d39193453a6bf868f8b62dac18201b0b4b7bccbd
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "pack-objects.h"
18 #include "progress.h"
19 #include "refs.h"
20 #include "streaming.h"
21 #include "thread-utils.h"
22 #include "pack-bitmap.h"
23
24 static const char *pack_usage[] = {
25 N_("git pack-objects --stdout [options...] [< ref-list | < object-list]"),
26 N_("git pack-objects [options...] base-name [< ref-list | < object-list]"),
27 NULL
28 };
29
30 /*
31 * Objects we are going to pack are collected in the `to_pack` structure.
32 * It contains an array (dynamically expanded) of the object data, and a map
33 * that can resolve SHA1s to their position in the array.
34 */
35 static struct packing_data to_pack;
36
37 static struct pack_idx_entry **written_list;
38 static uint32_t nr_result, nr_written;
39
40 static int non_empty;
41 static int reuse_delta = 1, reuse_object = 1;
42 static int keep_unreachable, unpack_unreachable, include_tag;
43 static unsigned long unpack_unreachable_expiration;
44 static int local;
45 static int incremental;
46 static int ignore_packed_keep;
47 static int allow_ofs_delta;
48 static struct pack_idx_option pack_idx_opts;
49 static const char *base_name;
50 static int progress = 1;
51 static int window = 10;
52 static unsigned long pack_size_limit;
53 static int depth = 50;
54 static int delta_search_threads;
55 static int pack_to_stdout;
56 static int num_preferred_base;
57 static struct progress *progress_state;
58 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
59 static int pack_compression_seen;
60
61 static struct packed_git *reuse_packfile;
62 static uint32_t reuse_packfile_objects;
63 static off_t reuse_packfile_offset;
64
65 static int use_bitmap_index = 1;
66 static int write_bitmap_index;
67 static uint16_t write_bitmap_options;
68
69 static unsigned long delta_cache_size = 0;
70 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
71 static unsigned long cache_max_small_delta_size = 1000;
72
73 static unsigned long window_memory_limit = 0;
74
75 /*
76 * stats
77 */
78 static uint32_t written, written_delta;
79 static uint32_t reused, reused_delta;
80
81 /*
82 * Indexed commits
83 */
84 static struct commit **indexed_commits;
85 static unsigned int indexed_commits_nr;
86 static unsigned int indexed_commits_alloc;
87
88 static void index_commit_for_bitmap(struct commit *commit)
89 {
90 if (indexed_commits_nr >= indexed_commits_alloc) {
91 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
92 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
93 }
94
95 indexed_commits[indexed_commits_nr++] = commit;
96 }
97
98 static void *get_delta(struct object_entry *entry)
99 {
100 unsigned long size, base_size, delta_size;
101 void *buf, *base_buf, *delta_buf;
102 enum object_type type;
103
104 buf = read_sha1_file(entry->idx.sha1, &type, &size);
105 if (!buf)
106 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
107 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
108 if (!base_buf)
109 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
110 delta_buf = diff_delta(base_buf, base_size,
111 buf, size, &delta_size, 0);
112 if (!delta_buf || delta_size != entry->delta_size)
113 die("delta size changed");
114 free(buf);
115 free(base_buf);
116 return delta_buf;
117 }
118
119 static unsigned long do_compress(void **pptr, unsigned long size)
120 {
121 git_zstream stream;
122 void *in, *out;
123 unsigned long maxsize;
124
125 memset(&stream, 0, sizeof(stream));
126 git_deflate_init(&stream, pack_compression_level);
127 maxsize = git_deflate_bound(&stream, size);
128
129 in = *pptr;
130 out = xmalloc(maxsize);
131 *pptr = out;
132
133 stream.next_in = in;
134 stream.avail_in = size;
135 stream.next_out = out;
136 stream.avail_out = maxsize;
137 while (git_deflate(&stream, Z_FINISH) == Z_OK)
138 ; /* nothing */
139 git_deflate_end(&stream);
140
141 free(in);
142 return stream.total_out;
143 }
144
145 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
146 const unsigned char *sha1)
147 {
148 git_zstream stream;
149 unsigned char ibuf[1024 * 16];
150 unsigned char obuf[1024 * 16];
151 unsigned long olen = 0;
152
153 memset(&stream, 0, sizeof(stream));
154 git_deflate_init(&stream, pack_compression_level);
155
156 for (;;) {
157 ssize_t readlen;
158 int zret = Z_OK;
159 readlen = read_istream(st, ibuf, sizeof(ibuf));
160 if (readlen == -1)
161 die(_("unable to read %s"), sha1_to_hex(sha1));
162
163 stream.next_in = ibuf;
164 stream.avail_in = readlen;
165 while ((stream.avail_in || readlen == 0) &&
166 (zret == Z_OK || zret == Z_BUF_ERROR)) {
167 stream.next_out = obuf;
168 stream.avail_out = sizeof(obuf);
169 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
170 sha1write(f, obuf, stream.next_out - obuf);
171 olen += stream.next_out - obuf;
172 }
173 if (stream.avail_in)
174 die(_("deflate error (%d)"), zret);
175 if (readlen == 0) {
176 if (zret != Z_STREAM_END)
177 die(_("deflate error (%d)"), zret);
178 break;
179 }
180 }
181 git_deflate_end(&stream);
182 return olen;
183 }
184
185 /*
186 * we are going to reuse the existing object data as is. make
187 * sure it is not corrupt.
188 */
189 static int check_pack_inflate(struct packed_git *p,
190 struct pack_window **w_curs,
191 off_t offset,
192 off_t len,
193 unsigned long expect)
194 {
195 git_zstream stream;
196 unsigned char fakebuf[4096], *in;
197 int st;
198
199 memset(&stream, 0, sizeof(stream));
200 git_inflate_init(&stream);
201 do {
202 in = use_pack(p, w_curs, offset, &stream.avail_in);
203 stream.next_in = in;
204 stream.next_out = fakebuf;
205 stream.avail_out = sizeof(fakebuf);
206 st = git_inflate(&stream, Z_FINISH);
207 offset += stream.next_in - in;
208 } while (st == Z_OK || st == Z_BUF_ERROR);
209 git_inflate_end(&stream);
210 return (st == Z_STREAM_END &&
211 stream.total_out == expect &&
212 stream.total_in == len) ? 0 : -1;
213 }
214
215 static void copy_pack_data(struct sha1file *f,
216 struct packed_git *p,
217 struct pack_window **w_curs,
218 off_t offset,
219 off_t len)
220 {
221 unsigned char *in;
222 unsigned long avail;
223
224 while (len) {
225 in = use_pack(p, w_curs, offset, &avail);
226 if (avail > len)
227 avail = (unsigned long)len;
228 sha1write(f, in, avail);
229 offset += avail;
230 len -= avail;
231 }
232 }
233
234 /* Return 0 if we will bust the pack-size limit */
235 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
236 unsigned long limit, int usable_delta)
237 {
238 unsigned long size, datalen;
239 unsigned char header[10], dheader[10];
240 unsigned hdrlen;
241 enum object_type type;
242 void *buf;
243 struct git_istream *st = NULL;
244
245 if (!usable_delta) {
246 if (entry->type == OBJ_BLOB &&
247 entry->size > big_file_threshold &&
248 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
249 buf = NULL;
250 else {
251 buf = read_sha1_file(entry->idx.sha1, &type, &size);
252 if (!buf)
253 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
254 }
255 /*
256 * make sure no cached delta data remains from a
257 * previous attempt before a pack split occurred.
258 */
259 free(entry->delta_data);
260 entry->delta_data = NULL;
261 entry->z_delta_size = 0;
262 } else if (entry->delta_data) {
263 size = entry->delta_size;
264 buf = entry->delta_data;
265 entry->delta_data = NULL;
266 type = (allow_ofs_delta && entry->delta->idx.offset) ?
267 OBJ_OFS_DELTA : OBJ_REF_DELTA;
268 } else {
269 buf = get_delta(entry);
270 size = entry->delta_size;
271 type = (allow_ofs_delta && entry->delta->idx.offset) ?
272 OBJ_OFS_DELTA : OBJ_REF_DELTA;
273 }
274
275 if (st) /* large blob case, just assume we don't compress well */
276 datalen = size;
277 else if (entry->z_delta_size)
278 datalen = entry->z_delta_size;
279 else
280 datalen = do_compress(&buf, size);
281
282 /*
283 * The object header is a byte of 'type' followed by zero or
284 * more bytes of length.
285 */
286 hdrlen = encode_in_pack_object_header(type, size, header);
287
288 if (type == OBJ_OFS_DELTA) {
289 /*
290 * Deltas with relative base contain an additional
291 * encoding of the relative offset for the delta
292 * base from this object's position in the pack.
293 */
294 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
295 unsigned pos = sizeof(dheader) - 1;
296 dheader[pos] = ofs & 127;
297 while (ofs >>= 7)
298 dheader[--pos] = 128 | (--ofs & 127);
299 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
300 if (st)
301 close_istream(st);
302 free(buf);
303 return 0;
304 }
305 sha1write(f, header, hdrlen);
306 sha1write(f, dheader + pos, sizeof(dheader) - pos);
307 hdrlen += sizeof(dheader) - pos;
308 } else if (type == OBJ_REF_DELTA) {
309 /*
310 * Deltas with a base reference contain
311 * an additional 20 bytes for the base sha1.
312 */
313 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
314 if (st)
315 close_istream(st);
316 free(buf);
317 return 0;
318 }
319 sha1write(f, header, hdrlen);
320 sha1write(f, entry->delta->idx.sha1, 20);
321 hdrlen += 20;
322 } else {
323 if (limit && hdrlen + datalen + 20 >= limit) {
324 if (st)
325 close_istream(st);
326 free(buf);
327 return 0;
328 }
329 sha1write(f, header, hdrlen);
330 }
331 if (st) {
332 datalen = write_large_blob_data(st, f, entry->idx.sha1);
333 close_istream(st);
334 } else {
335 sha1write(f, buf, datalen);
336 free(buf);
337 }
338
339 return hdrlen + datalen;
340 }
341
342 /* Return 0 if we will bust the pack-size limit */
343 static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
344 unsigned long limit, int usable_delta)
345 {
346 struct packed_git *p = entry->in_pack;
347 struct pack_window *w_curs = NULL;
348 struct revindex_entry *revidx;
349 off_t offset;
350 enum object_type type = entry->type;
351 unsigned long datalen;
352 unsigned char header[10], dheader[10];
353 unsigned hdrlen;
354
355 if (entry->delta)
356 type = (allow_ofs_delta && entry->delta->idx.offset) ?
