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#include "reachable.h"
24#include "sha1-array.h"
25#include "argv-array.h"
26
27static const char *pack_usage[] = {
28 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
29 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
30 NULL
31};
32
33/*
34 * Objects we are going to pack are collected in the `to_pack` structure.
35 * It contains an array (dynamically expanded) of the object data, and a map
36 * that can resolve SHA1s to their position in the array.
37 */
38static struct packing_data to_pack;
39
40static struct pack_idx_entry **written_list;
41static uint32_t nr_result, nr_written;
42
43static int non_empty;
44static int reuse_delta = 1, reuse_object = 1;
45static int keep_unreachable, unpack_unreachable, include_tag;
46static unsigned long unpack_unreachable_expiration;
47static int local;
48static int incremental;
49static int ignore_packed_keep;
50static int allow_ofs_delta;
51static struct pack_idx_option pack_idx_opts;
52static const char *base_name;
53static int progress = 1;
54static int window = 10;
55static unsigned long pack_size_limit;
56static int depth = 50;
57static int delta_search_threads;
58static int pack_to_stdout;
59static int num_preferred_base;
60static struct progress *progress_state;
61static int pack_compression_level = Z_DEFAULT_COMPRESSION;
62static int pack_compression_seen;
63
64static struct packed_git *reuse_packfile;
65static uint32_t reuse_packfile_objects;
66static off_t reuse_packfile_offset;
67
68static int use_bitmap_index = 1;
69static int write_bitmap_index;
70static uint16_t write_bitmap_options;
71
72static unsigned long delta_cache_size = 0;
73static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
74static unsigned long cache_max_small_delta_size = 1000;
75
76static unsigned long window_memory_limit = 0;
77
78/*
79 * stats
80 */
81static uint32_t written, written_delta;
82static uint32_t reused, reused_delta;
83
84/*
85 * Indexed commits
86 */
87static struct commit **indexed_commits;
88static unsigned int indexed_commits_nr;
89static unsigned int indexed_commits_alloc;
90
91static void index_commit_for_bitmap(struct commit *commit)
92{
93 if (indexed_commits_nr >= indexed_commits_alloc) {
94 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
95 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
96 }
97
98 indexed_commits[indexed_commits_nr++] = commit;
99}
100
101static void *get_delta(struct object_entry *entry)
102{
103 unsigned long size, base_size, delta_size;
104 void *buf, *base_buf, *delta_buf;
105 enum object_type type;
106
107 buf = read_sha1_file(entry->idx.sha1, &type, &size);
108 if (!buf)
109 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
110 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
111 if (!base_buf)
112 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
113 delta_buf = diff_delta(base_buf, base_size,
114 buf, size, &delta_size, 0);
115 if (!delta_buf || delta_size != entry->delta_size)
116 die("delta size changed");
117 free(buf);
118 free(base_buf);
119 return delta_buf;
120}
121
122static unsigned long do_compress(void **pptr, unsigned long size)
123{
124 git_zstream stream;
125 void *in, *out;
126 unsigned long maxsize;
127
128 git_deflate_init(&stream, pack_compression_level);
129 maxsize = git_deflate_bound(&stream, size);
130
131 in = *pptr;
132 out = xmalloc(maxsize);
133 *pptr = out;
134
135 stream.next_in = in;
136 stream.avail_in = size;
137 stream.next_out = out;
138 stream.avail_out = maxsize;
139 while (git_deflate(&stream, Z_FINISH) == Z_OK)
140 ; /* nothing */
141 git_deflate_end(&stream);
142
143 free(in);
144 return stream.total_out;
145}
146
147static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
148 const unsigned char *sha1)
149{
150 git_zstream stream;
151 unsigned char ibuf[1024 * 16];
152 unsigned char obuf[1024 * 16];
153 unsigned long olen = 0;
154
155 git_deflate_init(&stream, pack_compression_level);
156
157 for (;;) {
158 ssize_t readlen;
159 int zret = Z_OK;
160 readlen = read_istream(st, ibuf, sizeof(ibuf));
161 if (readlen == -1)
162 die(_("unable to read %s"), sha1_to_hex(sha1));
163
164 stream.next_in = ibuf;
165 stream.avail_in = readlen;
166 while ((stream.avail_in || readlen == 0) &&
167 (zret == Z_OK || zret == Z_BUF_ERROR)) {
168 stream.next_out = obuf;
169 stream.avail_out = sizeof(obuf);
170 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
171 sha1write(f, obuf, stream.next_out - obuf);
172 olen += stream.next_out - obuf;
173 }
174 if (stream.avail_in)
175 die(_("deflate error (%d)"), zret);
176 if (readlen == 0) {
177 if (zret != Z_STREAM_END)
178 die(_("deflate error (%d)"), zret);
179 break;
180 }
181 }
182 git_deflate_end(&stream);
183 return olen;
184}
185
186/*
187 * we are going to reuse the existing object data as is. make
188 * sure it is not corrupt.
189 */
190static int check_pack_inflate(struct packed_git *p,
191 struct pack_window **w_curs,
192 off_t offset,
193 off_t len,
194 unsigned long expect)
195{
196 git_zstream stream;
197 unsigned char fakebuf[4096], *in;
198 int st;
199
200 memset(&stream, 0, sizeof(stream));
201 git_inflate_init(&stream);
202 do {
203 in = use_pack(p, w_curs, offset, &stream.avail_in);
204 stream.next_in = in;
205 stream.next_out = fakebuf;
206 stream.avail_out = sizeof(fakebuf);
207 st = git_inflate(&stream, Z_FINISH);
208 offset += stream.next_in - in;
209 } while (st == Z_OK || st == Z_BUF_ERROR);
210 git_inflate_end(&stream);
211 return (st == Z_STREAM_END &&
212 stream.total_out == expect &&
213 stream.total_in == len) ? 0 : -1;
214}
215
216static void copy_pack_data(struct sha1file *f,
217 struct packed_git *p,
218 struct pack_window **w_curs,
219 off_t offset,
220 off_t len)
221{
222 unsigned char *in;
223 unsigned long avail;
224
225 while (len) {
226 in = use_pack(p, w_curs, offset, &avail);
227 if (avail > len)
228 avail = (unsigned long)len;
229 sha1write(f, in, avail);
230 offset += avail;
231 len -= avail;
232 }
233}
234
235/* Return 0 if we will bust the pack-size limit */
236static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
237 unsigned long limit, int usable_delta)
238{
239 unsigned long size, datalen;
240 unsigned char header[10], dheader[10];
241 unsigned hdrlen;
242 enum object_type type;
243 void *buf;
244 struct git_istream *st = NULL;
245
246 if (!usable_delta) {
247 if (entry->type == OBJ_BLOB &&
248 entry->size > big_file_threshold &&
249 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
250 buf = NULL;
251 else {
252 buf = read_sha1_file(entry->idx.sha1, &type, &size);
253 if (!buf)
254 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
255 }
256 /*
257 * make sure no cached delta data remains from a
258 * previous attempt before a pack split occurred.
259 */
260 free(entry->delta_data);
261 entry->delta_data = NULL;
262 entry->z_delta_size = 0;
263 } else if (entry->delta_data) {
264 size = entry->delta_size;
265 buf = entry->delta_data;
266 entry->delta_data = NULL;
267 type = (allow_ofs_delta && entry->delta->idx.offset) ?
268 OBJ_OFS_DELTA : OBJ_REF_DELTA;
269 } else {
270 buf = get_delta(entry);
271 size = entry->delta_size;
272 type = (allow_ofs_delta && entry->delta->idx.offset) ?
273 OBJ_OFS_DELTA : OBJ_REF_DELTA;
274 }
275
276 if (st) /* large blob case, just assume we don't compress well */
277 datalen = size;
278 else if (entry->z_delta_size)
279 datalen = entry->z_delta_size;
280 else
281 datalen = do_compress(&buf, size);
282
283 /*
284 * The object header is a byte of 'type' followed by zero or
285 * more bytes of length.
286 */
287 hdrlen = encode_in_pack_object_header(type, size, header);
288
289 if (type == OBJ_OFS_DELTA) {
290 /*
291 * Deltas with relative base contain an additional
292 * encoding of the relative offset for the delta
293 * base from this object's position in the pack.
