Fix what to do and how to detect when hardlinking fails
[git/git.git] / pack-objects.c
CommitLineData
c323ac7d
LT
1#include "cache.h"
2#include "object.h"
3#include "delta.h"
a733cb60 4#include "pack.h"
c38138cd 5#include "csum-file.h"
c323ac7d 6
64560374 7static const char pack_usage[] = "git-pack-objects [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
c323ac7d 8
c323ac7d
LT
9struct object_entry {
10 unsigned char sha1[20];
11 unsigned long size;
12 unsigned long offset;
13 unsigned int depth;
27225f2e 14 unsigned int hash;
a733cb60 15 enum object_type type;
c323ac7d
LT
16 unsigned long delta_size;
17 struct object_entry *delta;
18};
19
5f3de58f 20static unsigned char object_list_sha1[20];
1c4a2912 21static int non_empty = 0;
64560374 22static int local = 0;
eb019375 23static int incremental = 0;
c323ac7d
LT
24static struct object_entry **sorted_by_sha, **sorted_by_type;
25static struct object_entry *objects = NULL;
26static int nr_objects = 0, nr_alloc = 0;
c323ac7d 27static const char *base_name;
e1808845 28static unsigned char pack_file_sha1[20];
c323ac7d 29
c323ac7d
LT
30static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
31{
32 unsigned long othersize, delta_size;
33 char type[10];
34 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
35 void *delta_buf;
36
37 if (!otherbuf)
38 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
8ee378a0 39 delta_buf = diff_delta(otherbuf, othersize,
dcde55bc 40 buf, size, &delta_size, 0);
75c42d8c 41 if (!delta_buf || delta_size != entry->delta_size)
c323ac7d
LT
42 die("delta size changed");
43 free(buf);
44 free(otherbuf);
45 return delta_buf;
46}
47
a733cb60
LT
48/*
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"
54 */
55static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
56{
01247d87 57 int n = 1;
a733cb60
LT
58 unsigned char c;
59
60 if (type < OBJ_COMMIT || type > OBJ_DELTA)
61 die("bad type %d", type);
62
01247d87
LT
63 c = (type << 4) | (size & 15);
64 size >>= 4;
65 while (size) {
a733cb60 66 *hdr++ = c | 0x80;
01247d87
LT
67 c = size & 0x7f;
68 size >>= 7;
69 n++;
a733cb60
LT
70 }
71 *hdr = c;
72 return n;
73}
74
c38138cd 75static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
c323ac7d
LT
76{
77 unsigned long size;
78 char type[10];
79 void *buf = read_sha1_file(entry->sha1, type, &size);
a733cb60 80 unsigned char header[10];
c323ac7d 81 unsigned hdrlen, datalen;
a733cb60 82 enum object_type obj_type;
c323ac7d
LT
83
84 if (!buf)
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);
88
89 /*
e1ddc976
JH
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
92 * follows that.
c323ac7d 93 */
a733cb60 94 obj_type = entry->type;
c323ac7d 95 if (entry->delta) {
c323ac7d
LT
96 buf = delta_against(buf, size, entry);
97 size = entry->delta_size;
a733cb60 98 obj_type = OBJ_DELTA;
c323ac7d 99 }
a733cb60 100 hdrlen = encode_header(obj_type, size, header);
c38138cd 101 sha1write(f, header, hdrlen);
a733cb60
LT
102 if (entry->delta) {
103 sha1write(f, entry->delta, 20);
104 hdrlen += 20;
105 }
c38138cd 106 datalen = sha1write_compressed(f, buf, size);
c323ac7d
LT
107 free(buf);
108 return hdrlen + datalen;
109}
110
9d5ab962
JH
111static unsigned long write_one(struct sha1file *f,
112 struct object_entry *e,
113 unsigned long offset)
114{
115 if (e->offset)
116 /* offset starts from header size and cannot be zero
117 * if it is written already.