357 OBJ_OFS_DELTA : OBJ_REF_DELTA;
358 hdrlen = encode_in_pack_object_header(type, entry->size, header);
359
360 offset = entry->in_pack_offset;
361 revidx = find_pack_revindex(p, offset);
362 datalen = revidx[1].offset - offset;
363 if (!pack_to_stdout && p->index_version > 1 &&
364 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
365 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
366 unuse_pack(&w_curs);
367 return write_no_reuse_object(f, entry, limit, usable_delta);
368 }
369
370 offset += entry->in_pack_header_size;
371 datalen -= entry->in_pack_header_size;
372
373 if (!pack_to_stdout && p->index_version == 1 &&
374 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
375 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
376 unuse_pack(&w_curs);
377 return write_no_reuse_object(f, entry, limit, usable_delta);
378 }
379
380 if (type == OBJ_OFS_DELTA) {
381 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
382 unsigned pos = sizeof(dheader) - 1;
383 dheader[pos] = ofs & 127;
384 while (ofs >>= 7)
385 dheader[--pos] = 128 | (--ofs & 127);
386 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
387 unuse_pack(&w_curs);
388 return 0;
389 }
390 sha1write(f, header, hdrlen);
391 sha1write(f, dheader + pos, sizeof(dheader) - pos);
392 hdrlen += sizeof(dheader) - pos;
393 reused_delta++;
394 } else if (type == OBJ_REF_DELTA) {
395 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
396 unuse_pack(&w_curs);
397 return 0;
398 }
399 sha1write(f, header, hdrlen);
400 sha1write(f, entry->delta->idx.sha1, 20);
401 hdrlen += 20;
402 reused_delta++;
403 } else {
404 if (limit && hdrlen + datalen + 20 >= limit) {
405 unuse_pack(&w_curs);
406 return 0;
407 }
408 sha1write(f, header, hdrlen);
409 }
410 copy_pack_data(f, p, &w_curs, offset, datalen);
411 unuse_pack(&w_curs);
412 reused++;
413 return hdrlen + datalen;
414 }
415
416 /* Return 0 if we will bust the pack-size limit */
417 static unsigned long write_object(struct sha1file *f,
418 struct object_entry *entry,
419 off_t write_offset)
420 {
421 unsigned long limit, len;
422 int usable_delta, to_reuse;
423
424 if (!pack_to_stdout)
425 crc32_begin(f);
426
427 /* apply size limit if limited packsize and not first object */
428 if (!pack_size_limit || !nr_written)
429 limit = 0;
430 else if (pack_size_limit <= write_offset)
431 /*
432 * the earlier object did not fit the limit; avoid
433 * mistaking this with unlimited (i.e. limit = 0).
434 */
435 limit = 1;
436 else
437 limit = pack_size_limit - write_offset;
438
439 if (!entry->delta)
440 usable_delta = 0; /* no delta */
441 else if (!pack_size_limit)
442 usable_delta = 1; /* unlimited packfile */
443 else if (entry->delta->idx.offset == (off_t)-1)
444 usable_delta = 0; /* base was written to another pack */
445 else if (entry->delta->idx.offset)
446 usable_delta = 1; /* base already exists in this pack */
447 else
448 usable_delta = 0; /* base could end up in another pack */
449
450 if (!reuse_object)
451 to_reuse = 0; /* explicit */
452 else if (!entry->in_pack)
453 to_reuse = 0; /* can't reuse what we don't have */
454 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
455 /* check_object() decided it for us ... */
456 to_reuse = usable_delta;
457 /* ... but pack split may override that */
458 else if (entry->type != entry->in_pack_type)
459 to_reuse = 0; /* pack has delta which is unusable */
460 else if (entry->delta)
461 to_reuse = 0; /* we want to pack afresh */
462 else
463 to_reuse = 1; /* we have it in-pack undeltified,
464 * and we do not need to deltify it.
465 */
466
467 if (!to_reuse)
468 len = write_no_reuse_object(f, entry, limit, usable_delta);
469 else
470 len = write_reuse_object(f, entry, limit, usable_delta);
471 if (!len)
472 return 0;
473
474 if (usable_delta)
475 written_delta++;
476 written++;
477 if (!pack_to_stdout)
478 entry->idx.crc32 = crc32_end(f);
479 return len;
480 }
481
482 enum write_one_status {
483 WRITE_ONE_SKIP = -1, /* already written */
484 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
485 WRITE_ONE_WRITTEN = 1, /* normal */
486 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
487 };
488
489 static enum write_one_status write_one(struct sha1file *f,
490 struct object_entry *e,
491 off_t *offset)
492 {
493 unsigned long size;
494 int recursing;
495
496 /*
497 * we set offset to 1 (which is an impossible value) to mark
498 * the fact that this object is involved in "write its base
499 * first before writing a deltified object" recursion.
500 */
501 recursing = (e->idx.offset == 1);
502 if (recursing) {
503 warning("recursive delta detected for object %s",
504 sha1_to_hex(e->idx.sha1));
505 return WRITE_ONE_RECURSIVE;
506 } else if (e->idx.offset || e->preferred_base) {
507 /* offset is non zero if object is written already. */
508 return WRITE_ONE_SKIP;
509 }
510
511 /* if we are deltified, write out base object first. */
512 if (e->delta) {
513 e->idx.offset = 1; /* now recurse */
514 switch (write_one(f, e->delta, offset)) {
515 case WRITE_ONE_RECURSIVE:
516 /* we cannot depend on this one */
517 e->delta = NULL;
518 break;
519 default:
520 break;
521 case WRITE_ONE_BREAK:
522 e->idx.offset = recursing;
523 return WRITE_ONE_BREAK;
524 }
525 }
526
527 e->idx.offset = *offset;
528 size = write_object(f, e, *offset);
529 if (!size) {
530 e->idx.offset = recursing;
531 return WRITE_ONE_BREAK;
532 }
533 written_list[nr_written++] = &e->idx;
534
535 /* make sure off_t is sufficiently large not to wrap */
536 if (signed_add_overflows(*offset, size))
537 die("pack too large for current definition of off_t");
538 *offset += size;
539 return WRITE_ONE_WRITTEN;
540 }
541
542 static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
543 void *cb_data)
544 {
545 unsigned char peeled[20];
546 struct object_entry *entry = packlist_find(&to_pack, sha1, NULL);
547
548 if (entry)
549 entry->tagged = 1;
550 if (!peel_ref(path, peeled)) {
551 entry = packlist_find(&to_pack, peeled, NULL);
552 if (entry)
553 entry->tagged = 1;
554 }
555 return 0;
556 }
557
558 static inline void add_to_write_order(struct object_entry **wo,
559 unsigned int *endp,
560 struct object_entry *e)
561 {
562 if (e->filled)
563 return;
564 wo[(*endp)++] = e;
565 e->filled = 1;
566 }
567
568 static void add_descendants_to_write_order(struct object_entry **wo,
569 unsigned int *endp,
570 struct object_entry *e)
571 {
572 int add_to_order = 1;
573 while (e) {
574 if (add_to_order) {
575 struct object_entry *s;
576 /* add this node... */
577 add_to_write_order(wo, endp, e);
578 /* all its siblings... */
579 for (s = e->delta_sibling; s; s = s->delta_sibling) {
580 add_to_write_order(wo, endp, s);
581 }
582 }
583 /* drop down a level to add left subtree nodes if possible */
584 if (e->delta_child) {
585 add_to_order = 1;
586 e = e->delta_child;
587 } else {
588 add_to_order = 0;
589 /* our sibling might have some children, it is next */
590 if (e->delta_sibling) {
591 e = e->delta_sibling;
592 continue;
593 }
594 /* go back to our parent node */
595 e = e->delta;
596 while (e && !e->delta_sibling) {
597 /* we're on the right side of a subtree, keep
598 * going up until we can go right again */
599 e = e->delta;
600 }
601 if (!e) {
602 /* done- we hit our original root node */
603 return;
604 }
605 /* pass it off to sibling at this level */
606 e = e->delta_sibling;
607 }
608 };
609 }
610
611 static void add_family_to_write_order(struct object_entry **wo,
612 unsigned int *endp,
613 struct object_entry *e)
614 {
615 struct object_entry *root;
616
617 for (root = e; root->delta; root = root->delta)
618 ; /* nothing */
619 add_descendants_to_write_order(wo, endp, root);
620 }
621
622 static struct object_entry **compute_write_order(void)
623 {
624 unsigned int i, wo_end, last_untagged;
625
626 struct object_entry **wo = xmalloc(to_pack.nr_objects * sizeof(*wo));
627 struct object_entry *objects = to_pack.objects;
628
629 for (i = 0; i < to_pack.nr_objects; i++) {
630 objects[i].tagged = 0;
631 objects[i].filled = 0;
632 objects[i].delta_child = NULL;
633 objects[i].delta_sibling = NULL;
634 }
635
636 /*
637 * Fully connect delta_child/delta_sibling network.
638 * Make sure delta_sibling is sorted in the original
639 * recency order.
640 */
641 for (i = to_pack.nr_objects; i > 0;) {
642 struct object_entry *e = &objects[--i];
643 if (!e->delta)
644 continue;
645 /* Mark me as the first child */
646 e->delta_sibling = e->delta->delta_child;
647 e->delta->delta_child = e;
648 }
649
650 /*
651 * Mark objects that are at the tip of tags.
652 */
653 for_each_tag_ref(mark_tagged, NULL);
654
655 /*
656 * Give the objects in the original recency order until
657 * we see a tagged tip.
658 */
659 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
660 if (objects[i].tagged)
661 break;
662 add_to_write_order(wo, &wo_end, &objects[i]);
663 }
664 last_untagged = i;
665
666 /*
667 * Then fill all the tagged tips.
668 */
669 for (; i < to_pack.nr_objects; i++) {
670 if (objects[i].tagged)
671 add_to_write_order(wo, &wo_end, &objects[i]);
672 }
673
674 /*
675 * And then all remaining commits and tags.
676 */
677 for (i = last_untagged; i < to_pack.nr_objects; i++) {
678 if (objects[i].type != OBJ_COMMIT &&
679 objects[i].type != OBJ_TAG)
680 continue;
681 add_to_write_order(wo, &wo_end, &objects[i]);
682 }
683
684 /*
685 * And then all the trees.