294 */
295 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
296 unsigned pos = sizeof(dheader) - 1;
297 dheader[pos] = ofs & 127;
298 while (ofs >>= 7)
299 dheader[--pos] = 128 | (--ofs & 127);
300 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
301 if (st)
302 close_istream(st);
303 free(buf);
304 return 0;
305 }
306 sha1write(f, header, hdrlen);
307 sha1write(f, dheader + pos, sizeof(dheader) - pos);
308 hdrlen += sizeof(dheader) - pos;
309 } else if (type == OBJ_REF_DELTA) {
310 /*
311 * Deltas with a base reference contain
312 * an additional 20 bytes for the base sha1.
313 */
314 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
315 if (st)
316 close_istream(st);
317 free(buf);
318 return 0;
319 }
320 sha1write(f, header, hdrlen);
321 sha1write(f, entry->delta->idx.sha1, 20);
322 hdrlen += 20;
323 } else {
324 if (limit && hdrlen + datalen + 20 >= limit) {
325 if (st)
326 close_istream(st);
327 free(buf);
328 return 0;
329 }
330 sha1write(f, header, hdrlen);
331 }
332 if (st) {
333 datalen = write_large_blob_data(st, f, entry->idx.sha1);
334 close_istream(st);
335 } else {
336 sha1write(f, buf, datalen);
337 free(buf);
338 }
339
340 return hdrlen + datalen;
341}
342
343/* Return 0 if we will bust the pack-size limit */
344static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
345 unsigned long limit, int usable_delta)
346{
347 struct packed_git *p = entry->in_pack;
348 struct pack_window *w_curs = NULL;
349 struct revindex_entry *revidx;
350 off_t offset;
351 enum object_type type = entry->type;
352 unsigned long datalen;
353 unsigned char header[10], dheader[10];
354 unsigned hdrlen;
355
356 if (entry->delta)
357 type = (allow_ofs_delta && entry->delta->idx.offset) ?
358 OBJ_OFS_DELTA : OBJ_REF_DELTA;
359 hdrlen = encode_in_pack_object_header(type, entry->size, header);
360
361 offset = entry->in_pack_offset;
362 revidx = find_pack_revindex(p, offset);
363 datalen = revidx[1].offset - offset;
364 if (!pack_to_stdout && p->index_version > 1 &&
365 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
366 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
367 unuse_pack(&w_curs);
368 return write_no_reuse_object(f, entry, limit, usable_delta);
369 }
370
371 offset += entry->in_pack_header_size;
372 datalen -= entry->in_pack_header_size;
373
374 if (!pack_to_stdout && p->index_version == 1 &&
375 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
376 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
377 unuse_pack(&w_curs);
378 return write_no_reuse_object(f, entry, limit, usable_delta);
379 }
380
381 if (type == OBJ_OFS_DELTA) {
382 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
383 unsigned pos = sizeof(dheader) - 1;
384 dheader[pos] = ofs & 127;
385 while (ofs >>= 7)
386 dheader[--pos] = 128 | (--ofs & 127);
387 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
388 unuse_pack(&w_curs);
389 return 0;
390 }
391 sha1write(f, header, hdrlen);
392 sha1write(f, dheader + pos, sizeof(dheader) - pos);
393 hdrlen += sizeof(dheader) - pos;
394 reused_delta++;
395 } else if (type == OBJ_REF_DELTA) {
396 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
397 unuse_pack(&w_curs);
398 return 0;
399 }
400 sha1write(f, header, hdrlen);
401 sha1write(f, entry->delta->idx.sha1, 20);
402 hdrlen += 20;
403 reused_delta++;
404 } else {
405 if (limit && hdrlen + datalen + 20 >= limit) {
406 unuse_pack(&w_curs);
407 return 0;
408 }
409 sha1write(f, header, hdrlen);
410 }
411 copy_pack_data(f, p, &w_curs, offset, datalen);
412 unuse_pack(&w_curs);
413 reused++;
414 return hdrlen + datalen;
415}
416
417/* Return 0 if we will bust the pack-size limit */
418static unsigned long write_object(struct sha1file *f,
419 struct object_entry *entry,
420 off_t write_offset)
421{
422 unsigned long limit, len;
423 int usable_delta, to_reuse;
424
425 if (!pack_to_stdout)
426 crc32_begin(f);
427
428 /* apply size limit if limited packsize and not first object */
429 if (!pack_size_limit || !nr_written)
430 limit = 0;
431 else if (pack_size_limit <= write_offset)
432 /*
433 * the earlier object did not fit the limit; avoid
434 * mistaking this with unlimited (i.e. limit = 0).
435 */
436 limit = 1;
437 else
438 limit = pack_size_limit - write_offset;
439
440 if (!entry->delta)
441 usable_delta = 0; /* no delta */
442 else if (!pack_size_limit)
443 usable_delta = 1; /* unlimited packfile */
444 else if (entry->delta->idx.offset == (off_t)-1)
445 usable_delta = 0; /* base was written to another pack */
446 else if (entry->delta->idx.offset)
447 usable_delta = 1; /* base already exists in this pack */
448 else
449 usable_delta = 0; /* base could end up in another pack */
450
451 if (!reuse_object)
452 to_reuse = 0; /* explicit */
453 else if (!entry->in_pack)
454 to_reuse = 0; /* can't reuse what we don't have */
455 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
456 /* check_object() decided it for us ... */
457 to_reuse = usable_delta;
458 /* ... but pack split may override that */
459 else if (entry->type != entry->in_pack_type)
460 to_reuse = 0; /* pack has delta which is unusable */
461 else if (entry->delta)
462 to_reuse = 0; /* we want to pack afresh */
463 else
464 to_reuse = 1; /* we have it in-pack undeltified,
465 * and we do not need to deltify it.
466 */
467
468 if (!to_reuse)
469 len = write_no_reuse_object(f, entry, limit, usable_delta);
470 else
471 len = write_reuse_object(f, entry, limit, usable_delta);
472 if (!len)
473 return 0;
474
475 if (usable_delta)
476 written_delta++;
477 written++;
478 if (!pack_to_stdout)
479 entry->idx.crc32 = crc32_end(f);
480 return len;
481}
482
483enum write_one_status {
484 WRITE_ONE_SKIP = -1, /* already written */
485 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
486 WRITE_ONE_WRITTEN = 1, /* normal */
487 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
488};
489
490static enum write_one_status write_one(struct sha1file *f,
491 struct object_entry *e,
492 off_t *offset)
493{
494 unsigned long size;
495 int recursing;
496
497 /*
498 * we set offset to 1 (which is an impossible value) to mark
499 * the fact that this object is involved in "write its base
500 * first before writing a deltified object" recursion.
