avoid echo -e, there are systems where it does not work
[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"
21fcd1bd 6#include <sys/time.h>
c323ac7d 7
63ae26f8 8static const char pack_usage[] = "git-pack-objects [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
c323ac7d 9
c323ac7d
LT
10struct object_entry {
11 unsigned char sha1[20];
12 unsigned long size;
13 unsigned long offset;
14 unsigned int depth;
27225f2e 15 unsigned int hash;
a733cb60 16 enum object_type type;
c323ac7d
LT
17 unsigned long delta_size;
18 struct object_entry *delta;
19};
20
5f3de58f 21static unsigned char object_list_sha1[20];
1c4a2912 22static int non_empty = 0;
64560374 23static int local = 0;
eb019375 24static int incremental = 0;
c323ac7d
LT
25static struct object_entry **sorted_by_sha, **sorted_by_type;
26static struct object_entry *objects = NULL;
27static int nr_objects = 0, nr_alloc = 0;
c323ac7d 28static const char *base_name;
e1808845 29static unsigned char pack_file_sha1[20];
21fcd1bd 30static int progress = 0;
c323ac7d 31
c323ac7d
LT
32static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
33{
34 unsigned long othersize, delta_size;
35 char type[10];
36 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
37 void *delta_buf;
38
39 if (!otherbuf)
40 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
8ee378a0 41 delta_buf = diff_delta(otherbuf, othersize,
dcde55bc 42 buf, size, &delta_size, 0);
75c42d8c 43 if (!delta_buf || delta_size != entry->delta_size)
c323ac7d
LT
44 die("delta size changed");
45 free(buf);
46 free(otherbuf);
47 return delta_buf;
48}
49
a733cb60
LT
50/*
51 * The per-object header is a pretty dense thing, which is
52 * - first byte: low four bits are "size", then three bits of "type",
53 * and the high bit is "size continues".
54 * - each byte afterwards: low seven bits are size continuation,
55 * with the high bit being "size continues"
56 */
57static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
58{
01247d87 59 int n = 1;
a733cb60
LT
60 unsigned char c;
61
62 if (type < OBJ_COMMIT || type > OBJ_DELTA)
63 die("bad type %d", type);
64
01247d87
LT
65 c = (type << 4) | (size & 15);
66 size >>= 4;
67 while (size) {
a733cb60 68 *hdr++ = c | 0x80;
01247d87
LT
69 c = size & 0x7f;
70 size >>= 7;
71 n++;
a733cb60
LT
72 }
73 *hdr = c;
74 return n;
75}
76
c38138cd 77static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
c323ac7d
LT
78{
79 unsigned long size;
80 char type[10];
81 void *buf = read_sha1_file(entry->sha1, type, &size);
a733cb60 82 unsigned char header[10];
c323ac7d 83 unsigned hdrlen, datalen;
a733cb60 84 enum object_type obj_type;
c323ac7d
LT
85
86 if (!buf)
87 die("unable to read %s", sha1_to_hex(entry->sha1));
88 if (size != entry->size)
89 die("object %s size inconsistency (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
90
91 /*
e1ddc976
JH
92 * The object header is a byte of 'type' followed by zero or
93 * more bytes of length. For deltas, the 20 bytes of delta sha1
94 * follows that.
c323ac7d 95 */
a733cb60 96 obj_type = entry->type;
c323ac7d 97 if (entry->delta) {
c323ac7d
LT
98 buf = delta_against(buf, size, entry);
99 size = entry->delta_size;
a733cb60 100 obj_type = OBJ_DELTA;
c323ac7d 101 }
a733cb60 102 hdrlen = encode_header(obj_type, size, header);
c38138cd 103 sha1write(f, header, hdrlen);
a733cb60
LT
104 if (entry->delta) {
105 sha1write(f, entry->delta, 20);
106 hdrlen += 20;
107 }
c38138cd 108 datalen = sha1write_compressed(f, buf, size);
c323ac7d
LT
109 free(buf);
110 return hdrlen + datalen;
111}
112
9d5ab962
JH
113static unsigned long write_one(struct sha1file *f,
114 struct object_entry *e,
115 unsigned long offset)
116{
117 if (e->offset)
118 /* offset starts from header size and cannot be zero
119 * if it is written already.