686 */
687 for (i = last_untagged; i < to_pack.nr_objects; i++) {
688 if (objects[i].type != OBJ_TREE)
689 continue;
690 add_to_write_order(wo, &wo_end, &objects[i]);
691 }
692
693 /*
694 * Finally all the rest in really tight order
695 */
696 for (i = last_untagged; i < to_pack.nr_objects; i++) {
697 if (!objects[i].filled)
698 add_family_to_write_order(wo, &wo_end, &objects[i]);
699 }
700
701 if (wo_end != to_pack.nr_objects)
702 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
703
704 return wo;
705 }
706
707 static off_t write_reused_pack(struct sha1file *f)
708 {
709 unsigned char buffer[8192];
710 off_t to_write, total;
711 int fd;
712
713 if (!is_pack_valid(reuse_packfile))
714 die("packfile is invalid: %s", reuse_packfile->pack_name);
715
716 fd = git_open_noatime(reuse_packfile->pack_name);
717 if (fd < 0)
718 die_errno("unable to open packfile for reuse: %s",
719 reuse_packfile->pack_name);
720
721 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
722 die_errno("unable to seek in reused packfile");
723
724 if (reuse_packfile_offset < 0)
725 reuse_packfile_offset = reuse_packfile->pack_size - 20;
726
727 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
728
729 while (to_write) {
730 int read_pack = xread(fd, buffer, sizeof(buffer));
731
732 if (read_pack <= 0)
733 die_errno("unable to read from reused packfile");
734
735 if (read_pack > to_write)
736 read_pack = to_write;
737
738 sha1write(f, buffer, read_pack);
739 to_write -= read_pack;
740
741 /*
742 * We don't know the actual number of objects written,
743 * only how many bytes written, how many bytes total, and
744 * how many objects total. So we can fake it by pretending all
745 * objects we are writing are the same size. This gives us a
746 * smooth progress meter, and at the end it matches the true
747 * answer.
748 */
749 written = reuse_packfile_objects *
750 (((double)(total - to_write)) / total);
751 display_progress(progress_state, written);
752 }
753
754 close(fd);
755 written = reuse_packfile_objects;
756 display_progress(progress_state, written);
757 return reuse_packfile_offset - sizeof(struct pack_header);
758 }
759
760 static void write_pack_file(void)
761 {
762 uint32_t i = 0, j;
763 struct sha1file *f;
764 off_t offset;
765 uint32_t nr_remaining = nr_result;
766 time_t last_mtime = 0;
767 struct object_entry **write_order;
768
769 if (progress > pack_to_stdout)
770 progress_state = start_progress(_("Writing objects"), nr_result);
771 written_list = xmalloc(to_pack.nr_objects * sizeof(*written_list));
772 write_order = compute_write_order();
773
774 do {
775 unsigned char sha1[20];
776 char *pack_tmp_name = NULL;
777
778 if (pack_to_stdout)
779 f = sha1fd_throughput(1, "<stdout>", progress_state);
780 else
781 f = create_tmp_packfile(&pack_tmp_name);
782
783 offset = write_pack_header(f, nr_remaining);
784
785 if (reuse_packfile) {
786 off_t packfile_size;
787 assert(pack_to_stdout);
788
789 packfile_size = write_reused_pack(f);
790 offset += packfile_size;
791 }
792
793 nr_written = 0;
794 for (; i < to_pack.nr_objects; i++) {
795 struct object_entry *e = write_order[i];
796 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
797 break;
798 display_progress(progress_state, written);
799 }
800
801 /*
802 * Did we write the wrong # entries in the header?
803 * If so, rewrite it like in fast-import
804 */
805 if (pack_to_stdout) {
806 sha1close(f, sha1, CSUM_CLOSE);
807 } else if (nr_written == nr_remaining) {
808 sha1close(f, sha1, CSUM_FSYNC);
809 } else {
810 int fd = sha1close(f, sha1, 0);
811 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
812 nr_written, sha1, offset);
813 close(fd);
814 }
815
816 if (!pack_to_stdout) {
817 struct stat st;
818 struct strbuf tmpname = STRBUF_INIT;
819
820 /*
821 * Packs are runtime accessed in their mtime
822 * order since newer packs are more likely to contain
823 * younger objects. So if we are creating multiple
824 * packs then we should modify the mtime of later ones
825 * to preserve this property.
826 */
827 if (stat(pack_tmp_name, &st) < 0) {
828 warning("failed to stat %s: %s",
829 pack_tmp_name, strerror(errno));
830 } else if (!last_mtime) {
831 last_mtime = st.st_mtime;
832 } else {
833 struct utimbuf utb;
834 utb.actime = st.st_atime;
835 utb.modtime = --last_mtime;
836 if (utime(pack_tmp_name, &utb) < 0)
837 warning("failed utime() on %s: %s",
838 pack_tmp_name, strerror(errno));
839 }
840
841 strbuf_addf(&tmpname, "%s-", base_name);
842
843 if (write_bitmap_index) {
844 bitmap_writer_set_checksum(sha1);
845 bitmap_writer_build_type_index(written_list, nr_written);
846 }
847
848 finish_tmp_packfile(&tmpname, pack_tmp_name,
849 written_list, nr_written,
850 &pack_idx_opts, sha1);
851
852 if (write_bitmap_index) {
853 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
854
855 stop_progress(&progress_state);
856
857 bitmap_writer_show_progress(progress);
858 bitmap_writer_reuse_bitmaps(&to_pack);
859 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
860 bitmap_writer_build(&to_pack);
861 bitmap_writer_finish(written_list, nr_written,
862 tmpname.buf, write_bitmap_options);
863 write_bitmap_index = 0;
864 }
865
866 strbuf_release(&tmpname);
867 free(pack_tmp_name);
868 puts(sha1_to_hex(sha1));
869 }
870
871 /* mark written objects as written to previous pack */
872 for (j = 0; j < nr_written; j++) {
873 written_list[j]->offset = (off_t)-1;
874 }
875 nr_remaining -= nr_written;
876 } while (nr_remaining && i < to_pack.nr_objects);
877
878 free(written_list);
879 free(write_order);
880 stop_progress(&progress_state);
881 if (written != nr_result)
882 die("wrote %"PRIu32" objects while expecting %"PRIu32,
883 written, nr_result);
884 }
885
886 static void setup_delta_attr_check(struct git_attr_check *check)
887 {
888 static struct git_attr *attr_delta;
889
890 if (!attr_delta)
891 attr_delta = git_attr("delta");
892
893 check[0].attr = attr_delta;
894 }
895
896 static int no_try_delta(const char *path)
897 {
898 struct git_attr_check check[1];
899
900 setup_delta_attr_check(check);
901 if (git_check_attr(path, ARRAY_SIZE(check), check))
902 return 0;
903 if (ATTR_FALSE(check->value))
904 return 1;
905 return 0;
906 }
907
908 /*
909 * When adding an object, check whether we have already added it
910 * to our packing list. If so, we can skip. However, if we are
911 * being asked to excludei t, but the previous mention was to include
912 * it, make sure to adjust its flags and tweak our numbers accordingly.
913 *
914 * As an optimization, we pass out the index position where we would have
915 * found the item, since that saves us from having to look it up again a
916 * few lines later when we want to add the new entry.
917 */
918 static int have_duplicate_entry(const unsigned char *sha1,
919 int exclude,
920 uint32_t *index_pos)
921 {
922 struct object_entry *entry;
923
924 entry = packlist_find(&to_pack, sha1, index_pos);
925 if (!entry)
926 return 0;
927
928 if (exclude) {
929 if (!entry->preferred_base)
930 nr_result--;
931 entry->preferred_base = 1;
932 }
933
934 return 1;
935 }
936
937 /*
938 * Check whether we want the object in the pack (e.g., we do not want
939 * objects found in non-local stores if the "--local" option was used).
940 *
941 * As a side effect of this check, we will find the packed version of this
942 * object, if any. We therefore pass out the pack information to avoid having
943 * to look it up again later.
944 */
945 static int want_object_in_pack(const unsigned char *sha1,
946 int exclude,
947 struct packed_git **found_pack,
948 off_t *found_offset)
949 {
950 struct packed_git *p;
951
952 if (!exclude && local && has_loose_object_nonlocal(sha1))
953 return 0;
954
955 *found_pack = NULL;
956 *found_offset = 0;
957
958 for (p = packed_git; p; p = p->next) {
959 off_t offset = find_pack_entry_one(sha1, p);
960 if (offset) {
961 if (!*found_pack) {
962 if (!is_pack_valid(p)) {
963 warning("packfile %s cannot be accessed", p->pack_name);
964 continue;
965 }
966 *found_offset = offset;
967 *found_pack = p;
968 }
969 if (exclude)
970 return 1;
971 if (incremental)
972 return 0;
973 if (local && !p->pack_local)
974 return 0;
975 if (ignore_packed_keep && p->pack_local && p->pack_keep)
976 return 0;
977 }
978 }
979
980 return 1;
981 }
982
983 static void create_object_entry(const unsigned char *sha1,
984 enum object_type type,
985 uint32_t hash,
986 int exclude,
987 int no_try_delta,
988 uint32_t index_pos,
989 struct packed_git *found_pack,
990 off_t found_offset)
991 {
992 struct object_entry *entry;
993
994 entry = packlist_alloc(&to_pack, sha1, index_pos);
995 entry->hash = hash;
996 if (type)
997 entry->type = type;
998 if (exclude)
999 entry->preferred_base = 1;
1000 else
1001 nr_result++;
1002 if (found_pack) {
1003 entry->in_pack = found_pack;
1004 entry->in_pack_offset = found_offset;
1005 }
1006
1007 entry->no_try_delta = no_try_delta;
1008 }
1009
1010 static const char no_closure_warning[] = N_(
1011 "disabling bitmap writing, as some objects are not being packed"
1012 );
1013
1014 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1015 const char *name, int exclude)
1016 {
1017 struct packed_git *found_pack;
1018 off_t found_offset;
1019 uint32_t index_pos;
1020
1021 if (have_duplicate_entry(sha1, exclude, &index_pos))
1022 return 0;
1023
1024 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1025 /* The pack is missing an object, so it will not have closure */
1026 if (write_bitmap_index) {
1027 warning(_(no_closure_warning));
1028 write_bitmap_index = 0;
1029 }
1030 return 0;
1031 }
1032
1033 create_object_entry(sha1, type, pack_name_hash(name),
1034 exclude, name && no_try_delta(name),
1035 index_pos, found_pack, found_offset);
1036
1037 display_progress(progress_state, nr_result);
1038 return 1;
1039 }
1040
1041 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1042 enum object_type type,
1043 int flags, uint32_t name_hash,
1044 struct packed_git *pack, off_t offset)
1045 {
1046 uint32_t index_pos;
1047
1048 if (have_duplicate_entry(sha1, 0, &index_pos))
1049 return 0;
1050
1051 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1052
1053 display_progress(progress_state, nr_result);
1054 return 1;
1055 }
1056
1057 struct pbase_tree_cache {
1058 unsigned char sha1[20];
1059 int ref;
1060 int temporary;
1061 void *tree_data;
1062 unsigned long tree_size;
1063 };
1064
1065 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1066 static int pbase_tree_cache_ix(const unsigned char *sha1)
1067 {
1068 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1069 }
1070 static int pbase_tree_cache_ix_incr(int ix)
1071 {
1072 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1073 }
1074
1075 static struct pbase_tree {
1076 struct pbase_tree *next;
1077 /* This is a phony "cache" entry; we are not
1078 * going to evict it or find it through _get()
1079 * mechanism -- this is for the toplevel node that
1080 * would almost always change with any commit.