501 */
502 recursing = (e->idx.offset == 1);
503 if (recursing) {
504 warning("recursive delta detected for object %s",
505 sha1_to_hex(e->idx.sha1));
506 return WRITE_ONE_RECURSIVE;
507 } else if (e->idx.offset || e->preferred_base) {
508 /* offset is non zero if object is written already. */
509 return WRITE_ONE_SKIP;
510 }
511
512 /* if we are deltified, write out base object first. */
513 if (e->delta) {
514 e->idx.offset = 1; /* now recurse */
515 switch (write_one(f, e->delta, offset)) {
516 case WRITE_ONE_RECURSIVE:
517 /* we cannot depend on this one */
518 e->delta = NULL;
519 break;
520 default:
521 break;
522 case WRITE_ONE_BREAK:
523 e->idx.offset = recursing;
524 return WRITE_ONE_BREAK;
525 }
526 }
527
528 e->idx.offset = *offset;
529 size = write_object(f, e, *offset);
530 if (!size) {
531 e->idx.offset = recursing;
532 return WRITE_ONE_BREAK;
533 }
534 written_list[nr_written++] = &e->idx;
535
536 /* make sure off_t is sufficiently large not to wrap */
537 if (signed_add_overflows(*offset, size))
538 die("pack too large for current definition of off_t");
539 *offset += size;
540 return WRITE_ONE_WRITTEN;
541}
542
543static int mark_tagged(const char *path, const struct object_id *oid, int flag,
544 void *cb_data)
545{
546 unsigned char peeled[20];
547 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
548
549 if (entry)
550 entry->tagged = 1;
551 if (!peel_ref(path, peeled)) {
552 entry = packlist_find(&to_pack, peeled, NULL);
553 if (entry)
554 entry->tagged = 1;
555 }
556 return 0;
557}
558
559static inline void add_to_write_order(struct object_entry **wo,
560 unsigned int *endp,
561 struct object_entry *e)
562{
563 if (e->filled)
564 return;
565 wo[(*endp)++] = e;
566 e->filled = 1;
567}
568
569static void add_descendants_to_write_order(struct object_entry **wo,
570 unsigned int *endp,
571 struct object_entry *e)
572{
573 int add_to_order = 1;
574 while (e) {
575 if (add_to_order) {
576 struct object_entry *s;
577 /* add this node... */
578 add_to_write_order(wo, endp, e);
579 /* all its siblings... */
580 for (s = e->delta_sibling; s; s = s->delta_sibling) {
581 add_to_write_order(wo, endp, s);
582 }
583 }
584 /* drop down a level to add left subtree nodes if possible */
585 if (e->delta_child) {
586 add_to_order = 1;
587 e = e->delta_child;
588 } else {
589 add_to_order = 0;
590 /* our sibling might have some children, it is next */
591 if (e->delta_sibling) {
592 e = e->delta_sibling;
593 continue;
594 }
595 /* go back to our parent node */
596 e = e->delta;
597 while (e && !e->delta_sibling) {
598 /* we're on the right side of a subtree, keep
599 * going up until we can go right again */
600 e = e->delta;
601 }
602 if (!e) {
603 /* done- we hit our original root node */
604 return;
605 }
606 /* pass it off to sibling at this level */
607 e = e->delta_sibling;
608 }
609 };
610}
611
612static void add_family_to_write_order(struct object_entry **wo,
613 unsigned int *endp,
614 struct object_entry *e)
615{
616 struct object_entry *root;
617
618 for (root = e; root->delta; root = root->delta)
619 ; /* nothing */
620 add_descendants_to_write_order(wo, endp, root);
621}
622
623static struct object_entry **compute_write_order(void)
624{
625 unsigned int i, wo_end, last_untagged;
626
627 struct object_entry **wo;
628 struct object_entry *objects = to_pack.objects;
629
630 for (i = 0; i < to_pack.nr_objects; i++) {
631 objects[i].tagged = 0;
632 objects[i].filled = 0;
633 objects[i].delta_child = NULL;
634 objects[i].delta_sibling = NULL;
635 }
636
637 /*
638 * Fully connect delta_child/delta_sibling network.
639 * Make sure delta_sibling is sorted in the original
640 * recency order.
641 */
642 for (i = to_pack.nr_objects; i > 0;) {
643 struct object_entry *e = &objects[--i];
644 if (!e->delta)
645 continue;
646 /* Mark me as the first child */
647 e->delta_sibling = e->delta->delta_child;
648 e->delta->delta_child = e;
649 }
650
651 /*
652 * Mark objects that are at the tip of tags.
653 */
654 for_each_tag_ref(mark_tagged, NULL);
655
656 /*
657 * Give the objects in the original recency order until
658 * we see a tagged tip.
659 */
660 ALLOC_ARRAY(wo, to_pack.nr_objects);
661 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
662 if (objects[i].tagged)
663 break;
664 add_to_write_order(wo, &wo_end, &objects[i]);
665 }
666 last_untagged = i;
667
668 /*
669 * Then fill all the tagged tips.
670 */
671 for (; i < to_pack.nr_objects; i++) {
672 if (objects[i].tagged)
673 add_to_write_order(wo, &wo_end, &objects[i]);
674 }
675
676 /*
677 * And then all remaining commits and tags.
678 */
679 for (i = last_untagged; i < to_pack.nr_objects; i++) {
680 if (objects[i].type != OBJ_COMMIT &&
681 objects[i].type != OBJ_TAG)
682 continue;
683 add_to_write_order(wo, &wo_end, &objects[i]);
684 }
685
686 /*
687 * And then all the trees.
688 */
689 for (i = last_untagged; i < to_pack.nr_objects; i++) {
690 if (objects[i].type != OBJ_TREE)
691 continue;
692 add_to_write_order(wo, &wo_end, &objects[i]);
693 }
694
695 /*
696 * Finally all the rest in really tight order
697 */
698 for (i = last_untagged; i < to_pack.nr_objects; i++) {
699 if (!objects[i].filled)
700 add_family_to_write_order(wo, &wo_end, &objects[i]);
701 }
702
703 if (wo_end != to_pack.nr_objects)
704 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
705
706 return wo;
707}
708
709static off_t write_reused_pack(struct sha1file *f)
710{
711 unsigned char buffer[8192];
712 off_t to_write, total;
713 int fd;
714
715 if (!is_pack_valid(reuse_packfile))
716 die("packfile is invalid: %s", reuse_packfile->pack_name);
717
718 fd = git_open_noatime(reuse_packfile->pack_name);
719 if (fd < 0)
720 die_errno("unable to open packfile for reuse: %s",
721 reuse_packfile->pack_name);
722
723 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
724 die_errno("unable to seek in reused packfile");
725
726 if (reuse_packfile_offset < 0)
727 reuse_packfile_offset = reuse_packfile->pack_size - 20;
728
729 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
730
731 while (to_write) {
732 int read_pack = xread(fd, buffer, sizeof(buffer));
733
734 if (read_pack <= 0)
735 die_errno("unable to read from reused packfile");
736
737 if (read_pack > to_write)
738 read_pack = to_write;
739
740 sha1write(f, buffer, read_pack);
741 to_write -= read_pack;
742
743 /*
744 * We don't know the actual number of objects written,
745 * only how many bytes written, how many bytes total, and
746 * how many objects total. So we can fake it by pretending all
747 * objects we are writing are the same size. This gives us a
748 * smooth progress meter, and at the end it matches the true
749 * answer.
750 */
751 written = reuse_packfile_objects *
752 (((double)(total - to_write)) / total);
753 display_progress(progress_state, written);
754 }
755
756 close(fd);
757 written = reuse_packfile_objects;
758 display_progress(progress_state, written);
759 return reuse_packfile_offset - sizeof(struct pack_header);
760}
761
762static const char no_split_warning[] = N_(
763"disabling bitmap writing, packs are split due to pack.packSizeLimit"
764);
765
766static void write_pack_file(void)
767{
768 uint32_t i = 0, j;
769 struct sha1file *f;
770 off_t offset;
771 uint32_t nr_remaining = nr_result;
772 time_t last_mtime = 0;
773 struct object_entry **write_order;
774
775 if (progress > pack_to_stdout)
776 progress_state = start_progress(_("Writing objects"), nr_result);
777 ALLOC_ARRAY(written_list, to_pack.nr_objects);
778 write_order = compute_write_order();
779
780 do {
781 unsigned char sha1[20];
782 char *pack_tmp_name = NULL;
783
784 if (pack_to_stdout)
785 f = sha1fd_throughput(1, "<stdout>", progress_state);
786 else
787 f = create_tmp_packfile(&pack_tmp_name);
788
789 offset = write_pack_header(f, nr_remaining);
790
791 if (reuse_packfile) {
792 off_t packfile_size;
793 assert(pack_to_stdout);
794
795 packfile_size = write_reused_pack(f);
796 offset += packfile_size;
797 }
798
799 nr_written = 0;
800 for (; i < to_pack.nr_objects; i++) {
801 struct object_entry *e = write_order[i];
802 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
803 break;
804 display_progress(progress_state, written);
805 }
806
807 /*
808 * Did we write the wrong # entries in the header?
809 * If so, rewrite it like in fast-import
810 */
811 if (pack_to_stdout) {
812 sha1close(f, sha1, CSUM_CLOSE);
813 } else if (nr_written == nr_remaining) {
814 sha1close(f, sha1, CSUM_FSYNC);
815 } else {
816 int fd = sha1close(f, sha1, 0);
817 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
818 nr_written, sha1, offset);
819 close(fd);
820 if (write_bitmap_index) {
821 warning(_(no_split_warning));
822 write_bitmap_index = 0;
823 }
824 }
825
826 if (!pack_to_stdout) {
827 struct stat st;
828 struct strbuf tmpname = STRBUF_INIT;
829
830 /*
831 * Packs are runtime accessed in their mtime
832 * order since newer packs are more likely to contain
833 * younger objects. So if we are creating multiple
834 * packs then we should modify the mtime of later ones
835 * to preserve this property.