118 */
119 return offset;
120 e->offset = offset;
121 offset += write_object(f, e);
122 /* if we are delitified, write out its base object. */
123 if (e->delta)
124 offset = write_one(f, e->delta, offset);
125 return offset;
126}
127
c323ac7d
LT
128static void write_pack_file(void)
129{
130 int i;
d22b9290 131 struct sha1file *f;
a733cb60 132 unsigned long offset;
c323ac7d 133 unsigned long mb;
a733cb60 134 struct pack_header hdr;
c323ac7d 135
d22b9290
LT
136 if (!base_name)
137 f = sha1fd(1, "<stdout>");
138 else
5f3de58f 139 f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "pack");
a733cb60 140 hdr.hdr_signature = htonl(PACK_SIGNATURE);
01247d87 141 hdr.hdr_version = htonl(PACK_VERSION);
a733cb60
LT
142 hdr.hdr_entries = htonl(nr_objects);
143 sha1write(f, &hdr, sizeof(hdr));
144 offset = sizeof(hdr);
9d5ab962
JH
145 for (i = 0; i < nr_objects; i++)
146 offset = write_one(f, objects + i, offset);
147
e1808845 148 sha1close(f, pack_file_sha1, 1);
c323ac7d
LT
149 mb = offset >> 20;
150 offset &= 0xfffff;
151}
152
153static void write_index_file(void)
154{
155 int i;
5f3de58f 156 struct sha1file *f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "idx");
c323ac7d
LT
157 struct object_entry **list = sorted_by_sha;
158 struct object_entry **last = list + nr_objects;
159 unsigned int array[256];
160
161 /*
162 * Write the first-level table (the list is sorted,
163 * but we use a 256-entry lookup to be able to avoid
e1808845 164 * having to do eight extra binary search iterations).
c323ac7d
LT
165 */
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)
171 break;
172 next++;
173 }
174 array[i] = htonl(next - sorted_by_sha);
175 list = next;
176 }
c38138cd 177 sha1write(f, array, 256 * sizeof(int));
c323ac7d
LT
178
179 /*
180 * Write the actual SHA1 entries..
181 */
182 list = sorted_by_sha;
49397104 183 for (i = 0; i < nr_objects; i++) {
c323ac7d
LT
184 struct object_entry *entry = *list++;
185 unsigned int offset = htonl(entry->offset);
c38138cd
LT
186 sha1write(f, &offset, 4);
187 sha1write(f, entry->sha1, 20);
c323ac7d 188 }
e1808845
LT
189 sha1write(f, pack_file_sha1, 20);
190 sha1close(f, NULL, 1);
c323ac7d
LT
191}
192
5f3de58f 193static int add_object_entry(unsigned char *sha1, unsigned int hash)
c323ac7d
LT
194{
195 unsigned int idx = nr_objects;
196 struct object_entry *entry;
197
64560374
LT
198 if (incremental || local) {
199 struct packed_git *p;
200
201 for (p = packed_git; p; p = p->next) {
202 struct pack_entry e;
203
204 if (find_pack_entry_one(sha1, &e, p)) {
205 if (incremental)
206 return 0;
207 if (local && !p->pack_local)
208 return 0;
209 }
210 }
211 }
eb019375 212
c323ac7d
LT
213 if (idx >= nr_alloc) {
214 unsigned int needed = (idx + 1024) * 3 / 2;
215 objects = xrealloc(objects, needed * sizeof(*entry));
216 nr_alloc = needed;
217 }
218 entry = objects + idx;
219 memset(entry, 0, sizeof(*entry));
220 memcpy(entry->sha1, sha1, 20);
27225f2e 221 entry->hash = hash;
c323ac7d 222 nr_objects = idx+1;
5f3de58f 223 return 1;
c323ac7d
LT
224}
225
226static void check_object(struct object_entry *entry)
227{
36e4d74a
JH
228 char type[20];
229
230 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
231 if (!strcmp(type, "commit")) {
a733cb60 232 entry->type = OBJ_COMMIT;
36e4d74a 233 } else if (!strcmp(type, "tree")) {
a733cb60 234 entry->type = OBJ_TREE;
36e4d74a 235 } else if (!strcmp(type, "blob")) {
a733cb60 236 entry->type = OBJ_BLOB;