1081 */
1082 struct pbase_tree_cache pcache;
1083 } *pbase_tree;
1084
1085 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1086 {
1087 struct pbase_tree_cache *ent, *nent;
1088 void *data;
1089 unsigned long size;
1090 enum object_type type;
1091 int neigh;
1092 int my_ix = pbase_tree_cache_ix(sha1);
1093 int available_ix = -1;
1094
1095 /* pbase-tree-cache acts as a limited hashtable.
1096 * your object will be found at your index or within a few
1097 * slots after that slot if it is cached.
1098 */
1099 for (neigh = 0; neigh < 8; neigh++) {
1100 ent = pbase_tree_cache[my_ix];
1101 if (ent && !hashcmp(ent->sha1, sha1)) {
1102 ent->ref++;
1103 return ent;
1104 }
1105 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1106 ((0 <= available_ix) &&
1107 (!ent && pbase_tree_cache[available_ix])))
1108 available_ix = my_ix;
1109 if (!ent)
1110 break;
1111 my_ix = pbase_tree_cache_ix_incr(my_ix);
1112 }
1113
1114 /* Did not find one. Either we got a bogus request or
1115 * we need to read and perhaps cache.
1116 */
1117 data = read_sha1_file(sha1, &type, &size);
1118 if (!data)
1119 return NULL;
1120 if (type != OBJ_TREE) {
1121 free(data);
1122 return NULL;
1123 }
1124
1125 /* We need to either cache or return a throwaway copy */
1126
1127 if (available_ix < 0)
1128 ent = NULL;
1129 else {
1130 ent = pbase_tree_cache[available_ix];
1131 my_ix = available_ix;
1132 }
1133
1134 if (!ent) {
1135 nent = xmalloc(sizeof(*nent));
1136 nent->temporary = (available_ix < 0);
1137 }
1138 else {
1139 /* evict and reuse */
1140 free(ent->tree_data);
1141 nent = ent;
1142 }
1143 hashcpy(nent->sha1, sha1);
1144 nent->tree_data = data;
1145 nent->tree_size = size;
1146 nent->ref = 1;
1147 if (!nent->temporary)
1148 pbase_tree_cache[my_ix] = nent;
1149 return nent;
1150 }
1151
1152 static void pbase_tree_put(struct pbase_tree_cache *cache)
1153 {
1154 if (!cache->temporary) {
1155 cache->ref--;
1156 return;
1157 }
1158 free(cache->tree_data);
1159 free(cache);
1160 }
1161
1162 static int name_cmp_len(const char *name)
1163 {
1164 int i;
1165 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1166 ;
1167 return i;
1168 }
1169
1170 static void add_pbase_object(struct tree_desc *tree,
1171 const char *name,
1172 int cmplen,
1173 const char *fullname)
1174 {
1175 struct name_entry entry;
1176 int cmp;
1177
1178 while (tree_entry(tree,&entry)) {
1179 if (S_ISGITLINK(entry.mode))
1180 continue;
1181 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1182 memcmp(name, entry.path, cmplen);
1183 if (cmp > 0)
1184 continue;
1185 if (cmp < 0)
1186 return;
1187 if (name[cmplen] != '/') {
1188 add_object_entry(entry.sha1,
1189 object_type(entry.mode),
1190 fullname, 1);
1191 return;
1192 }
1193 if (S_ISDIR(entry.mode)) {
1194 struct tree_desc sub;
1195 struct pbase_tree_cache *tree;
1196 const char *down = name+cmplen+1;
1197 int downlen = name_cmp_len(down);
1198
1199 tree = pbase_tree_get(entry.sha1);
1200 if (!tree)
1201 return;
1202 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1203
1204 add_pbase_object(&sub, down, downlen, fullname);
1205 pbase_tree_put(tree);
1206 }
1207 }
1208 }
1209
1210 static unsigned *done_pbase_paths;
1211 static int done_pbase_paths_num;
1212 static int done_pbase_paths_alloc;
1213 static int done_pbase_path_pos(unsigned hash)
1214 {
1215 int lo = 0;
1216 int hi = done_pbase_paths_num;
1217 while (lo < hi) {
1218 int mi = (hi + lo) / 2;
1219 if (done_pbase_paths[mi] == hash)
1220 return mi;
1221 if (done_pbase_paths[mi] < hash)
1222 hi = mi;
1223 else
1224 lo = mi + 1;
1225 }
1226 return -lo-1;
1227 }
1228
1229 static int check_pbase_path(unsigned hash)
1230 {
1231 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1232 if (0 <= pos)
1233 return 1;
1234 pos = -pos - 1;
1235 ALLOC_GROW(done_pbase_paths,
1236 done_pbase_paths_num + 1,
1237 done_pbase_paths_alloc);
1238 done_pbase_paths_num++;
1239 if (pos < done_pbase_paths_num)
1240 memmove(done_pbase_paths + pos + 1,
1241 done_pbase_paths + pos,
1242 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1243 done_pbase_paths[pos] = hash;
1244 return 0;
1245 }
1246
1247 static void add_preferred_base_object(const char *name)
1248 {
1249 struct pbase_tree *it;
1250 int cmplen;
1251 unsigned hash = pack_name_hash(name);
1252
1253 if (!num_preferred_base || check_pbase_path(hash))
1254 return;
1255
1256 cmplen = name_cmp_len(name);
1257 for (it = pbase_tree; it; it = it->next) {
1258 if (cmplen == 0) {
1259 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1260 }
1261 else {
1262 struct tree_desc tree;
1263 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1264 add_pbase_object(&tree, name, cmplen, name);
1265 }
1266 }
1267 }
1268
1269 static void add_preferred_base(unsigned char *sha1)
1270 {
1271 struct pbase_tree *it;
1272 void *data;
1273 unsigned long size;
1274 unsigned char tree_sha1[20];
1275
1276 if (window <= num_preferred_base++)
1277 return;
1278
1279 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1280 if (!data)
1281 return;
1282
1283 for (it = pbase_tree; it; it = it->next) {
1284 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1285 free(data);
1286 return;
1287 }
1288 }
1289
1290 it = xcalloc(1, sizeof(*it));
1291 it->next = pbase_tree;
1292 pbase_tree = it;
1293
1294 hashcpy(it->pcache.sha1, tree_sha1);
1295 it->pcache.tree_data = data;
1296 it->pcache.tree_size = size;
1297 }
1298
1299 static void cleanup_preferred_base(void)
1300 {
1301 struct pbase_tree *it;
1302 unsigned i;
1303
1304 it = pbase_tree;
1305 pbase_tree = NULL;
1306 while (it) {
1307 struct pbase_tree *this = it;
1308 it = this->next;
1309 free(this->pcache.tree_data);
1310 free(this);
1311 }
1312
1313 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1314 if (!pbase_tree_cache[i])
1315 continue;
1316 free(pbase_tree_cache[i]->tree_data);
1317 free(pbase_tree_cache[i]);
1318 pbase_tree_cache[i] = NULL;
1319 }
1320
1321 free(done_pbase_paths);
1322 done_pbase_paths = NULL;
1323 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1324 }
1325
1326 static void check_object(struct object_entry *entry)
1327 {
1328 if (entry->in_pack) {
1329 struct packed_git *p = entry->in_pack;
1330 struct pack_window *w_curs = NULL;
1331 const unsigned char *base_ref = NULL;
1332 struct object_entry *base_entry;
1333 unsigned long used, used_0;
1334 unsigned long avail;
1335 off_t ofs;
1336 unsigned char *buf, c;
1337
1338 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1339
1340 /*
1341 * We want in_pack_type even if we do not reuse delta
1342 * since non-delta representations could still be reused.
1343 */
1344 used = unpack_object_header_buffer(buf, avail,
1345 &entry->in_pack_type,
1346 &entry->size);
1347 if (used == 0)
1348 goto give_up;
1349
1350 /*
1351 * Determine if this is a delta and if so whether we can
1352 * reuse it or not. Otherwise let's find out as cheaply as
1353 * possible what the actual type and size for this object is.
1354 */
1355 switch (entry->in_pack_type) {
1356 default:
1357 /* Not a delta hence we've already got all we need. */
1358 entry->type = entry->in_pack_type;
1359 entry->in_pack_header_size = used;
1360 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1361 goto give_up;
1362 unuse_pack(&w_curs);
1363 return;
1364 case OBJ_REF_DELTA:
1365 if (reuse_delta && !entry->preferred_base)
1366 base_ref = use_pack(p, &w_curs,
1367 entry->in_pack_offset + used, NULL);
1368 entry->in_pack_header_size = used + 20;
1369 break;
1370 case OBJ_OFS_DELTA:
1371 buf = use_pack(p, &w_curs,
1372 entry->in_pack_offset + used, NULL);
1373 used_0 = 0;
1374 c = buf[used_0++];
1375 ofs = c & 127;
1376 while (c & 128) {
1377 ofs += 1;
1378 if (!ofs || MSB(ofs, 7)) {
1379 error("delta base offset overflow in pack for %s",
1380 sha1_to_hex(entry->idx.sha1));
1381 goto give_up;
1382 }
1383 c = buf[used_0++];
1384 ofs = (ofs << 7) + (c & 127);
1385 }
1386 ofs = entry->in_pack_offset - ofs;
1387 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1388 error("delta base offset out of bound for %s",
1389 sha1_to_hex(entry->idx.sha1));
1390 goto give_up;
1391 }
1392 if (reuse_delta && !entry->preferred_base) {
1393 struct revindex_entry *revidx;
1394 revidx = find_pack_revindex(p, ofs);
1395 if (!revidx)
1396 goto give_up;
1397 base_ref = nth_packed_object_sha1(p, revidx->nr);
1398 }
1399 entry->in_pack_header_size = used + used_0;
1400 break;
1401 }
1402
1403 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1404 /*
1405 * If base_ref was set above that means we wish to
1406 * reuse delta data, and we even found that base
1407 * in the list of objects we want to pack. Goodie!
1408 *
1409 * Depth value does not matter - find_deltas() will
1410 * never consider reused delta as the base object to
1411 * deltify other objects against, in order to avoid
1412 * circular deltas.
1413 */
1414 entry->type = entry->in_pack_type;
1415 entry->delta = base_entry;
1416 entry->delta_size = entry->size;
1417 entry->delta_sibling = base_entry->delta_child;
1418 base_entry->delta_child = entry;
1419 unuse_pack(&w_curs);
1420 return;
1421 }
1422
1423 if (entry->type) {
1424 /*
1425 * This must be a delta and we already know what the
1426 * final object type is. Let's extract the actual
1427 * object size from the delta header.