836 */
837 if (stat(pack_tmp_name, &st) < 0) {
838 warning_errno("failed to stat %s", pack_tmp_name);
839 } else if (!last_mtime) {
840 last_mtime = st.st_mtime;
841 } else {
842 struct utimbuf utb;
843 utb.actime = st.st_atime;
844 utb.modtime = --last_mtime;
845 if (utime(pack_tmp_name, &utb) < 0)
846 warning_errno("failed utime() on %s", pack_tmp_name);
847 }
848
849 strbuf_addf(&tmpname, "%s-", base_name);
850
851 if (write_bitmap_index) {
852 bitmap_writer_set_checksum(sha1);
853 bitmap_writer_build_type_index(written_list, nr_written);
854 }
855
856 finish_tmp_packfile(&tmpname, pack_tmp_name,
857 written_list, nr_written,
858 &pack_idx_opts, sha1);
859
860 if (write_bitmap_index) {
861 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
862
863 stop_progress(&progress_state);
864
865 bitmap_writer_show_progress(progress);
866 bitmap_writer_reuse_bitmaps(&to_pack);
867 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
868 bitmap_writer_build(&to_pack);
869 bitmap_writer_finish(written_list, nr_written,
870 tmpname.buf, write_bitmap_options);
871 write_bitmap_index = 0;
872 }
873
874 strbuf_release(&tmpname);
875 free(pack_tmp_name);
876 puts(sha1_to_hex(sha1));
877 }
878
879 /* mark written objects as written to previous pack */
880 for (j = 0; j < nr_written; j++) {
881 written_list[j]->offset = (off_t)-1;
882 }
883 nr_remaining -= nr_written;
884 } while (nr_remaining && i < to_pack.nr_objects);
885
886 free(written_list);
887 free(write_order);
888 stop_progress(&progress_state);
889 if (written != nr_result)
890 die("wrote %"PRIu32" objects while expecting %"PRIu32,
891 written, nr_result);
892}
893
894static void setup_delta_attr_check(struct git_attr_check *check)
895{
896 static struct git_attr *attr_delta;
897
898 if (!attr_delta)
899 attr_delta = git_attr("delta");
900
901 check[0].attr = attr_delta;
902}
903
904static int no_try_delta(const char *path)
905{
906 struct git_attr_check check[1];
907
908 setup_delta_attr_check(check);
909 if (git_check_attr(path, ARRAY_SIZE(check), check))
910 return 0;
911 if (ATTR_FALSE(check->value))
912 return 1;
913 return 0;
914}
915
916/*
917 * When adding an object, check whether we have already added it
918 * to our packing list. If so, we can skip. However, if we are
919 * being asked to excludei t, but the previous mention was to include
920 * it, make sure to adjust its flags and tweak our numbers accordingly.
921 *
922 * As an optimization, we pass out the index position where we would have
923 * found the item, since that saves us from having to look it up again a
924 * few lines later when we want to add the new entry.
925 */
926static int have_duplicate_entry(const unsigned char *sha1,
927 int exclude,
928 uint32_t *index_pos)
929{
930 struct object_entry *entry;
931
932 entry = packlist_find(&to_pack, sha1, index_pos);
933 if (!entry)
934 return 0;
935
936 if (exclude) {
937 if (!entry->preferred_base)
938 nr_result--;
939 entry->preferred_base = 1;
940 }
941
942 return 1;
943}
944
945/*
946 * Check whether we want the object in the pack (e.g., we do not want
947 * objects found in non-local stores if the "--local" option was used).
948 *
949 * As a side effect of this check, we will find the packed version of this
950 * object, if any. We therefore pass out the pack information to avoid having
951 * to look it up again later.
952 */
953static int want_object_in_pack(const unsigned char *sha1,
954 int exclude,
955 struct packed_git **found_pack,
956 off_t *found_offset)
957{
958 struct packed_git *p;
959
960 if (!exclude && local && has_loose_object_nonlocal(sha1))
961 return 0;
962
963 *found_pack = NULL;
964 *found_offset = 0;
965
966 for (p = packed_git; p; p = p->next) {
967 off_t offset = find_pack_entry_one(sha1, p);
968 if (offset) {
969 if (!*found_pack) {
970 if (!is_pack_valid(p))
971 continue;
972 *found_offset = offset;
973 *found_pack = p;
974 }
975 if (exclude)
976 return 1;
977 if (incremental)
978 return 0;
979 if (local && !p->pack_local)
980 return 0;
981 if (ignore_packed_keep && p->pack_local && p->pack_keep)
982 return 0;
983 }
984 }
985
986 return 1;
987}
988
989static void create_object_entry(const unsigned char *sha1,
990 enum object_type type,
991 uint32_t hash,
992 int exclude,
993 int no_try_delta,
994 uint32_t index_pos,
995 struct packed_git *found_pack,
996 off_t found_offset)
997{
998 struct object_entry *entry;
999
1000 entry = packlist_alloc(&to_pack, sha1, index_pos);
1001 entry->hash = hash;
1002 if (type)
1003 entry->type = type;
1004 if (exclude)
1005 entry->preferred_base = 1;
1006 else
1007 nr_result++;
1008 if (found_pack) {
1009 entry->in_pack = found_pack;
1010 entry->in_pack_offset = found_offset;
1011 }
1012
1013 entry->no_try_delta = no_try_delta;
1014}
1015
1016static const char no_closure_warning[] = N_(
1017"disabling bitmap writing, as some objects are not being packed"
1018);
1019
1020static int add_object_entry(const unsigned char *sha1, enum object_type type,
1021 const char *name, int exclude)
1022{
1023 struct packed_git *found_pack;
1024 off_t found_offset;
1025 uint32_t index_pos;
1026
1027 if (have_duplicate_entry(sha1, exclude, &index_pos))
1028 return 0;
1029
1030 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1031 /* The pack is missing an object, so it will not have closure */
1032 if (write_bitmap_index) {
1033 warning(_(no_closure_warning));
1034 write_bitmap_index = 0;
1035 }
1036 return 0;
1037 }
1038
1039 create_object_entry(sha1, type, pack_name_hash(name),
1040 exclude, name && no_try_delta(name),
1041 index_pos, found_pack, found_offset);
1042
1043 display_progress(progress_state, nr_result);
1044 return 1;
1045}
1046
1047static int add_object_entry_from_bitmap(const unsigned char *sha1,
1048 enum object_type type,
1049 int flags, uint32_t name_hash,
1050 struct packed_git *pack, off_t offset)
1051{
1052 uint32_t index_pos;
1053
1054 if (have_duplicate_entry(sha1, 0, &index_pos))
1055 return 0;
1056
1057 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1058
1059 display_progress(progress_state, nr_result);
1060 return 1;
1061}
1062
1063struct pbase_tree_cache {
1064 unsigned char sha1[20];
1065 int ref;
1066 int temporary;
1067 void *tree_data;
1068 unsigned long tree_size;
1069};
1070
1071static struct pbase_tree_cache *(pbase_tree_cache[256]);
1072static int pbase_tree_cache_ix(const unsigned char *sha1)
1073{
1074 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1075}
1076static int pbase_tree_cache_ix_incr(int ix)
1077{
1078 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1079}
1080
1081static struct pbase_tree {
1082 struct pbase_tree *next;
1083 /* This is a phony "cache" entry; we are not
1084 * going to evict it or find it through _get()
1085 * mechanism -- this is for the toplevel node that
1086 * would almost always change with any commit.
1087 */
1088 struct pbase_tree_cache pcache;
1089} *pbase_tree;
1090
1091static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1092{
1093 struct pbase_tree_cache *ent, *nent;
1094 void *data;
1095 unsigned long size;
1096 enum object_type type;
1097 int neigh;
1098 int my_ix = pbase_tree_cache_ix(sha1);
1099 int available_ix = -1;
1100
1101 /* pbase-tree-cache acts as a limited hashtable.
1102 * your object will be found at your index or within a few
1103 * slots after that slot if it is cached.
1104 */
1105 for (neigh = 0; neigh < 8; neigh++) {
1106 ent = pbase_tree_cache[my_ix];
1107 if (ent && !hashcmp(ent->sha1, sha1)) {
1108 ent->ref++;
1109 return ent;
1110 }
1111 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1112 ((0 <= available_ix) &&
1113 (!ent && pbase_tree_cache[available_ix])))
1114 available_ix = my_ix;
1115 if (!ent)
1116 break;
1117 my_ix = pbase_tree_cache_ix_incr(my_ix);
1118 }
1119
1120 /* Did not find one. Either we got a bogus request or
1121 * we need to read and perhaps cache.