a69d0943 237 } else if (!strcmp(type, "tag")) {
a733cb60 238 entry->type = OBJ_TAG;
36e4d74a
JH
239 } else
240 die("unable to pack object %s of type %s",
241 sha1_to_hex(entry->sha1), type);
242 }
243 else
244 die("unable to get type of object %s",
245 sha1_to_hex(entry->sha1));
c323ac7d
LT
246}
247
248static void get_object_details(void)
249{
250 int i;
251 struct object_entry *entry = objects;
252
253 for (i = 0; i < nr_objects; i++)
254 check_object(entry++);
255}
256
257typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
258
259static entry_sort_t current_sort;
260
261static int sort_comparator(const void *_a, const void *_b)
262{
263 struct object_entry *a = *(struct object_entry **)_a;
264 struct object_entry *b = *(struct object_entry **)_b;
265 return current_sort(a,b);
266}
267
268static struct object_entry **create_sorted_list(entry_sort_t sort)
269{
270 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
271 int i;
272
273 for (i = 0; i < nr_objects; i++)
274 list[i] = objects + i;
275 current_sort = sort;
276 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
277 return list;
278}
279
280static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
281{
282 return memcmp(a->sha1, b->sha1, 20);
283}
284
285static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
286{
287 if (a->type < b->type)
288 return -1;
289 if (a->type > b->type)
290 return 1;
27225f2e
LT
291 if (a->hash < b->hash)
292 return -1;
293 if (a->hash > b->hash)
294 return 1;
c323ac7d
LT
295 if (a->size < b->size)
296 return -1;
297 if (a->size > b->size)
298 return 1;
299 return a < b ? -1 : (a > b);
300}
301
302struct unpacked {
303 struct object_entry *entry;
304 void *data;
305};
306
307/*
521a4f4c
LT
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
313 * one.
c323ac7d 314 */
d116a45a 315static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
c323ac7d
LT
316{
317 struct object_entry *cur_entry = cur->entry;
318 struct object_entry *old_entry = old->entry;
521a4f4c 319 unsigned long size, oldsize, delta_size, sizediff;
75c42d8c 320 long max_size;
c323ac7d
LT
321 void *delta_buf;
322
323 /* Don't bother doing diffs between different types */
324 if (cur_entry->type != old_entry->type)
325 return -1;
326
c323ac7d 327 size = cur_entry->size;
75c42d8c
LT
328 if (size < 50)
329 return -1;
c323ac7d 330 oldsize = old_entry->size;
521a4f4c
LT
331 sizediff = oldsize > size ? oldsize - size : size - oldsize;
332 if (sizediff > size / 8)
c323ac7d 333 return -1;
d116a45a
LT
334 if (old_entry->depth >= max_depth)
335 return 0;
c323ac7d
LT
336
337 /*
338 * NOTE!
339 *
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
342 * delete).
343 */
75c42d8c
LT
344 max_size = size / 2 - 20;
345 if (cur_entry->delta)
346 max_size = cur_entry->delta_size-1;
27225f2e
LT
347 if (sizediff >= max_size)
348 return -1;
8ee378a0
JH
349 delta_buf = diff_delta(old->data, oldsize,
350 cur->data, size, &delta_size, max_size);
c323ac7d 351 if (!delta_buf)
75c42d8c
LT
352 return 0;
353 cur_entry->delta = old_entry;
354 cur_entry->delta_size = delta_size;
d116a45a 355 cur_entry->depth = old_entry->depth + 1;
c323ac7d 356 free(delta_buf);
eb41ab11 357 return 0;
c323ac7d
LT
358}
359
d116a45a 360static void find_deltas(struct object_entry **list, int window, int depth)
c323ac7d 361{
521a4f4c 362 int i, idx;
c323ac7d
LT
363 unsigned int array_size = window * sizeof(struct unpacked);
364 struct unpacked *array = xmalloc(array_size);