1428 */
1429 entry->size = get_size_from_delta(p, &w_curs,
1430 entry->in_pack_offset + entry->in_pack_header_size);
1431 if (entry->size == 0)
1432 goto give_up;
1433 unuse_pack(&w_curs);
1434 return;
1435 }
1436
1437 /*
1438 * No choice but to fall back to the recursive delta walk
1439 * with sha1_object_info() to find about the object type
1440 * at this point...
1441 */
1442 give_up:
1443 unuse_pack(&w_curs);
1444 }
1445
1446 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1447 /*
1448 * The error condition is checked in prepare_pack(). This is
1449 * to permit a missing preferred base object to be ignored
1450 * as a preferred base. Doing so can result in a larger
1451 * pack file, but the transfer will still take place.
1452 */
1453 }
1454
1455 static int pack_offset_sort(const void *_a, const void *_b)
1456 {
1457 const struct object_entry *a = *(struct object_entry **)_a;
1458 const struct object_entry *b = *(struct object_entry **)_b;
1459
1460 /* avoid filesystem trashing with loose objects */
1461 if (!a->in_pack && !b->in_pack)
1462 return hashcmp(a->idx.sha1, b->idx.sha1);
1463
1464 if (a->in_pack < b->in_pack)
1465 return -1;
1466 if (a->in_pack > b->in_pack)
1467 return 1;
1468 return a->in_pack_offset < b->in_pack_offset ? -1 :
1469 (a->in_pack_offset > b->in_pack_offset);
1470 }
1471
1472 static void get_object_details(void)
1473 {
1474 uint32_t i;
1475 struct object_entry **sorted_by_offset;
1476
1477 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1478 for (i = 0; i < to_pack.nr_objects; i++)
1479 sorted_by_offset[i] = to_pack.objects + i;
1480 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1481
1482 for (i = 0; i < to_pack.nr_objects; i++) {
1483 struct object_entry *entry = sorted_by_offset[i];
1484 check_object(entry);
1485 if (big_file_threshold < entry->size)
1486 entry->no_try_delta = 1;
1487 }
1488
1489 free(sorted_by_offset);
1490 }
1491
1492 /*
1493 * We search for deltas in a list sorted by type, by filename hash, and then
1494 * by size, so that we see progressively smaller and smaller files.
1495 * That's because we prefer deltas to be from the bigger file
1496 * to the smaller -- deletes are potentially cheaper, but perhaps
1497 * more importantly, the bigger file is likely the more recent
1498 * one. The deepest deltas are therefore the oldest objects which are
1499 * less susceptible to be accessed often.
1500 */
1501 static int type_size_sort(const void *_a, const void *_b)
1502 {
1503 const struct object_entry *a = *(struct object_entry **)_a;
1504 const struct object_entry *b = *(struct object_entry **)_b;
1505
1506 if (a->type > b->type)
1507 return -1;
1508 if (a->type < b->type)
1509 return 1;
1510 if (a->hash > b->hash)
1511 return -1;
1512 if (a->hash < b->hash)
1513 return 1;
1514 if (a->preferred_base > b->preferred_base)
1515 return -1;
1516 if (a->preferred_base < b->preferred_base)
1517 return 1;
1518 if (a->size > b->size)
1519 return -1;
1520 if (a->size < b->size)
1521 return 1;
1522 return a < b ? -1 : (a > b); /* newest first */
1523 }
1524
1525 struct unpacked {
1526 struct object_entry *entry;
1527 void *data;
1528 struct delta_index *index;
1529 unsigned depth;
1530 };
1531
1532 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1533 unsigned long delta_size)
1534 {
1535 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1536 return 0;
1537
1538 if (delta_size < cache_max_small_delta_size)
1539 return 1;
1540
1541 /* cache delta, if objects are large enough compared to delta size */
1542 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1543 return 1;
1544
1545 return 0;
1546 }
1547
1548 #ifndef NO_PTHREADS
1549
1550 static pthread_mutex_t read_mutex;
1551 #define read_lock() pthread_mutex_lock(&read_mutex)
1552 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1553
1554 static pthread_mutex_t cache_mutex;
1555 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1556 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1557
1558 static pthread_mutex_t progress_mutex;
1559 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1560 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1561
1562 #else
1563
1564 #define read_lock() (void)0
1565 #define read_unlock() (void)0
1566 #define cache_lock() (void)0
1567 #define cache_unlock() (void)0
1568 #define progress_lock() (void)0
1569 #define progress_unlock() (void)0
1570
1571 #endif
1572
1573 static int try_delta(struct unpacked *trg, struct unpacked *src,
1574 unsigned max_depth, unsigned long *mem_usage)
1575 {
1576 struct object_entry *trg_entry = trg->entry;
1577 struct object_entry *src_entry = src->entry;
1578 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1579 unsigned ref_depth;
1580 enum object_type type;
1581 void *delta_buf;
1582
1583 /* Don't bother doing diffs between different types */
1584 if (trg_entry->type != src_entry->type)
1585 return -1;
1586
1587 /*
1588 * We do not bother to try a delta that we discarded on an
1589 * earlier try, but only when reusing delta data. Note that
1590 * src_entry that is marked as the preferred_base should always
1591 * be considered, as even if we produce a suboptimal delta against
1592 * it, we will still save the transfer cost, as we already know
1593 * the other side has it and we won't send src_entry at all.
1594 */
1595 if (reuse_delta && trg_entry->in_pack &&
1596 trg_entry->in_pack == src_entry->in_pack &&
1597 !src_entry->preferred_base &&
1598 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1599 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1600 return 0;
1601
1602 /* Let's not bust the allowed depth. */
1603 if (src->depth >= max_depth)
1604 return 0;
1605
1606 /* Now some size filtering heuristics. */
1607 trg_size = trg_entry->size;
1608 if (!trg_entry->delta) {
1609 max_size = trg_size/2 - 20;
1610 ref_depth = 1;
1611 } else {
1612 max_size = trg_entry->delta_size;
1613 ref_depth = trg->depth;
1614 }
1615 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1616 (max_depth - ref_depth + 1);
1617 if (max_size == 0)
1618 return 0;
1619 src_size = src_entry->size;
1620 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1621 if (sizediff >= max_size)
1622 return 0;
1623 if (trg_size < src_size / 32)
1624 return 0;
1625
1626 /* Load data if not already done */
1627 if (!trg->data) {
1628 read_lock();
1629 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1630 read_unlock();
1631 if (!trg->data)
1632 die("object %s cannot be read",
1633 sha1_to_hex(trg_entry->idx.sha1));
1634 if (sz != trg_size)
1635 die("object %s inconsistent object length (%lu vs %lu)",
1636 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1637 *mem_usage += sz;
1638 }
1639 if (!src->data) {
1640 read_lock();
1641 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1642 read_unlock();
1643 if (!src->data) {
1644 if (src_entry->preferred_base) {
1645 static int warned = 0;
1646 if (!warned++)
1647 warning("object %s cannot be read",
1648 sha1_to_hex(src_entry->idx.sha1));
1649 /*
1650 * Those objects are not included in the
1651 * resulting pack. Be resilient and ignore
1652 * them if they can't be read, in case the
1653 * pack could be created nevertheless.
1654 */
1655 return 0;
1656 }
1657 die("object %s cannot be read",
1658 sha1_to_hex(src_entry->idx.sha1));
1659 }
1660 if (sz != src_size)
1661 die("object %s inconsistent object length (%lu vs %lu)",
1662 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1663 *mem_usage += sz;
1664 }
1665 if (!src->index) {
1666 src->index = create_delta_index(src->data, src_size);
1667 if (!src->index) {
1668 static int warned = 0;
1669 if (!warned++)
1670 warning("suboptimal pack - out of memory");
1671 return 0;
1672 }
1673 *mem_usage += sizeof_delta_index(src->index);
1674 }
1675
1676 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1677 if (!delta_buf)
1678 return 0;
1679
1680 if (trg_entry->delta) {
1681 /* Prefer only shallower same-sized deltas. */
1682 if (delta_size == trg_entry->delta_size &&
1683 src->depth + 1 >= trg->depth) {
1684 free(delta_buf);
1685 return 0;
1686 }
1687 }
1688
1689 /*
1690 * Handle memory allocation outside of the cache
1691 * accounting lock. Compiler will optimize the strangeness
1692 * away when NO_PTHREADS is defined.
1693 */
1694 free(trg_entry->delta_data);
1695 cache_lock();
1696 if (trg_entry->delta_data) {
1697 delta_cache_size -= trg_entry->delta_size;
1698 trg_entry->delta_data = NULL;
1699 }
1700 if (delta_cacheable(src_size, trg_size, delta_size)) {
1701 delta_cache_size += delta_size;
1702 cache_unlock();
1703 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1704 } else {
1705 cache_unlock();
1706 free(delta_buf);
1707 }
1708
1709 trg_entry->delta = src_entry;
1710 trg_entry->delta_size = delta_size;
1711 trg->depth = src->depth + 1;
1712
1713 return 1;
1714 }
1715
1716 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1717 {
1718 struct object_entry *child = me->delta_child;
1719 unsigned int m = n;
1720 while (child) {
1721 unsigned int c = check_delta_limit(child, n + 1);
1722 if (m < c)
1723 m = c;
1724 child = child->delta_sibling;
1725 }
1726 return m;
1727 }
1728
1729 static unsigned long free_unpacked(struct unpacked *n)
1730 {
1731 unsigned long freed_mem = sizeof_delta_index(n->index);
1732 free_delta_index(n->index);
1733 n->index = NULL;
1734 if (n->data) {
1735 freed_mem += n->entry->size;
1736 free(n->data);
1737 n->data = NULL;
1738 }
1739 n->entry = NULL;
1740 n->depth = 0;
1741 return freed_mem;
1742 }
1743
1744 static void find_deltas(struct object_entry **list, unsigned *list_size,
1745 int window, int depth, unsigned *processed)
1746 {
1747 uint32_t i, idx = 0, count = 0;
1748 struct unpacked *array;
1749 unsigned long mem_usage = 0;
1750
1751 array = xcalloc(window, sizeof(struct unpacked));
1752
1753 for (;;) {
1754 struct object_entry *entry;
1755 struct unpacked *n = array + idx;
1756 int j, max_depth, best_base = -1;
1757
1758 progress_lock();
1759 if (!*list_size) {
1760 progress_unlock();
1761 break;
1762 }
1763 entry = *list++;
1764 (*list_size)--;
1765 if (!entry->preferred_base) {
1766 (*processed)++;
1767 display_progress(progress_state, *processed);
1768 }
1769 progress_unlock();
1770
1771 mem_usage -= free_unpacked(n);
1772 n->entry = entry;
1773
1774 while (window_memory_limit &&
1775 mem_usage > window_memory_limit &&
1776 count > 1) {
1777 uint32_t tail = (idx + window - count) % window;
1778 mem_usage -= free_unpacked(array + tail);
1779 count--;
1780 }
1781
1782 /* We do not compute delta to *create* objects we are not
1783 * going to pack.