1122 */
1123 data = read_sha1_file(sha1, &type, &size);
1124 if (!data)
1125 return NULL;
1126 if (type != OBJ_TREE) {
1127 free(data);
1128 return NULL;
1129 }
1130
1131 /* We need to either cache or return a throwaway copy */
1132
1133 if (available_ix < 0)
1134 ent = NULL;
1135 else {
1136 ent = pbase_tree_cache[available_ix];
1137 my_ix = available_ix;
1138 }
1139
1140 if (!ent) {
1141 nent = xmalloc(sizeof(*nent));
1142 nent->temporary = (available_ix < 0);
1143 }
1144 else {
1145 /* evict and reuse */
1146 free(ent->tree_data);
1147 nent = ent;
1148 }
1149 hashcpy(nent->sha1, sha1);
1150 nent->tree_data = data;
1151 nent->tree_size = size;
1152 nent->ref = 1;
1153 if (!nent->temporary)
1154 pbase_tree_cache[my_ix] = nent;
1155 return nent;
1156}
1157
1158static void pbase_tree_put(struct pbase_tree_cache *cache)
1159{
1160 if (!cache->temporary) {
1161 cache->ref--;
1162 return;
1163 }
1164 free(cache->tree_data);
1165 free(cache);
1166}
1167
1168static int name_cmp_len(const char *name)
1169{
1170 int i;
1171 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1172 ;
1173 return i;
1174}
1175
1176static void add_pbase_object(struct tree_desc *tree,
1177 const char *name,
1178 int cmplen,
1179 const char *fullname)
1180{
1181 struct name_entry entry;
1182 int cmp;
1183
1184 while (tree_entry(tree,&entry)) {
1185 if (S_ISGITLINK(entry.mode))
1186 continue;
1187 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1188 memcmp(name, entry.path, cmplen);
1189 if (cmp > 0)
1190 continue;
1191 if (cmp < 0)
1192 return;
1193 if (name[cmplen] != '/') {
1194 add_object_entry(entry.oid->hash,
1195 object_type(entry.mode),
1196 fullname, 1);
1197 return;
1198 }
1199 if (S_ISDIR(entry.mode)) {
1200 struct tree_desc sub;
1201 struct pbase_tree_cache *tree;
1202 const char *down = name+cmplen+1;
1203 int downlen = name_cmp_len(down);
1204
1205 tree = pbase_tree_get(entry.oid->hash);
1206 if (!tree)
1207 return;
1208 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1209
1210 add_pbase_object(&sub, down, downlen, fullname);
1211 pbase_tree_put(tree);
1212 }
1213 }
1214}
1215
1216static unsigned *done_pbase_paths;
1217static int done_pbase_paths_num;
1218static int done_pbase_paths_alloc;
1219static int done_pbase_path_pos(unsigned hash)
1220{
1221 int lo = 0;
1222 int hi = done_pbase_paths_num;
1223 while (lo < hi) {
1224 int mi = (hi + lo) / 2;
1225 if (done_pbase_paths[mi] == hash)
1226 return mi;
1227 if (done_pbase_paths[mi] < hash)
1228 hi = mi;
1229 else
1230 lo = mi + 1;
1231 }
1232 return -lo-1;
1233}
1234
1235static int check_pbase_path(unsigned hash)
1236{
1237 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1238 if (0 <= pos)
1239 return 1;
1240 pos = -pos - 1;
1241 ALLOC_GROW(done_pbase_paths,
1242 done_pbase_paths_num + 1,
1243 done_pbase_paths_alloc);
1244 done_pbase_paths_num++;
1245 if (pos < done_pbase_paths_num)
1246 memmove(done_pbase_paths + pos + 1,
1247 done_pbase_paths + pos,
1248 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1249 done_pbase_paths[pos] = hash;
1250 return 0;
1251}
1252
1253static void add_preferred_base_object(const char *name)
1254{
1255 struct pbase_tree *it;
1256 int cmplen;
1257 unsigned hash = pack_name_hash(name);
1258
1259 if (!num_preferred_base || check_pbase_path(hash))
1260 return;
1261
1262 cmplen = name_cmp_len(name);
1263 for (it = pbase_tree; it; it = it->next) {
1264 if (cmplen == 0) {
1265 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1266 }
1267 else {
1268 struct tree_desc tree;
1269 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1270 add_pbase_object(&tree, name, cmplen, name);
1271 }
1272 }
1273}
1274
1275static void add_preferred_base(unsigned char *sha1)
1276{
1277 struct pbase_tree *it;
1278 void *data;
1279 unsigned long size;
1280 unsigned char tree_sha1[20];
1281
1282 if (window <= num_preferred_base++)
1283 return;
1284
1285 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1286 if (!data)
1287 return;
1288
1289 for (it = pbase_tree; it; it = it->next) {
1290 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1291 free(data);
1292 return;
1293 }
1294 }
1295
1296 it = xcalloc(1, sizeof(*it));
1297 it->next = pbase_tree;
1298 pbase_tree = it;
1299
1300 hashcpy(it->pcache.sha1, tree_sha1);
1301 it->pcache.tree_data = data;
1302 it->pcache.tree_size = size;
1303}
1304
1305static void cleanup_preferred_base(void)
1306{
1307 struct pbase_tree *it;
1308 unsigned i;
1309
1310 it = pbase_tree;
1311 pbase_tree = NULL;
1312 while (it) {
1313 struct pbase_tree *this = it;
1314 it = this->next;
1315 free(this->pcache.tree_data);
1316 free(this);
1317 }
1318
1319 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1320 if (!pbase_tree_cache[i])
1321 continue;
1322 free(pbase_tree_cache[i]->tree_data);
1323 free(pbase_tree_cache[i]);
1324 pbase_tree_cache[i] = NULL;
1325 }
1326
1327 free(done_pbase_paths);
1328 done_pbase_paths = NULL;
1329 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1330}
1331
1332static void check_object(struct object_entry *entry)
1333{
1334 if (entry->in_pack) {
1335 struct packed_git *p = entry->in_pack;
1336 struct pack_window *w_curs = NULL;
1337 const unsigned char *base_ref = NULL;
1338 struct object_entry *base_entry;
1339 unsigned long used, used_0;
1340 unsigned long avail;
1341 off_t ofs;
1342 unsigned char *buf, c;
1343
1344 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1345
1346 /*
1347 * We want in_pack_type even if we do not reuse delta
1348 * since non-delta representations could still be reused.
1349 */
1350 used = unpack_object_header_buffer(buf, avail,
1351 &entry->in_pack_type,
1352 &entry->size);
1353 if (used == 0)
1354 goto give_up;
1355
1356 /*
1357 * Determine if this is a delta and if so whether we can
1358 * reuse it or not. Otherwise let's find out as cheaply as
1359 * possible what the actual type and size for this object is.
1360 */
1361 switch (entry->in_pack_type) {
1362 default:
1363 /* Not a delta hence we've already got all we need. */
1364 entry->type = entry->in_pack_type;
1365 entry->in_pack_header_size = used;
1366 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1367 goto give_up;
1368 unuse_pack(&w_curs);
1369 return;
1370 case OBJ_REF_DELTA:
1371 if (reuse_delta && !entry->preferred_base)
1372 base_ref = use_pack(p, &w_curs,
1373 entry->in_pack_offset + used, NULL);
1374 entry->in_pack_header_size = used + 20;
1375 break;
1376 case OBJ_OFS_DELTA:
1377 buf = use_pack(p, &w_curs,
1378 entry->in_pack_offset + used, NULL);
1379 used_0 = 0;
1380 c = buf[used_0++];
1381 ofs = c & 127;
1382 while (c & 128) {
1383 ofs += 1;
1384 if (!ofs || MSB(ofs, 7)) {
1385 error("delta base offset overflow in pack for %s",
1386 sha1_to_hex(entry->idx.sha1));
1387 goto give_up;
1388 }
1389 c = buf[used_0++];
1390 ofs = (ofs << 7) + (c & 127);
1391 }
1392 ofs = entry->in_pack_offset - ofs;
1393 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1394 error("delta base offset out of bound for %s",
1395 sha1_to_hex(entry->idx.sha1));
1396 goto give_up;
1397 }
1398 if (reuse_delta && !entry->preferred_base) {
1399 struct revindex_entry *revidx;
1400 revidx = find_pack_revindex(p, ofs);
1401 if (!revidx)
1402 goto give_up;
1403 base_ref = nth_packed_object_sha1(p, revidx->nr);
1404 }
1405 entry->in_pack_header_size = used + used_0;
1406 break;
1407 }
1408
1409 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1410 /*
1411 * If base_ref was set above that means we wish to
1412 * reuse delta data, and we even found that base
1413 * in the list of objects we want to pack. Goodie!
