
270 return 0;
273 * NOTE!
275 * We always delta from the bigger to the smaller, since that's
276 * more space-efficient (deletes don't have to say _what_ they
277 * delete).
279 max_size = size / 2 - 20;
280 if (cur_entry->delta)
281 max_size = cur_entry->delta_size-1;
282 if (sizediff >= max_size)
283 return -1;
284 delta_buf = diff_delta(old->data, oldsize,
285 cur->data, size, &delta_size, max_size);
286 if (!delta_buf)
287 return 0;
288 cur_entry->delta = old_entry;
289 cur_entry->delta_size = delta_size;
290 cur_entry->depth = old_entry->depth + 1;
291 free(delta_buf);
292 return 0;
295 static void find_deltas(struct object_entry **list, int window, int depth)
297 int i, idx;
298 unsigned int array_size = window * sizeof(struct unpacked);
299 struct unpacked *array = xmalloc(array_size);
301 memset(array, 0, array_size);
302 i = nr_objects;
303 idx = 0;
304 while (--i >= 0) {
305 struct object_entry *entry = list[i];
306 struct unpacked *n = array + idx;
307 unsigned long size;
308 char type[10];
309 int j;
311 free(n->data);
312 n->entry = entry;
313 n->data = read_sha1_file(entry->sha1, type, &size);
314 if (size != entry->size)
315 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
316 j = window;
317 while (--j > 0) {
318 unsigned int other_idx = idx + j;
319 struct unpacked *m;
320 if (other_idx >= window)
321 other_idx -= window;
322 m = array + other_idx;
323 if (!m->entry)
324 break;
325 if (try_delta(n, m, depth) < 0)
326 break;
328 idx++;
329 if (idx >= window)
330 idx = 0;
334 int main(int argc, char **argv)
336 char line[PATH_MAX + 20];
337 int window = 10, depth = 10;
338 int i;
340 for (i = 1; i < argc; i++) {
341 const char *arg = argv[i];
343 if (*arg == '-') {
344 if (!strncmp("--window=", arg, 9)) {
345 char *end;
346 window = strtoul(arg+9, &end, 0);
347 if (!arg[9] || *end)
348 usage(pack_usage);
349 continue;
351 if (!strncmp("--depth=", arg, 8)) {
352 char *end;
353 depth = strtoul(arg+8, &end, 0);
354 if (!arg[8] || *end)
355 usage(pack_usage);
356 continue;
358 usage(pack_usage);
360 if (base_name)
361 usage(pack_usage);
362 base_name = arg;
365 if (!base_name)
366 usage(pack_usage);
368 while (fgets(line, sizeof(line), stdin) != NULL) {
369 unsigned int hash;
370 char *p;
371 unsigned char sha1[20];
373 if (get_sha1_hex(line, sha1))
374 die("expected sha1, got garbage");
375 hash = 0;
376 p = line+40;
377 while (*p) {
378 unsigned char c = *p++;
379 if (isspace(c))
380 continue;
381 hash = hash * 11 + c;
383 add_object_entry(sha1, hash);
385 get_object_details();
387 printf("Packing %d objects\n", nr_objects);
389 sorted_by_sha = create_sorted_list(sha1_sort);
390 sorted_by_type = create_sorted_list(type_size_sort);
391 if (window && depth)
392 find_deltas(sorted_by_type, window+1, depth);
394 write_pack_file();
395 write_index_file();
396 return 0;