1784 */
1785 if (entry->preferred_base)
1786 goto next;
1787
1788 /*
1789 * If the current object is at pack edge, take the depth the
1790 * objects that depend on the current object into account
1791 * otherwise they would become too deep.
1792 */
1793 max_depth = depth;
1794 if (entry->delta_child) {
1795 max_depth -= check_delta_limit(entry, 0);
1796 if (max_depth <= 0)
1797 goto next;
1798 }
1799
1800 j = window;
1801 while (--j > 0) {
1802 int ret;
1803 uint32_t other_idx = idx + j;
1804 struct unpacked *m;
1805 if (other_idx >= window)
1806 other_idx -= window;
1807 m = array + other_idx;
1808 if (!m->entry)
1809 break;
1810 ret = try_delta(n, m, max_depth, &mem_usage);
1811 if (ret < 0)
1812 break;
1813 else if (ret > 0)
1814 best_base = other_idx;
1815 }
1816
1817 /*
1818 * If we decided to cache the delta data, then it is best
1819 * to compress it right away. First because we have to do
1820 * it anyway, and doing it here while we're threaded will
1821 * save a lot of time in the non threaded write phase,
1822 * as well as allow for caching more deltas within
1823 * the same cache size limit.
1824 * ...
1825 * But only if not writing to stdout, since in that case
1826 * the network is most likely throttling writes anyway,
1827 * and therefore it is best to go to the write phase ASAP
1828 * instead, as we can afford spending more time compressing
1829 * between writes at that moment.
1830 */
1831 if (entry->delta_data && !pack_to_stdout) {
1832 entry->z_delta_size = do_compress(&entry->delta_data,
1833 entry->delta_size);
1834 cache_lock();
1835 delta_cache_size -= entry->delta_size;
1836 delta_cache_size += entry->z_delta_size;
1837 cache_unlock();
1838 }
1839
1840 /* if we made n a delta, and if n is already at max
1841 * depth, leaving it in the window is pointless. we
1842 * should evict it first.
1843 */
1844 if (entry->delta && max_depth <= n->depth)
1845 continue;
1846
1847 /*
1848 * Move the best delta base up in the window, after the
1849 * currently deltified object, to keep it longer. It will
1850 * be the first base object to be attempted next.
1851 */
1852 if (entry->delta) {
1853 struct unpacked swap = array[best_base];
1854 int dist = (window + idx - best_base) % window;
1855 int dst = best_base;
1856 while (dist--) {
1857 int src = (dst + 1) % window;
1858 array[dst] = array[src];
1859 dst = src;
1860 }
1861 array[dst] = swap;
1862 }
1863
1864 next:
1865 idx++;
1866 if (count + 1 < window)
1867 count++;
1868 if (idx >= window)
1869 idx = 0;
1870 }
1871
1872 for (i = 0; i < window; ++i) {
1873 free_delta_index(array[i].index);
1874 free(array[i].data);
1875 }
1876 free(array);
1877 }
1878
1879 #ifndef NO_PTHREADS
1880
1881 static void try_to_free_from_threads(size_t size)
1882 {
1883 read_lock();
1884 release_pack_memory(size);
1885 read_unlock();
1886 }
1887
1888 static try_to_free_t old_try_to_free_routine;
1889
1890 /*
1891 * The main thread waits on the condition that (at least) one of the workers
1892 * has stopped working (which is indicated in the .working member of
1893 * struct thread_params).
1894 * When a work thread has completed its work, it sets .working to 0 and
1895 * signals the main thread and waits on the condition that .data_ready
1896 * becomes 1.
1897 */
1898
1899 struct thread_params {
1900 pthread_t thread;
1901 struct object_entry **list;
1902 unsigned list_size;
1903 unsigned remaining;
1904 int window;
1905 int depth;
1906 int working;
1907 int data_ready;
1908 pthread_mutex_t mutex;
1909 pthread_cond_t cond;
1910 unsigned *processed;
1911 };
1912
1913 static pthread_cond_t progress_cond;
1914
1915 /*
1916 * Mutex and conditional variable can't be statically-initialized on Windows.
1917 */
1918 static void init_threaded_search(void)
1919 {
1920 init_recursive_mutex(&read_mutex);
1921 pthread_mutex_init(&cache_mutex, NULL);
1922 pthread_mutex_init(&progress_mutex, NULL);
1923 pthread_cond_init(&progress_cond, NULL);
1924 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1925 }
1926
1927 static void cleanup_threaded_search(void)
1928 {
1929 set_try_to_free_routine(old_try_to_free_routine);
1930 pthread_cond_destroy(&progress_cond);
1931 pthread_mutex_destroy(&read_mutex);
1932 pthread_mutex_destroy(&cache_mutex);
1933 pthread_mutex_destroy(&progress_mutex);
1934 }
1935
1936 static void *threaded_find_deltas(void *arg)
1937 {
1938 struct thread_params *me = arg;
1939
1940 while (me->remaining) {
1941 find_deltas(me->list, &me->remaining,
1942 me->window, me->depth, me->processed);
1943
1944 progress_lock();
1945 me->working = 0;
1946 pthread_cond_signal(&progress_cond);
1947 progress_unlock();
1948
1949 /*
1950 * We must not set ->data_ready before we wait on the
1951 * condition because the main thread may have set it to 1
1952 * before we get here. In order to be sure that new
1953 * work is available if we see 1 in ->data_ready, it
1954 * was initialized to 0 before this thread was spawned
1955 * and we reset it to 0 right away.
1956 */
1957 pthread_mutex_lock(&me->mutex);
1958 while (!me->data_ready)
1959 pthread_cond_wait(&me->cond, &me->mutex);
1960 me->data_ready = 0;
1961 pthread_mutex_unlock(&me->mutex);
1962 }
1963 /* leave ->working 1 so that this doesn't get more work assigned */
1964 return NULL;
1965 }
1966
1967 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1968 int window, int depth, unsigned *processed)
1969 {
1970 struct thread_params *p;
1971 int i, ret, active_threads = 0;
1972
1973 init_threaded_search();
1974
1975 if (!delta_search_threads) /* --threads=0 means autodetect */
1976 delta_search_threads = online_cpus();
1977 if (delta_search_threads <= 1) {
1978 find_deltas(list, &list_size, window, depth, processed);
1979 cleanup_threaded_search();
1980 return;
1981 }
1982 if (progress > pack_to_stdout)
1983 fprintf(stderr, "Delta compression using up to %d threads.\n",
1984 delta_search_threads);
1985 p = xcalloc(delta_search_threads, sizeof(*p));
1986
1987 /* Partition the work amongst work threads. */
1988 for (i = 0; i < delta_search_threads; i++) {
1989 unsigned sub_size = list_size / (delta_search_threads - i);
1990
1991 /* don't use too small segments or no deltas will be found */
1992 if (sub_size < 2*window && i+1 < delta_search_threads)
1993 sub_size = 0;
1994
1995 p[i].window = window;
1996 p[i].depth = depth;
1997 p[i].processed = processed;
1998 p[i].working = 1;
1999 p[i].data_ready = 0;
2000
2001 /* try to split chunks on "path" boundaries */
2002 while (sub_size && sub_size < list_size &&
2003 list[sub_size]->hash &&
2004 list[sub_size]->hash == list[sub_size-1]->hash)
2005 sub_size++;
2006
2007 p[i].list = list;
2008 p[i].list_size = sub_size;
2009 p[i].remaining = sub_size;
2010
2011 list += sub_size;
2012 list_size -= sub_size;
2013 }
2014
2015 /* Start work threads. */
2016 for (i = 0; i < delta_search_threads; i++) {
2017 if (!p[i].list_size)
2018 continue;
2019 pthread_mutex_init(&p[i].mutex, NULL);
2020 pthread_cond_init(&p[i].cond, NULL);
2021 ret = pthread_create(&p[i].thread, NULL,
2022 threaded_find_deltas, &p[i]);
2023 if (ret)
2024 die("unable to create thread: %s", strerror(ret));
2025 active_threads++;
2026 }
2027
2028 /*
2029 * Now let's wait for work completion. Each time a thread is done
2030 * with its work, we steal half of the remaining work from the
2031 * thread with the largest number of unprocessed objects and give
2032 * it to that newly idle thread. This ensure good load balancing
2033 * until the remaining object list segments are simply too short
2034 * to be worth splitting anymore.
2035 */
2036 while (active_threads) {
2037 struct thread_params *target = NULL;
2038 struct thread_params *victim = NULL;
2039 unsigned sub_size = 0;
2040
2041 progress_lock();
2042 for (;;) {
2043 for (i = 0; !target && i < delta_search_threads; i++)
2044 if (!p[i].working)
2045 target = &p[i];
2046 if (target)
2047 break;
2048 pthread_cond_wait(&progress_cond, &progress_mutex);
2049 }
2050
2051 for (i = 0; i < delta_search_threads; i++)
2052 if (p[i].remaining > 2*window &&
2053 (!victim || victim->remaining < p[i].remaining))
2054 victim = &p[i];
2055 if (victim) {
2056 sub_size = victim->remaining / 2;
2057 list = victim->list + victim->list_size - sub_size;
2058 while (sub_size && list[0]->hash &&
2059 list[0]->hash == list[-1]->hash) {
2060 list++;
2061 sub_size--;
2062 }
2063 if (!sub_size) {
2064 /*
2065 * It is possible for some "paths" to have
2066 * so many objects that no hash boundary
2067 * might be found. Let's just steal the
2068 * exact half in that case.