120 */
121 return offset;
122 e->offset = offset;
123 offset += write_object(f, e);
82f9d58a 124 /* if we are deltified, write out its base object. */
9d5ab962
JH
125 if (e->delta)
126 offset = write_one(f, e->delta, offset);
127 return offset;
128}
129
c323ac7d
LT
130static void write_pack_file(void)
131{
132 int i;
d22b9290 133 struct sha1file *f;
a733cb60 134 unsigned long offset;
c323ac7d 135 unsigned long mb;
a733cb60 136 struct pack_header hdr;
c323ac7d 137
d22b9290
LT
138 if (!base_name)
139 f = sha1fd(1, "<stdout>");
140 else
5f3de58f 141 f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "pack");
a733cb60 142 hdr.hdr_signature = htonl(PACK_SIGNATURE);
01247d87 143 hdr.hdr_version = htonl(PACK_VERSION);
a733cb60
LT
144 hdr.hdr_entries = htonl(nr_objects);
145 sha1write(f, &hdr, sizeof(hdr));
146 offset = sizeof(hdr);
9d5ab962
JH
147 for (i = 0; i < nr_objects; i++)
148 offset = write_one(f, objects + i, offset);
149
e1808845 150 sha1close(f, pack_file_sha1, 1);
c323ac7d
LT
151 mb = offset >> 20;
152 offset &= 0xfffff;
153}
154
155static void write_index_file(void)
156{
157 int i;
5f3de58f 158 struct sha1file *f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "idx");
c323ac7d
LT
159 struct object_entry **list = sorted_by_sha;
160 struct object_entry **last = list + nr_objects;
161 unsigned int array[256];
162
163 /*
164 * Write the first-level table (the list is sorted,
165 * but we use a 256-entry lookup to be able to avoid
e1808845 166 * having to do eight extra binary search iterations).
c323ac7d
LT
167 */
168 for (i = 0; i < 256; i++) {
169 struct object_entry **next = list;
170 while (next < last) {
171 struct object_entry *entry = *next;
172 if (entry->sha1[0] != i)
173 break;
174 next++;
175 }
176 array[i] = htonl(next - sorted_by_sha);
177 list = next;
178 }
c38138cd 179 sha1write(f, array, 256 * sizeof(int));
c323ac7d
LT
180
181 /*
182 * Write the actual SHA1 entries..
183 */
184 list = sorted_by_sha;
49397104 185 for (i = 0; i < nr_objects; i++) {
c323ac7d
LT
186 struct object_entry *entry = *list++;
187 unsigned int offset = htonl(entry->offset);
c38138cd
LT
188 sha1write(f, &offset, 4);
189 sha1write(f, entry->sha1, 20);
c323ac7d 190 }
e1808845
LT
191 sha1write(f, pack_file_sha1, 20);
192 sha1close(f, NULL, 1);
c323ac7d
LT
193}
194
5f3de58f 195static int add_object_entry(unsigned char *sha1, unsigned int hash)
c323ac7d
LT
196{
197 unsigned int idx = nr_objects;
198 struct object_entry *entry;
199
64560374
LT
200 if (incremental || local) {
201 struct packed_git *p;
202
203 for (p = packed_git; p; p = p->next) {
204 struct pack_entry e;
205
206 if (find_pack_entry_one(sha1, &e, p)) {
207 if (incremental)
208 return 0;
209 if (local && !p->pack_local)
210 return 0;
211 }
212 }
213 }
eb019375 214
c323ac7d
LT
215 if (idx >= nr_alloc) {
216 unsigned int needed = (idx + 1024) * 3 / 2;
217 objects = xrealloc(objects, needed * sizeof(*entry));
218 nr_alloc = needed;
219 }
220 entry = objects + idx;
221 memset(entry, 0, sizeof(*entry));
222 memcpy(entry->sha1, sha1, 20);
27225f2e 223 entry->hash = hash;
c323ac7d 224 nr_objects = idx+1;
5f3de58f 225 return 1;
c323ac7d
LT
226}
227
228static void check_object(struct object_entry *entry)
229{
36e4d74a
JH
230 char type[20];
231
232 if (!sha1_object_info(entry->sha1, type, &entry->size)) {