1414 *
1415 * Depth value does not matter - find_deltas() will
1416 * never consider reused delta as the base object to
1417 * deltify other objects against, in order to avoid
1418 * circular deltas.
1419 */
1420 entry->type = entry->in_pack_type;
1421 entry->delta = base_entry;
1422 entry->delta_size = entry->size;
1423 entry->delta_sibling = base_entry->delta_child;
1424 base_entry->delta_child = entry;
1425 unuse_pack(&w_curs);
1426 return;
1427 }
1428
1429 if (entry->type) {
1430 /*
1431 * This must be a delta and we already know what the
1432 * final object type is. Let's extract the actual
1433 * object size from the delta header.
1434 */
1435 entry->size = get_size_from_delta(p, &w_curs,
1436 entry->in_pack_offset + entry->in_pack_header_size);
1437 if (entry->size == 0)
1438 goto give_up;
1439 unuse_pack(&w_curs);
1440 return;
1441 }
1442
1443 /*
1444 * No choice but to fall back to the recursive delta walk
1445 * with sha1_object_info() to find about the object type
1446 * at this point...
1447 */
1448 give_up:
1449 unuse_pack(&w_curs);
1450 }
1451
1452 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1453 /*
1454 * The error condition is checked in prepare_pack(). This is
1455 * to permit a missing preferred base object to be ignored
1456 * as a preferred base. Doing so can result in a larger
1457 * pack file, but the transfer will still take place.
1458 */
1459}
1460
1461static int pack_offset_sort(const void *_a, const void *_b)
1462{
1463 const struct object_entry *a = *(struct object_entry **)_a;
1464 const struct object_entry *b = *(struct object_entry **)_b;
1465
1466 /* avoid filesystem trashing with loose objects */
1467 if (!a->in_pack && !b->in_pack)
1468 return hashcmp(a->idx.sha1, b->idx.sha1);
1469
1470 if (a->in_pack < b->in_pack)
1471 return -1;
1472 if (a->in_pack > b->in_pack)
1473 return 1;
1474 return a->in_pack_offset < b->in_pack_offset ? -1 :
1475 (a->in_pack_offset > b->in_pack_offset);
1476}
1477
1478static void get_object_details(void)
1479{
1480 uint32_t i;
1481 struct object_entry **sorted_by_offset;
1482
1483 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1484 for (i = 0; i < to_pack.nr_objects; i++)
1485 sorted_by_offset[i] = to_pack.objects + i;
1486 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1487
1488 for (i = 0; i < to_pack.nr_objects; i++) {
1489 struct object_entry *entry = sorted_by_offset[i];
1490 check_object(entry);
1491 if (big_file_threshold < entry->size)
1492 entry->no_try_delta = 1;
1493 }
1494
1495 free(sorted_by_offset);
1496}
1497
1498/*
1499 * We search for deltas in a list sorted by type, by filename hash, and then
1500 * by size, so that we see progressively smaller and smaller files.
1501 * That's because we prefer deltas to be from the bigger file
1502 * to the smaller -- deletes are potentially cheaper, but perhaps
1503 * more importantly, the bigger file is likely the more recent
1504 * one. The deepest deltas are therefore the oldest objects which are
1505 * less susceptible to be accessed often.
1506 */
1507static int type_size_sort(const void *_a, const void *_b)
1508{
1509 const struct object_entry *a = *(struct object_entry **)_a;
1510 const struct object_entry *b = *(struct object_entry **)_b;
1511
1512 if (a->type > b->type)
1513 return -1;
1514 if (a->type < b->type)
1515 return 1;
1516 if (a->hash > b->hash)
1517 return -1;
1518 if (a->hash < b->hash)
1519 return 1;
1520 if (a->preferred_base > b->preferred_base)
1521 return -1;
1522 if (a->preferred_base < b->preferred_base)
1523 return 1;
1524 if (a->size > b->size)
1525 return -1;
1526 if (a->size < b->size)
1527 return 1;
1528 return a < b ? -1 : (a > b); /* newest first */
1529}
1530
1531struct unpacked {
1532 struct object_entry *entry;
1533 void *data;
1534 struct delta_index *index;
1535 unsigned depth;
1536};
1537
1538static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1539 unsigned long delta_size)
1540{
1541 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1542 return 0;
1543
1544 if (delta_size < cache_max_small_delta_size)
1545 return 1;
1546
1547 /* cache delta, if objects are large enough compared to delta size */
1548 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1549 return 1;
1550
1551 return 0;
1552}
1553
1554#ifndef NO_PTHREADS
1555
1556static pthread_mutex_t read_mutex;
1557#define read_lock() pthread_mutex_lock(&read_mutex)
1558#define read_unlock() pthread_mutex_unlock(&read_mutex)
1559
1560static pthread_mutex_t cache_mutex;
1561#define cache_lock() pthread_mutex_lock(&cache_mutex)
1562#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1563
1564static pthread_mutex_t progress_mutex;
1565#define progress_lock() pthread_mutex_lock(&progress_mutex)
1566#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1567
1568#else
1569
1570#define read_lock() (void)0
1571#define read_unlock() (void)0
1572#define cache_lock() (void)0
1573#define cache_unlock() (void)0
1574#define progress_lock() (void)0
1575#define progress_unlock() (void)0
1576
1577#endif
1578
1579static int try_delta(struct unpacked *trg, struct unpacked *src,
1580 unsigned max_depth, unsigned long *mem_usage)
1581{
1582 struct object_entry *trg_entry = trg->entry;
1583 struct object_entry *src_entry = src->entry;
1584 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1585 unsigned ref_depth;
1586 enum object_type type;
1587 void *delta_buf;
1588
1589 /* Don't bother doing diffs between different types */
1590 if (trg_entry->type != src_entry->type)
1591 return -1;
1592
1593 /*
1594 * We do not bother to try a delta that we discarded on an
1595 * earlier try, but only when reusing delta data. Note that
1596 * src_entry that is marked as the preferred_base should always
1597 * be considered, as even if we produce a suboptimal delta against
1598 * it, we will still save the transfer cost, as we already know
1599 * the other side has it and we won't send src_entry at all.
1600 */
1601 if (reuse_delta && trg_entry->in_pack &&
1602 trg_entry->in_pack == src_entry->in_pack &&
1603 !src_entry->preferred_base &&
1604 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1605 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1606 return 0;
1607
1608 /* Let's not bust the allowed depth. */
1609 if (src->depth >= max_depth)
1610 return 0;
1611
1612 /* Now some size filtering heuristics. */
1613 trg_size = trg_entry->size;
1614 if (!trg_entry->delta) {
1615 max_size = trg_size/2 - 20;
1616 ref_depth = 1;
1617 } else {
1618 max_size = trg_entry->delta_size;
1619 ref_depth = trg->depth;
1620 }
1621 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1622 (max_depth - ref_depth + 1);
1623 if (max_size == 0)
1624 return 0;
1625 src_size = src_entry->size;
1626 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1627 if (sizediff >= max_size)
1628 return 0;
1629 if (trg_size < src_size / 32)
1630 return 0;
1631
1632 /* Load data if not already done */
1633 if (!trg->data) {
1634 read_lock();
1635 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1636 read_unlock();
1637 if (!trg->data)
1638 die("object %s cannot be read",
1639 sha1_to_hex(trg_entry->idx.sha1));
1640 if (sz != trg_size)
1641 die("object %s inconsistent object length (%lu vs %lu)",
1642 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1643 *mem_usage += sz;
1644 }
1645 if (!src->data) {
1646 read_lock();
1647 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1648 read_unlock();
1649 if (!src->data) {
1650 if (src_entry->preferred_base) {
1651 static int warned = 0;
1652 if (!warned++)
1653 warning("object %s cannot be read",
1654 sha1_to_hex(src_entry->idx.sha1));
1655 /*
1656 * Those objects are not included in the
1657 * resulting pack. Be resilient and ignore
1658 * them if they can't be read, in case the
1659 * pack could be created nevertheless.