2069 */
2070 sub_size = victim->remaining / 2;
2071 list -= sub_size;
2072 }
2073 target->list = list;
2074 victim->list_size -= sub_size;
2075 victim->remaining -= sub_size;
2076 }
2077 target->list_size = sub_size;
2078 target->remaining = sub_size;
2079 target->working = 1;
2080 progress_unlock();
2081
2082 pthread_mutex_lock(&target->mutex);
2083 target->data_ready = 1;
2084 pthread_cond_signal(&target->cond);
2085 pthread_mutex_unlock(&target->mutex);
2086
2087 if (!sub_size) {
2088 pthread_join(target->thread, NULL);
2089 pthread_cond_destroy(&target->cond);
2090 pthread_mutex_destroy(&target->mutex);
2091 active_threads--;
2092 }
2093 }
2094 cleanup_threaded_search();
2095 free(p);
2096 }
2097
2098 #else
2099 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2100 #endif
2101
2102 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
2103 {
2104 unsigned char peeled[20];
2105
2106 if (starts_with(path, "refs/tags/") && /* is a tag? */
2107 !peel_ref(path, peeled) && /* peelable? */
2108 packlist_find(&to_pack, peeled, NULL)) /* object packed? */
2109 add_object_entry(sha1, OBJ_TAG, NULL, 0);
2110 return 0;
2111 }
2112
2113 static void prepare_pack(int window, int depth)
2114 {
2115 struct object_entry **delta_list;
2116 uint32_t i, nr_deltas;
2117 unsigned n;
2118
2119 get_object_details();
2120
2121 /*
2122 * If we're locally repacking then we need to be doubly careful
2123 * from now on in order to make sure no stealth corruption gets
2124 * propagated to the new pack. Clients receiving streamed packs
2125 * should validate everything they get anyway so no need to incur
2126 * the additional cost here in that case.
2127 */
2128 if (!pack_to_stdout)
2129 do_check_packed_object_crc = 1;
2130
2131 if (!to_pack.nr_objects || !window || !depth)
2132 return;
2133
2134 delta_list = xmalloc(to_pack.nr_objects * sizeof(*delta_list));
2135 nr_deltas = n = 0;
2136
2137 for (i = 0; i < to_pack.nr_objects; i++) {
2138 struct object_entry *entry = to_pack.objects + i;
2139
2140 if (entry->delta)
2141 /* This happens if we decided to reuse existing
2142 * delta from a pack. "reuse_delta &&" is implied.
2143 */
2144 continue;
2145
2146 if (entry->size < 50)
2147 continue;
2148
2149 if (entry->no_try_delta)
2150 continue;
2151
2152 if (!entry->preferred_base) {
2153 nr_deltas++;
2154 if (entry->type < 0)
2155 die("unable to get type of object %s",
2156 sha1_to_hex(entry->idx.sha1));
2157 } else {
2158 if (entry->type < 0) {
2159 /*
2160 * This object is not found, but we
2161 * don't have to include it anyway.
2162 */
2163 continue;
2164 }
2165 }
2166
2167 delta_list[n++] = entry;
2168 }
2169
2170 if (nr_deltas && n > 1) {
2171 unsigned nr_done = 0;
2172 if (progress)
2173 progress_state = start_progress(_("Compressing objects"),
2174 nr_deltas);
2175 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2176 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2177 stop_progress(&progress_state);
2178 if (nr_done != nr_deltas)
2179 die("inconsistency with delta count");
2180 }
2181 free(delta_list);
2182 }
2183
2184 static int git_pack_config(const char *k, const char *v, void *cb)
2185 {
2186 if (!strcmp(k, "pack.window")) {
2187 window = git_config_int(k, v);
2188 return 0;
2189 }
2190 if (!strcmp(k, "pack.windowmemory")) {
2191 window_memory_limit = git_config_ulong(k, v);
2192 return 0;
2193 }
2194 if (!strcmp(k, "pack.depth")) {
2195 depth = git_config_int(k, v);
2196 return 0;
2197 }
2198 if (!strcmp(k, "pack.compression")) {
2199 int level = git_config_int(k, v);
2200 if (level == -1)
2201 level = Z_DEFAULT_COMPRESSION;
2202 else if (level < 0 || level > Z_BEST_COMPRESSION)
2203 die("bad pack compression level %d", level);
2204 pack_compression_level = level;
2205 pack_compression_seen = 1;
2206 return 0;
2207 }
2208 if (!strcmp(k, "pack.deltacachesize")) {
2209 max_delta_cache_size = git_config_int(k, v);
2210 return 0;
2211 }
2212 if (!strcmp(k, "pack.deltacachelimit")) {
2213 cache_max_small_delta_size = git_config_int(k, v);
2214 return 0;
2215 }
2216 if (!strcmp(k, "pack.writebitmaphashcache")) {
2217 if (git_config_bool(k, v))
2218 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2219 else
2220 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2221 }
2222 if (!strcmp(k, "pack.usebitmaps")) {
2223 use_bitmap_index = git_config_bool(k, v);
2224 return 0;
2225 }
2226 if (!strcmp(k, "pack.threads")) {
2227 delta_search_threads = git_config_int(k, v);
2228 if (delta_search_threads < 0)
2229 die("invalid number of threads specified (%d)",
2230 delta_search_threads);
2231 #ifdef NO_PTHREADS
2232 if (delta_search_threads != 1)
2233 warning("no threads support, ignoring %s", k);
2234 #endif
2235 return 0;
2236 }
2237 if (!strcmp(k, "pack.indexversion")) {
2238 pack_idx_opts.version = git_config_int(k, v);
2239 if (pack_idx_opts.version > 2)
2240 die("bad pack.indexversion=%"PRIu32,
2241 pack_idx_opts.version);
2242 return 0;
2243 }
2244 return git_default_config(k, v, cb);
2245 }
2246
2247 static void read_object_list_from_stdin(void)
2248 {
2249 char line[40 + 1 + PATH_MAX + 2];
2250 unsigned char sha1[20];
2251
2252 for (;;) {
2253 if (!fgets(line, sizeof(line), stdin)) {
2254 if (feof(stdin))
2255 break;
2256 if (!ferror(stdin))
2257 die("fgets returned NULL, not EOF, not error!");
2258 if (errno != EINTR)
2259 die_errno("fgets");
2260 clearerr(stdin);
2261 continue;
2262 }
2263 if (line[0] == '-') {
2264 if (get_sha1_hex(line+1, sha1))
2265 die("expected edge sha1, got garbage:\n %s",
2266 line);
2267 add_preferred_base(sha1);
2268 continue;
2269 }
2270 if (get_sha1_hex(line, sha1))
2271 die("expected sha1, got garbage:\n %s", line);
2272
2273 add_preferred_base_object(line+41);
2274 add_object_entry(sha1, 0, line+41, 0);
2275 }
2276 }
2277
2278 #define OBJECT_ADDED (1u<<20)
2279
2280 static void show_commit(struct commit *commit, void *data)
2281 {
2282 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2283 commit->object.flags |= OBJECT_ADDED;
2284
2285 if (write_bitmap_index)
2286 index_commit_for_bitmap(commit);
2287 }
2288
2289 static void show_object(struct object *obj,
2290 const struct name_path *path, const char *last,
2291 void *data)
2292 {
2293 char *name = path_name(path, last);
2294
2295 add_preferred_base_object(name);
2296 add_object_entry(obj->sha1, obj->type, name, 0);
2297 obj->flags |= OBJECT_ADDED;
2298
2299 /*
2300 * We will have generated the hash from the name,
2301 * but not saved a pointer to it - we can free it
2302 */
2303 free((char *)name);
2304 }
2305
2306 static void show_edge(struct commit *commit)
2307 {
2308 add_preferred_base(commit->object.sha1);
2309 }
2310
2311 struct in_pack_object {
2312 off_t offset;
2313 struct object *object;
2314 };
2315
2316 struct in_pack {
2317 int alloc;
2318 int nr;
2319 struct in_pack_object *array;
2320 };
2321
2322 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2323 {
2324 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2325 in_pack->array[in_pack->nr].object = object;
2326 in_pack->nr++;
2327 }
2328
2329 /*
2330 * Compare the objects in the offset order, in order to emulate the
2331 * "git rev-list --objects" output that produced the pack originally.
2332 */
2333 static int ofscmp(const void *a_, const void *b_)
2334 {
2335 struct in_pack_object *a = (struct in_pack_object *)a_;
2336 struct in_pack_object *b = (struct in_pack_object *)b_;
2337
2338 if (a->offset < b->offset)
2339 return -1;
2340 else if (a->offset > b->offset)
2341 return 1;
2342 else
2343 return hashcmp(a->object->sha1, b->object->sha1);
2344 }
2345
2346 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2347 {
2348 struct packed_git *p;
2349 struct in_pack in_pack;
2350 uint32_t i;
2351
2352 memset(&in_pack, 0, sizeof(in_pack));
2353
2354 for (p = packed_git; p; p = p->next) {
2355 const unsigned char *sha1;
2356 struct object *o;
2357
2358 if (!p->pack_local || p->pack_keep)
2359 continue;
2360 if (open_pack_index(p))
2361 die("cannot open pack index");
2362
2363 ALLOC_GROW(in_pack.array,
2364 in_pack.nr + p->num_objects,
2365 in_pack.alloc);
2366
2367 for (i = 0; i < p->num_objects; i++) {
2368 sha1 = nth_packed_object_sha1(p, i);
2369 o = lookup_unknown_object(sha1);
2370 if (!(o->flags & OBJECT_ADDED))
2371 mark_in_pack_object(o, p, &in_pack);
2372 o->flags |= OBJECT_ADDED;
2373 }
2374 }
2375
2376 if (in_pack.nr) {
2377 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2378 ofscmp);
2379 for (i = 0; i < in_pack.nr; i++) {
2380 struct object *o = in_pack.array[i].object;
2381 add_object_entry(o->sha1, o->type, "", 0);
2382 }
2383 }
2384 free(in_pack.array);
2385 }
2386
2387 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2388 {
2389 static struct packed_git *last_found = (void *)1;
2390 struct packed_git *p;
2391
2392 p = (last_found != (void *)1) ? last_found : packed_git;
2393
2394 while (p) {
2395 if ((!p->pack_local || p->pack_keep) &&
2396 find_pack_entry_one(sha1, p)) {
2397 last_found = p;
2398 return 1;
2399 }
2400 if (p == last_found)
2401 p = packed_git;
2402 else
2403 p = p->next;
2404 if (p == last_found)
2405 p = p->next;
2406 }
2407 return 0;
2408 }
2409
2410 static void loosen_unused_packed_objects(struct rev_info *revs)
2411 {
2412 struct packed_git *p;
2413 uint32_t i;
2414 const unsigned char *sha1;
2415
2416 for (p = packed_git; p; p = p->next) {
2417 if (!p->pack_local || p->pack_keep)
2418 continue;
2419
2420 if (unpack_unreachable_expiration &&
2421 p->mtime < unpack_unreachable_expiration)
2422 continue;
2423
2424 if (open_pack_index(p))
2425 die("cannot open pack index");
2426
2427 for (i = 0; i < p->num_objects; i++) {
2428 sha1 = nth_packed_object_sha1(p, i);
2429 if (!packlist_find(&to_pack, sha1, NULL) &&
2430 !has_sha1_pack_kept_or_nonlocal(sha1))
2431 if (force_object_loose(sha1, p->mtime))
2432 die("unable to force loose object");
2433 }
2434 }
2435 }
2436
2437 /*
2438 * This tracks any options which a reader of the pack might
2439 * not understand, and which would therefore prevent blind reuse
2440 * of what we have on disk.