233 if (!strcmp(type, "commit")) {
a733cb60 234 entry->type = OBJ_COMMIT;
36e4d74a 235 } else if (!strcmp(type, "tree")) {
a733cb60 236 entry->type = OBJ_TREE;
36e4d74a 237 } else if (!strcmp(type, "blob")) {
a733cb60 238 entry->type = OBJ_BLOB;
a69d0943 239 } else if (!strcmp(type, "tag")) {
a733cb60 240 entry->type = OBJ_TAG;
36e4d74a
JH
241 } else
242 die("unable to pack object %s of type %s",
243 sha1_to_hex(entry->sha1), type);
244 }
245 else
246 die("unable to get type of object %s",
247 sha1_to_hex(entry->sha1));
c323ac7d
LT
248}
249
250static void get_object_details(void)
251{
252 int i;
253 struct object_entry *entry = objects;
254
255 for (i = 0; i < nr_objects; i++)
256 check_object(entry++);
257}
258
259typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
260
261static entry_sort_t current_sort;
262
263static int sort_comparator(const void *_a, const void *_b)
264{
265 struct object_entry *a = *(struct object_entry **)_a;
266 struct object_entry *b = *(struct object_entry **)_b;
267 return current_sort(a,b);
268}
269
270static struct object_entry **create_sorted_list(entry_sort_t sort)
271{
272 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
273 int i;
274
275 for (i = 0; i < nr_objects; i++)
276 list[i] = objects + i;
277 current_sort = sort;
278 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
279 return list;
280}
281
282static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
283{
284 return memcmp(a->sha1, b->sha1, 20);
285}
286
287static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
288{
289 if (a->type < b->type)
290 return -1;
291 if (a->type > b->type)
292 return 1;
27225f2e
LT
293 if (a->hash < b->hash)
294 return -1;
295 if (a->hash > b->hash)
296 return 1;
c323ac7d
LT
297 if (a->size < b->size)
298 return -1;
299 if (a->size > b->size)
300 return 1;
301 return a < b ? -1 : (a > b);
302}
303
304struct unpacked {
305 struct object_entry *entry;
306 void *data;
307};
308
309/*
521a4f4c
LT
310 * We search for deltas _backwards_ in a list sorted by type and
311 * by size, so that we see progressively smaller and smaller files.
312 * That's because we prefer deltas to be from the bigger file
313 * to the smaller - deletes are potentially cheaper, but perhaps
314 * more importantly, the bigger file is likely the more recent
315 * one.
c323ac7d 316 */
d116a45a 317static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
c323ac7d
LT
318{
319 struct object_entry *cur_entry = cur->entry;
320 struct object_entry *old_entry = old->entry;
521a4f4c 321 unsigned long size, oldsize, delta_size, sizediff;
75c42d8c 322 long max_size;
c323ac7d
LT
323 void *delta_buf;
324
325 /* Don't bother doing diffs between different types */
326 if (cur_entry->type != old_entry->type)
327 return -1;
328
c323ac7d 329 size = cur_entry->size;
75c42d8c
LT
330 if (size < 50)
331 return -1;
c323ac7d 332 oldsize = old_entry->size;
521a4f4c
LT
333 sizediff = oldsize > size ? oldsize - size : size - oldsize;
334 if (sizediff > size / 8)
c323ac7d 335 return -1;
d116a45a
LT
336 if (old_entry->depth >= max_depth)
337 return 0;
c323ac7d
LT
338
339 /*
340 * NOTE!
341 *
342 * We always delta from the bigger to the smaller, since that's
343 * more space-efficient (deletes don't have to say _what_ they
344 * delete).