1660 */
1661 return 0;
1662 }
1663 die("object %s cannot be read",
1664 sha1_to_hex(src_entry->idx.sha1));
1665 }
1666 if (sz != src_size)
1667 die("object %s inconsistent object length (%lu vs %lu)",
1668 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1669 *mem_usage += sz;
1670 }
1671 if (!src->index) {
1672 src->index = create_delta_index(src->data, src_size);
1673 if (!src->index) {
1674 static int warned = 0;
1675 if (!warned++)
1676 warning("suboptimal pack - out of memory");
1677 return 0;
1678 }
1679 *mem_usage += sizeof_delta_index(src->index);
1680 }
1681
1682 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1683 if (!delta_buf)
1684 return 0;
1685
1686 if (trg_entry->delta) {
1687 /* Prefer only shallower same-sized deltas. */
1688 if (delta_size == trg_entry->delta_size &&
1689 src->depth + 1 >= trg->depth) {
1690 free(delta_buf);
1691 return 0;
1692 }
1693 }
1694
1695 /*
1696 * Handle memory allocation outside of the cache
1697 * accounting lock. Compiler will optimize the strangeness
1698 * away when NO_PTHREADS is defined.
1699 */
1700 free(trg_entry->delta_data);
1701 cache_lock();
1702 if (trg_entry->delta_data) {
1703 delta_cache_size -= trg_entry->delta_size;
1704 trg_entry->delta_data = NULL;
1705 }
1706 if (delta_cacheable(src_size, trg_size, delta_size)) {
1707 delta_cache_size += delta_size;
1708 cache_unlock();
1709 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1710 } else {
1711 cache_unlock();
1712 free(delta_buf);
1713 }
1714
1715 trg_entry->delta = src_entry;
1716 trg_entry->delta_size = delta_size;
1717 trg->depth = src->depth + 1;
1718
1719 return 1;
1720}
1721
1722static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1723{
1724 struct object_entry *child = me->delta_child;
1725 unsigned int m = n;
1726 while (child) {
1727 unsigned int c = check_delta_limit(child, n + 1);
1728 if (m < c)
1729 m = c;
1730 child = child->delta_sibling;
1731 }
1732 return m;
1733}
1734
1735static unsigned long free_unpacked(struct unpacked *n)
1736{
1737 unsigned long freed_mem = sizeof_delta_index(n->index);
1738 free_delta_index(n->index);
1739 n->index = NULL;
1740 if (n->data) {
1741 freed_mem += n->entry->size;
1742 free(n->data);
1743 n->data = NULL;
1744 }
1745 n->entry = NULL;
1746 n->depth = 0;
1747 return freed_mem;
1748}
1749
1750static void find_deltas(struct object_entry **list, unsigned *list_size,
1751 int window, int depth, unsigned *processed)
1752{
1753 uint32_t i, idx = 0, count = 0;
1754 struct unpacked *array;
1755 unsigned long mem_usage = 0;
1756
1757 array = xcalloc(window, sizeof(struct unpacked));
1758
1759 for (;;) {
1760 struct object_entry *entry;
1761 struct unpacked *n = array + idx;
1762 int j, max_depth, best_base = -1;
1763
1764 progress_lock();
1765 if (!*list_size) {
1766 progress_unlock();
1767 break;
1768 }
1769 entry = *list++;
1770 (*list_size)--;
1771 if (!entry->preferred_base) {
1772 (*processed)++;
1773 display_progress(progress_state, *processed);
1774 }
1775 progress_unlock();
1776
1777 mem_usage -= free_unpacked(n);
1778 n->entry = entry;
1779
1780 while (window_memory_limit &&
1781 mem_usage > window_memory_limit &&
1782 count > 1) {
1783 uint32_t tail = (idx + window - count) % window;
1784 mem_usage -= free_unpacked(array + tail);
1785 count--;
1786 }
1787
1788 /* We do not compute delta to *create* objects we are not
1789 * going to pack.
1790 */
1791 if (entry->preferred_base)
1792 goto next;
1793
1794 /*
1795 * If the current object is at pack edge, take the depth the
1796 * objects that depend on the current object into account
1797 * otherwise they would become too deep.
1798 */
1799 max_depth = depth;
1800 if (entry->delta_child) {
1801 max_depth -= check_delta_limit(entry, 0);
1802 if (max_depth <= 0)
1803 goto next;
1804 }
1805
1806 j = window;
1807 while (--j > 0) {
1808 int ret;
1809 uint32_t other_idx = idx + j;
1810 struct unpacked *m;
1811 if (other_idx >= window)
1812 other_idx -= window;
1813 m = array + other_idx;
1814 if (!m->entry)
1815 break;
1816 ret = try_delta(n, m, max_depth, &mem_usage);
1817 if (ret < 0)
1818 break;
1819 else if (ret > 0)
1820 best_base = other_idx;
1821 }
1822
1823 /*
1824 * If we decided to cache the delta data, then it is best
1825 * to compress it right away. First because we have to do
1826 * it anyway, and doing it here while we're threaded will
1827 * save a lot of time in the non threaded write phase,
1828 * as well as allow for caching more deltas within
1829 * the same cache size limit.
1830 * ...
1831 * But only if not writing to stdout, since in that case
1832 * the network is most likely throttling writes anyway,
1833 * and therefore it is best to go to the write phase ASAP
1834 * instead, as we can afford spending more time compressing
1835 * between writes at that moment.
1836 */
1837 if (entry->delta_data && !pack_to_stdout) {
1838 entry->z_delta_size = do_compress(&entry->delta_data,
1839 entry->delta_size);
1840 cache_lock();
1841 delta_cache_size -= entry->delta_size;
1842 delta_cache_size += entry->z_delta_size;
1843 cache_unlock();
1844 }
1845
1846 /* if we made n a delta, and if n is already at max
1847 * depth, leaving it in the window is pointless. we
1848 * should evict it first.
1849 */
1850 if (entry->delta && max_depth <= n->depth)
1851 continue;
1852
1853 /*
1854 * Move the best delta base up in the window, after the
1855 * currently deltified object, to keep it longer. It will
1856 * be the first base object to be attempted next.
1857 */
1858 if (entry->delta) {
1859 struct unpacked swap = array[best_base];
1860 int dist = (window + idx - best_base) % window;
1861 int dst = best_base;
1862 while (dist--) {
1863 int src = (dst + 1) % window;
1864 array[dst] = array[src];
1865 dst = src;
1866 }
1867 array[dst] = swap;
1868 }
1869
1870 next:
1871 idx++;
1872 if (count + 1 < window)
1873 count++;
1874 if (idx >= window)
1875 idx = 0;
1876 }
1877
1878 for (i = 0; i < window; ++i) {
1879 free_delta_index(array[i].index);
1880 free(array[i].data);
1881 }
1882 free(array);
1883}
1884
1885#ifndef NO_PTHREADS
1886
1887static void try_to_free_from_threads(size_t size)
1888{
1889 read_lock();
1890 release_pack_memory(size);
1891 read_unlock();
1892}
1893
1894static try_to_free_t old_try_to_free_routine;
1895
1896/*
1897 * The main thread waits on the condition that (at least) one of the workers
1898 * has stopped working (which is indicated in the .working member of
1899 * struct thread_params).
1900 * When a work thread has completed its work, it sets .working to 0 and
1901 * signals the main thread and waits on the condition that .data_ready
1902 * becomes 1.
1903 */
1904
1905struct thread_params {
1906 pthread_t thread;
1907 struct object_entry **list;
1908 unsigned list_size;
1909 unsigned remaining;
1910 int window;
1911 int depth;
1912 int working;
1913 int data_ready;
1914 pthread_mutex_t mutex;
1915 pthread_cond_t cond;
1916 unsigned *processed;
1917};
1918
1919static pthread_cond_t progress_cond;
1920
1921/*
1922 * Mutex and conditional variable can't be statically-initialized on Windows.
1923 */
1924static void init_threaded_search(void)
1925{
1926 init_recursive_mutex(&read_mutex);
1927 pthread_mutex_init(&cache_mutex, NULL);
1928 pthread_mutex_init(&progress_mutex, NULL);
1929 pthread_cond_init(&progress_cond, NULL);
1930 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1931}
1932
1933static void cleanup_threaded_search(void)
1934{
1935 set_try_to_free_routine(old_try_to_free_routine);
1936 pthread_cond_destroy(&progress_cond);
1937 pthread_mutex_destroy(&read_mutex);
1938 pthread_mutex_destroy(&cache_mutex);
1939 pthread_mutex_destroy(&progress_mutex);
1940}
1941
1942static void *threaded_find_deltas(void *arg)
1943{
1944 struct thread_params *me = arg;
1945
1946 while (me->remaining) {
1947 find_deltas(me->list, &me->remaining,
1948 me->window, me->depth, me->processed);
1949
1950 progress_lock();
1951 me->working = 0;
1952 pthread_cond_signal(&progress_cond);
1953 progress_unlock();
1954
1955 /*
1956 * We must not set ->data_ready before we wait on the
1957 * condition because the main thread may have set it to 1
1958 * before we get here. In order to be sure that new
1959 * work is available if we see 1 in ->data_ready, it
1960 * was initialized to 0 before this thread was spawned
1961 * and we reset it to 0 right away.