2441 */
2442 static int pack_options_allow_reuse(void)
2443 {
2444 return allow_ofs_delta;
2445 }
2446
2447 static int get_object_list_from_bitmap(struct rev_info *revs)
2448 {
2449 if (prepare_bitmap_walk(revs) < 0)
2450 return -1;
2451
2452 if (pack_options_allow_reuse() &&
2453 !reuse_partial_packfile_from_bitmap(
2454 &reuse_packfile,
2455 &reuse_packfile_objects,
2456 &reuse_packfile_offset)) {
2457 assert(reuse_packfile_objects);
2458 nr_result += reuse_packfile_objects;
2459 display_progress(progress_state, nr_result);
2460 }
2461
2462 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2463 return 0;
2464 }
2465
2466 static void get_object_list(int ac, const char **av)
2467 {
2468 struct rev_info revs;
2469 char line[1000];
2470 int flags = 0;
2471
2472 init_revisions(&revs, NULL);
2473 save_commit_buffer = 0;
2474 setup_revisions(ac, av, &revs, NULL);
2475
2476 /* make sure shallows are read */
2477 is_repository_shallow();
2478
2479 while (fgets(line, sizeof(line), stdin) != NULL) {
2480 int len = strlen(line);
2481 if (len && line[len - 1] == '\n')
2482 line[--len] = 0;
2483 if (!len)
2484 break;
2485 if (*line == '-') {
2486 if (!strcmp(line, "--not")) {
2487 flags ^= UNINTERESTING;
2488 write_bitmap_index = 0;
2489 continue;
2490 }
2491 if (starts_with(line, "--shallow ")) {
2492 unsigned char sha1[20];
2493 if (get_sha1_hex(line + 10, sha1))
2494 die("not an SHA-1 '%s'", line + 10);
2495 register_shallow(sha1);
2496 use_bitmap_index = 0;
2497 continue;
2498 }
2499 die("not a rev '%s'", line);
2500 }
2501 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2502 die("bad revision '%s'", line);
2503 }
2504
2505 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2506 return;
2507
2508 if (prepare_revision_walk(&revs))
2509 die("revision walk setup failed");
2510 mark_edges_uninteresting(&revs, show_edge);
2511 traverse_commit_list(&revs, show_commit, show_object, NULL);
2512
2513 if (keep_unreachable)
2514 add_objects_in_unpacked_packs(&revs);
2515 if (unpack_unreachable)
2516 loosen_unused_packed_objects(&revs);
2517 }
2518
2519 static int option_parse_index_version(const struct option *opt,
2520 const char *arg, int unset)
2521 {
2522 char *c;
2523 const char *val = arg;
2524 pack_idx_opts.version = strtoul(val, &c, 10);
2525 if (pack_idx_opts.version > 2)
2526 die(_("unsupported index version %s"), val);
2527 if (*c == ',' && c[1])
2528 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2529 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2530 die(_("bad index version '%s'"), val);
2531 return 0;
2532 }
2533
2534 static int option_parse_unpack_unreachable(const struct option *opt,
2535 const char *arg, int unset)
2536 {
2537 if (unset) {
2538 unpack_unreachable = 0;
2539 unpack_unreachable_expiration = 0;
2540 }
2541 else {
2542 unpack_unreachable = 1;
2543 if (arg)
2544 unpack_unreachable_expiration = approxidate(arg);
2545 }
2546 return 0;
2547 }
2548
2549 static int option_parse_ulong(const struct option *opt,
2550 const char *arg, int unset)
2551 {
2552 if (unset)
2553 die(_("option %s does not accept negative form"),
2554 opt->long_name);
2555
2556 if (!git_parse_ulong(arg, opt->value))
2557 die(_("unable to parse value '%s' for option %s"),
2558 arg, opt->long_name);
2559 return 0;
2560 }
2561
2562 #define OPT_ULONG(s, l, v, h) \
2563 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2564 PARSE_OPT_NONEG, option_parse_ulong }
2565
2566 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2567 {
2568 int use_internal_rev_list = 0;
2569 int thin = 0;
2570 int all_progress_implied = 0;
2571 const char *rp_av[6];
2572 int rp_ac = 0;
2573 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2574 struct option pack_objects_options[] = {
2575 OPT_SET_INT('q', "quiet", &progress,
2576 N_("do not show progress meter"), 0),
2577 OPT_SET_INT(0, "progress", &progress,
2578 N_("show progress meter"), 1),
2579 OPT_SET_INT(0, "all-progress", &progress,
2580 N_("show progress meter during object writing phase"), 2),
2581 OPT_BOOL(0, "all-progress-implied",
2582 &all_progress_implied,
2583 N_("similar to --all-progress when progress meter is shown")),
2584 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2585 N_("write the pack index file in the specified idx format version"),
2586 0, option_parse_index_version },
2587 OPT_ULONG(0, "max-pack-size", &pack_size_limit,
2588 N_("maximum size of each output pack file")),
2589 OPT_BOOL(0, "local", &local,
2590 N_("ignore borrowed objects from alternate object store")),
2591 OPT_BOOL(0, "incremental", &incremental,
2592 N_("ignore packed objects")),
2593 OPT_INTEGER(0, "window", &window,
2594 N_("limit pack window by objects")),
2595 OPT_ULONG(0, "window-memory", &window_memory_limit,
2596 N_("limit pack window by memory in addition to object limit")),
2597 OPT_INTEGER(0, "depth", &depth,
2598 N_("maximum length of delta chain allowed in the resulting pack")),
2599 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2600 N_("reuse existing deltas")),
2601 OPT_BOOL(0, "reuse-object", &reuse_object,
2602 N_("reuse existing objects")),
2603 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2604 N_("use OFS_DELTA objects")),
2605 OPT_INTEGER(0, "threads", &delta_search_threads,
2606 N_("use threads when searching for best delta matches")),
2607 OPT_BOOL(0, "non-empty", &non_empty,
2608 N_("do not create an empty pack output")),
2609 OPT_BOOL(0, "revs", &use_internal_rev_list,
2610 N_("read revision arguments from standard input")),
2611 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2612 N_("limit the objects to those that are not yet packed"),
2613 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2614 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2615 N_("include objects reachable from any reference"),
2616 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2617 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2618 N_("include objects referred by reflog entries"),
2619 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2620 OPT_BOOL(0, "stdout", &pack_to_stdout,
2621 N_("output pack to stdout")),
2622 OPT_BOOL(0, "include-tag", &include_tag,
2623 N_("include tag objects that refer to objects to be packed")),
2624 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2625 N_("keep unreachable objects")),
2626 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2627 N_("unpack unreachable objects newer than <time>"),
2628 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2629 OPT_BOOL(0, "thin", &thin,
2630 N_("create thin packs")),
2631 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2632 N_("ignore packs that have companion .keep file")),
2633 OPT_INTEGER(0, "compression", &pack_compression_level,
2634 N_("pack compression level")),
2635 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2636 N_("do not hide commits by grafts"), 0),
2637 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2638 N_("use a bitmap index if available to speed up counting objects")),
2639 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2640 N_("write a bitmap index together with the pack index")),
2641 OPT_END(),
2642 };
2643
2644 check_replace_refs = 0;
2645
2646 reset_pack_idx_option(&pack_idx_opts);
2647 git_config(git_pack_config, NULL);
2648 if (!pack_compression_seen && core_compression_seen)
2649 pack_compression_level = core_compression_level;
2650
2651 progress = isatty(2);
2652 argc = parse_options(argc, argv, prefix, pack_objects_options,
2653 pack_usage, 0);
2654
2655 if (argc) {
2656 base_name = argv[0];
2657 argc--;
2658 }
2659 if (pack_to_stdout != !base_name || argc)
2660 usage_with_options(pack_usage, pack_objects_options);
2661
2662 rp_av[rp_ac++] = "pack-objects";
2663 if (thin) {
2664 use_internal_rev_list = 1;
2665 rp_av[rp_ac++] = "--objects-edge";
2666 } else
2667 rp_av[rp_ac++] = "--objects";
2668
2669 if (rev_list_all) {
2670 use_internal_rev_list = 1;
2671 rp_av[rp_ac++] = "--all";
2672 }
2673 if (rev_list_reflog) {
2674 use_internal_rev_list = 1;
2675 rp_av[rp_ac++] = "--reflog";
2676 }
2677 if (rev_list_unpacked) {
2678 use_internal_rev_list = 1;
2679 rp_av[rp_ac++] = "--unpacked";
2680 }
2681
2682 if (!reuse_object)
2683 reuse_delta = 0;
2684 if (pack_compression_level == -1)
2685 pack_compression_level = Z_DEFAULT_COMPRESSION;
2686 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2687 die("bad pack compression level %d", pack_compression_level);
2688 #ifdef NO_PTHREADS
2689 if (delta_search_threads != 1)
2690 warning("no threads support, ignoring --threads");
2691 #endif
2692 if (!pack_to_stdout && !pack_size_limit)
2693 pack_size_limit = pack_size_limit_cfg;
2694 if (pack_to_stdout && pack_size_limit)
2695 die("--max-pack-size cannot be used to build a pack for transfer.");
2696 if (pack_size_limit && pack_size_limit < 1024*1024) {
2697 warning("minimum pack size limit is 1 MiB");
2698 pack_size_limit = 1024*1024;
2699 }
2700
2701 if (!pack_to_stdout && thin)
2702 die("--thin cannot be used to build an indexable pack.");
2703
2704 if (keep_unreachable && unpack_unreachable)
2705 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2706
2707 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2708 use_bitmap_index = 0;
2709
2710 if (pack_to_stdout || !rev_list_all)
2711 write_bitmap_index = 0;
2712
2713 if (progress && all_progress_implied)
2714 progress = 2;
2715
2716 prepare_packed_git();
2717
2718 if (progress)
2719 progress_state = start_progress(_("Counting objects"), 0);
2720 if (!use_internal_rev_list)
2721 read_object_list_from_stdin();
2722 else {
2723 rp_av[rp_ac] = NULL;
2724 get_object_list(rp_ac, rp_av);
2725 }
2726 cleanup_preferred_base();
2727 if (include_tag && nr_result)
2728 for_each_ref(add_ref_tag, NULL);
2729 stop_progress(&progress_state);
2730
2731 if (non_empty && !nr_result)
2732 return 0;
2733 if (nr_result)
2734 prepare_pack(window, depth);
2735 write_pack_file();
2736 if (progress)
2737 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2738 " reused %"PRIu32" (delta %"PRIu32")\n",
2739 written, written_delta, reused, reused_delta);
2740 return 0;
2741 }