345 */
75c42d8c
LT
346 max_size = size / 2 - 20;
347 if (cur_entry->delta)
348 max_size = cur_entry->delta_size-1;
27225f2e
LT
349 if (sizediff >= max_size)
350 return -1;
8ee378a0
JH
351 delta_buf = diff_delta(old->data, oldsize,
352 cur->data, size, &delta_size, max_size);
c323ac7d 353 if (!delta_buf)
75c42d8c
LT
354 return 0;
355 cur_entry->delta = old_entry;
356 cur_entry->delta_size = delta_size;
d116a45a 357 cur_entry->depth = old_entry->depth + 1;
c323ac7d 358 free(delta_buf);
eb41ab11 359 return 0;
c323ac7d
LT
360}
361
d116a45a 362static void find_deltas(struct object_entry **list, int window, int depth)
c323ac7d 363{
521a4f4c 364 int i, idx;
c323ac7d
LT
365 unsigned int array_size = window * sizeof(struct unpacked);
366 struct unpacked *array = xmalloc(array_size);
21fcd1bd 367 int eye_candy;
c323ac7d
LT
368
369 memset(array, 0, array_size);
521a4f4c
LT
370 i = nr_objects;
371 idx = 0;
21fcd1bd
JH
372 eye_candy = i - (nr_objects / 20);
373
521a4f4c 374 while (--i >= 0) {
c323ac7d
LT
375 struct object_entry *entry = list[i];
376 struct unpacked *n = array + idx;
377 unsigned long size;
378 char type[10];
379 int j;
380
21fcd1bd
JH
381 if (progress && i <= eye_candy) {
382 eye_candy -= nr_objects / 20;
383 fputc('.', stderr);
384 }
c323ac7d
LT
385 free(n->data);
386 n->entry = entry;
387 n->data = read_sha1_file(entry->sha1, type, &size);
388 if (size != entry->size)
389 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
78817c15
LT
390 j = window;
391 while (--j > 0) {
392 unsigned int other_idx = idx + j;
c323ac7d 393 struct unpacked *m;
78817c15
LT
394 if (other_idx >= window)
395 other_idx -= window;
c323ac7d
LT
396 m = array + other_idx;
397 if (!m->entry)
398 break;
d116a45a 399 if (try_delta(n, m, depth) < 0)
c323ac7d
LT
400 break;
401 }
521a4f4c
LT
402 idx++;
403 if (idx >= window)
404 idx = 0;
c323ac7d 405 }
adee7bdf
SV
406
407 for (i = 0; i < window; ++i)
408 free(array[i].data);
409 free(array);
c323ac7d
LT
410}
411
f3123c4a
JH
412static void prepare_pack(int window, int depth)
413{
414 get_object_details();
415
21fcd1bd
JH
416 if (progress)
417 fprintf(stderr, "Packing %d objects", nr_objects);
f3123c4a
JH
418 sorted_by_type = create_sorted_list(type_size_sort);
419 if (window && depth)
420 find_deltas(sorted_by_type, window+1, depth);
21fcd1bd
JH
421 if (progress)
422 fputc('\n', stderr);
f3123c4a
JH
423 write_pack_file();
424}
425
426static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
427{
428 static const char cache[] = "pack-cache/pack-%s.%s";
429 char *cached_pack, *cached_idx;
430 int ifd, ofd, ifd_ix = -1;
431
432 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
433 ifd = open(cached_pack, O_RDONLY);
434 if (ifd < 0)
435 return 0;
436
437 if (!pack_to_stdout) {
438 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
439 ifd_ix = open(cached_idx, O_RDONLY);
440 if (ifd_ix < 0) {
441 close(ifd);
442 return 0;
443 }
444 }
445
446 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
447 sha1_to_hex(sha1));
448
449 if (pack_to_stdout) {
450 if (copy_fd(ifd, 1))
451 exit(1);
452 close(ifd);
453 }
454 else {
455 char name[PATH_MAX];
456 snprintf(name, sizeof(name),
457 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
458 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
459 if (ofd < 0)
460 die("unable to open %s (%s)", name, strerror(errno));
461 if (copy_fd(ifd, ofd))
462 exit(1);
463 close(ifd);
464
465 snprintf(name, sizeof(name),
466 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
467 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