365
366 memset(array, 0, array_size);
521a4f4c
LT
367 i = nr_objects;
368 idx = 0;
369 while (--i >= 0) {
c323ac7d
LT
370 struct object_entry *entry = list[i];
371 struct unpacked *n = array + idx;
372 unsigned long size;
373 char type[10];
374 int j;
375
376 free(n->data);
377 n->entry = entry;
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);
78817c15
LT
381 j = window;
382 while (--j > 0) {
383 unsigned int other_idx = idx + j;
c323ac7d 384 struct unpacked *m;
78817c15
LT
385 if (other_idx >= window)
386 other_idx -= window;
c323ac7d
LT
387 m = array + other_idx;
388 if (!m->entry)
389 break;
d116a45a 390 if (try_delta(n, m, depth) < 0)
c323ac7d
LT
391 break;
392 }
521a4f4c
LT
393 idx++;
394 if (idx >= window)
395 idx = 0;
c323ac7d 396 }
adee7bdf
SV
397
398 for (i = 0; i < window; ++i)
399 free(array[i].data);
400 free(array);
c323ac7d
LT
401}
402
403int main(int argc, char **argv)
404{
5f3de58f 405 SHA_CTX ctx;
27225f2e 406 char line[PATH_MAX + 20];
d22b9290 407 int window = 10, depth = 10, pack_to_stdout = 0;
84c8d8ae 408 struct object_entry **list;
c323ac7d
LT
409 int i;
410
411 for (i = 1; i < argc; i++) {
412 const char *arg = argv[i];
413
414 if (*arg == '-') {
1c4a2912
LT
415 if (!strcmp("--non-empty", arg)) {
416 non_empty = 1;
417 continue;
418 }
64560374
LT
419 if (!strcmp("--local", arg)) {
420 local = 1;
421 continue;
422 }
eb019375
LT
423 if (!strcmp("--incremental", arg)) {
424 incremental = 1;
425 continue;
426 }
c323ac7d
LT
427 if (!strncmp("--window=", arg, 9)) {
428 char *end;
f846bbff 429 window = strtoul(arg+9, &end, 0);
c323ac7d
LT
430 if (!arg[9] || *end)
431 usage(pack_usage);
432 continue;
433 }
d116a45a
LT
434 if (!strncmp("--depth=", arg, 8)) {
435 char *end;
436 depth = strtoul(arg+8, &end, 0);
437 if (!arg[8] || *end)
438 usage(pack_usage);
439 continue;
440 }
d22b9290
LT
441 if (!strcmp("--stdout", arg)) {
442 pack_to_stdout = 1;
443 continue;
444 }
c323ac7d
LT
445 usage(pack_usage);
446 }
447 if (base_name)
448 usage(pack_usage);
449 base_name = arg;
450 }
451
d22b9290 452 if (pack_to_stdout != !base_name)
c323ac7d
LT
453 usage(pack_usage);
454
64560374 455 prepare_packed_git();
c323ac7d 456 while (fgets(line, sizeof(line), stdin) != NULL) {
27225f2e
LT
457 unsigned int hash;
458 char *p;
c323ac7d 459 unsigned char sha1[20];
27225f2e 460
c323ac7d
LT
461 if (get_sha1_hex(line, sha1))
462 die("expected sha1, got garbage");
27225f2e
LT
463 hash = 0;
464 p = line+40;
465 while (*p) {
466 unsigned char c = *p++;
467 if (isspace(c))
468 continue;
469 hash = hash * 11 + c;
470 }
84c8d8ae 471 add_object_entry(sha1, hash);
c323ac7d 472 }
1c4a2912
LT
473 if (non_empty && !nr_objects)
474 return 0;
c323ac7d
LT
475 get_object_details();
476
d22b9290 477 fprintf(stderr, "Packing %d objects\n", nr_objects);
c323ac7d
LT
478
479 sorted_by_sha = create_sorted_list(sha1_sort);
84c8d8ae
JH
480 SHA1_Init(&ctx);
481 list = sorted_by_sha;
482 for (i = 0; i < nr_objects; i++) {
483 struct object_entry *entry = *list++;
484 SHA1_Update(&ctx, entry->sha1, 20);
485 }
486 SHA1_Final(object_list_sha1, &ctx);
487
c323ac7d 488 sorted_by_type = create_sorted_list(type_size_sort);
d116a45a
LT
489 if (window && depth)
490 find_deltas(sorted_by_type, window+1, depth);
c323ac7d
LT
491
492 write_pack_file();
5f3de58f 493 if (!pack_to_stdout) {
d22b9290 494 write_index_file();
5f3de58f
LT
495 puts(sha1_to_hex(object_list_sha1));
496 }
c323ac7d
LT
497 return 0;
498}