468 if (ofd < 0)
469 die("unable to open %s (%s)", name, strerror(errno));
470 if (copy_fd(ifd_ix, ofd))
471 exit(1);
472 close(ifd_ix);
473 puts(sha1_to_hex(sha1));
474 }
475
476 return 1;
477}
478
c323ac7d
LT
479int main(int argc, char **argv)
480{
5f3de58f 481 SHA_CTX ctx;
27225f2e 482 char line[PATH_MAX + 20];
d22b9290 483 int window = 10, depth = 10, pack_to_stdout = 0;
84c8d8ae 484 struct object_entry **list;
c323ac7d 485 int i;
21fcd1bd
JH
486 struct timeval prev_tv;
487 int eye_candy = 0;
488 int eye_candy_incr = 500;
489
c323ac7d 490
53228a5f
JH
491 setup_git_directory();
492
c323ac7d
LT
493 for (i = 1; i < argc; i++) {
494 const char *arg = argv[i];
495
496 if (*arg == '-') {
1c4a2912
LT
497 if (!strcmp("--non-empty", arg)) {
498 non_empty = 1;
499 continue;
500 }
64560374
LT
501 if (!strcmp("--local", arg)) {
502 local = 1;
503 continue;
504 }
eb019375
LT
505 if (!strcmp("--incremental", arg)) {
506 incremental = 1;
507 continue;
508 }
c323ac7d
LT
509 if (!strncmp("--window=", arg, 9)) {
510 char *end;
f846bbff 511 window = strtoul(arg+9, &end, 0);
c323ac7d
LT
512 if (!arg[9] || *end)
513 usage(pack_usage);
514 continue;
515 }
d116a45a
LT
516 if (!strncmp("--depth=", arg, 8)) {
517 char *end;
518 depth = strtoul(arg+8, &end, 0);
519 if (!arg[8] || *end)
520 usage(pack_usage);
521 continue;
522 }
d22b9290
LT
523 if (!strcmp("--stdout", arg)) {
524 pack_to_stdout = 1;
525 continue;
526 }
c323ac7d
LT
527 usage(pack_usage);
528 }
529 if (base_name)
530 usage(pack_usage);
531 base_name = arg;
532 }
533
d22b9290 534 if (pack_to_stdout != !base_name)
c323ac7d
LT
535 usage(pack_usage);
536
21fcd1bd
JH
537 progress = isatty(2);
538
64560374 539 prepare_packed_git();
21fcd1bd
JH
540 if (progress) {
541 fprintf(stderr, "Generating pack...\n");
542 gettimeofday(&prev_tv, NULL);
543 }
c323ac7d 544 while (fgets(line, sizeof(line), stdin) != NULL) {
27225f2e
LT
545 unsigned int hash;
546 char *p;
c323ac7d 547 unsigned char sha1[20];
27225f2e 548
21fcd1bd
JH
549 if (progress && (eye_candy <= nr_objects)) {
550 fprintf(stderr, "Counting objects...%d\r", nr_objects);
551 if (eye_candy && (50 <= eye_candy_incr)) {
552 struct timeval tv;
553 int time_diff;
554 gettimeofday(&tv, NULL);
555 time_diff = (tv.tv_sec - prev_tv.tv_sec);
556 time_diff <<= 10;
557 time_diff += (tv.tv_usec - prev_tv.tv_usec);
558 if ((1 << 9) < time_diff)
559 eye_candy_incr += 50;
560 else if (50 < eye_candy_incr)
561 eye_candy_incr -= 50;
562 }
563 eye_candy += eye_candy_incr;
564 }
c323ac7d 565 if (get_sha1_hex(line, sha1))
ef07618f 566 die("expected sha1, got garbage:\n %s", line);
27225f2e
LT
567 hash = 0;
568 p = line+40;
569 while (*p) {
570 unsigned char c = *p++;
571 if (isspace(c))
572 continue;
573 hash = hash * 11 + c;
574 }
84c8d8ae 575 add_object_entry(sha1, hash);
c323ac7d 576 }
21fcd1bd
JH
577 if (progress)
578 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1c4a2912
LT
579 if (non_empty && !nr_objects)
580 return 0;
c323ac7d
LT
581
582 sorted_by_sha = create_sorted_list(sha1_sort);
84c8d8ae
JH
583 SHA1_Init(&ctx);
584 list = sorted_by_sha;
585 for (i = 0; i < nr_objects; i++) {
586 struct object_entry *entry = *list++;
587 SHA1_Update(&ctx, entry->sha1, 20);
588 }
589 SHA1_Final(object_list_sha1, &ctx);
590
f3123c4a
JH
591 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
592 ;
593 else {
594 prepare_pack(window, depth);
595 if (!pack_to_stdout) {
596 write_index_file();
597 puts(sha1_to_hex(object_list_sha1));
598 }
5f3de58f 599 }
c323ac7d
LT
600 return 0;
601}