1962 */
1963 pthread_mutex_lock(&me->mutex);
1964 while (!me->data_ready)
1965 pthread_cond_wait(&me->cond, &me->mutex);
1966 me->data_ready = 0;
1967 pthread_mutex_unlock(&me->mutex);
1968 }
1969 /* leave ->working 1 so that this doesn't get more work assigned */
1970 return NULL;
1971}
1972
1973static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1974 int window, int depth, unsigned *processed)
1975{
1976 struct thread_params *p;
1977 int i, ret, active_threads = 0;
1978
1979 init_threaded_search();
1980
1981 if (delta_search_threads <= 1) {
1982 find_deltas(list, &list_size, window, depth, processed);
1983 cleanup_threaded_search();
1984 return;
1985 }
1986 if (progress > pack_to_stdout)
1987 fprintf(stderr, "Delta compression using up to %d threads.\n",
1988 delta_search_threads);
1989 p = xcalloc(delta_search_threads, sizeof(*p));
1990
1991 /* Partition the work amongst work threads. */
1992 for (i = 0; i < delta_search_threads; i++) {
1993 unsigned sub_size = list_size / (delta_search_threads - i);
1994
1995 /* don't use too small segments or no deltas will be found */
1996 if (sub_size < 2*window && i+1 < delta_search_threads)
1997 sub_size = 0;
1998
1999 p[i].window = window;
2000 p[i].depth = depth;
2001 p[i].processed = processed;
2002 p[i].working = 1;
2003 p[i].data_ready = 0;
2004
2005 /* try to split chunks on "path" boundaries */
2006 while (sub_size && sub_size < list_size &&
2007 list[sub_size]->hash &&
2008 list[sub_size]->hash == list[sub_size-1]->hash)
2009 sub_size++;
2010
2011 p[i].list = list;
2012 p[i].list_size = sub_size;
2013 p[i].remaining = sub_size;
2014
2015 list += sub_size;
2016 list_size -= sub_size;
2017 }
2018
2019 /* Start work threads. */
2020 for (i = 0; i < delta_search_threads; i++) {
2021 if (!p[i].list_size)
2022 continue;
2023 pthread_mutex_init(&p[i].mutex, NULL);
2024 pthread_cond_init(&p[i].cond, NULL);
2025 ret = pthread_create(&p[i].thread, NULL,
2026 threaded_find_deltas, &p[i]);
2027 if (ret)
2028 die("unable to create thread: %s", strerror(ret));
2029 active_threads++;
2030 }
2031
2032 /*
2033 * Now let's wait for work completion. Each time a thread is done
2034 * with its work, we steal half of the remaining work from the
2035 * thread with the largest number of unprocessed objects and give
2036 * it to that newly idle thread. This ensure good load balancing
2037 * until the remaining object list segments are simply too short
2038 * to be worth splitting anymore.
2039 */
2040 while (active_threads) {
2041 struct thread_params *target = NULL;
2042 struct thread_params *victim = NULL;
2043 unsigned sub_size = 0;
2044
2045 progress_lock();
2046 for (;;) {
2047 for (i = 0; !target && i < delta_search_threads; i++)
2048 if (!p[i].working)
2049 target = &p[i];
2050 if (target)
2051 break;
2052 pthread_cond_wait(&progress_cond, &progress_mutex);
2053 }
2054
2055 for (i = 0; i < delta_search_threads; i++)
2056 if (p[i].remaining > 2*window &&
2057 (!victim || victim->remaining < p[i].remaining))
2058 victim = &p[i];
2059 if (victim) {
2060 sub_size = victim->remaining / 2;
2061 list = victim->list + victim->list_size - sub_size;
2062 while (sub_size && list[0]->hash &&
2063 list[0]->hash == list[-1]->hash) {
2064 list++;
2065 sub_size--;
2066 }
2067 if (!sub_size) {
2068 /*
2069 * It is possible for some "paths" to have
2070 * so many objects that no hash boundary
2071 * might be found. Let's just steal the
2072 * exact half in that case.
2073 */
2074 sub_size = victim->remaining / 2;
2075 list -= sub_size;
2076 }
2077 target->list = list;
2078 victim->list_size -= sub_size;
2079 victim->remaining -= sub_size;
2080 }
2081 target->list_size = sub_size;
2082 target->remaining = sub_size;
2083 target->working = 1;
2084 progress_unlock();
2085
2086 pthread_mutex_lock(&target->mutex);
2087 target->data_ready = 1;
2088 pthread_cond_signal(&target->cond);
2089 pthread_mutex_unlock(&target->mutex);
2090
2091 if (!sub_size) {
2092 pthread_join(target->thread, NULL);
2093 pthread_cond_destroy(&target->cond);
2094 pthread_mutex_destroy(&target->mutex);
2095 active_threads--;
2096 }
2097 }
2098 cleanup_threaded_search();
2099 free(p);
2100}
2101
2102#else
2103#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2104#endif
2105
2106static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2107{
2108 struct object_id peeled;
2109
2110 if (starts_with(path, "refs/tags/") && /* is a tag? */
2111 !peel_ref(path, peeled.hash) && /* peelable? */
2112 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2113 add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2114 return 0;
2115}
2116
2117static void prepare_pack(int window, int depth)
2118{
2119 struct object_entry **delta_list;
2120 uint32_t i, nr_deltas;
2121 unsigned n;
2122
2123 get_object_details();
2124
2125 /*
2126 * If we're locally repacking then we need to be doubly careful
2127 * from now on in order to make sure no stealth corruption gets
2128 * propagated to the new pack. Clients receiving streamed packs
2129 * should validate everything they get anyway so no need to incur
2130 * the additional cost here in that case.
2131 */
2132 if (!pack_to_stdout)
2133 do_check_packed_object_crc = 1;
2134
2135 if (!to_pack.nr_objects || !window || !depth)
2136 return;
2137
2138 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2139 nr_deltas = n = 0;
2140
2141 for (i = 0; i < to_pack.nr_objects; i++) {
2142 struct object_entry *entry = to_pack.objects + i;
2143
2144 if (entry->delta)
2145 /* This happens if we decided to reuse existing
2146 * delta from a pack. "reuse_delta &&" is implied.
2147 */
2148 continue;
2149
2150 if (entry->size < 50)
2151 continue;
2152
2153 if (entry->no_try_delta)
2154 continue;
2155
2156 if (!entry->preferred_base) {
2157 nr_deltas++;
2158 if (entry->type < 0)
2159 die("unable to get type of object %s",
2160 sha1_to_hex(entry->idx.sha1));
2161 } else {
2162 if (entry->type < 0) {
2163 /*
2164 * This object is not found, but we
2165 * don't have to include it anyway.
2166 */
2167 continue;
2168 }
2169 }
2170
2171 delta_list[n++] = entry;
2172 }
2173
2174 if (nr_deltas && n > 1) {
2175 unsigned nr_done = 0;
2176 if (progress)
2177 progress_state = start_progress(_("Compressing objects"),
2178 nr_deltas);
2179 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2180 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2181 stop_progress(&progress_state);
2182 if (nr_done != nr_deltas)
2183 die("inconsistency with delta count");
2184 }
2185 free(delta_list);
2186}
2187
2188static int git_pack_config(const char *k, const char *v, void *cb)
2189{
2190 if (!strcmp(k, "pack.window")) {
2191 window = git_config_int(k, v);
2192 return 0;
2193 }
2194 if (!strcmp(k, "pack.windowmemory")) {
2195 window_memory_limit = git_config_ulong(k, v);
2196 return 0;
2197 }
2198 if (!strcmp(k, "pack.depth")) {
2199 depth = git_config_int(k, v);
2200 return 0;
2201 }
2202 if (!strcmp(k, "pack.compression")) {
2203 int level = git_config_int(k, v);
2204 if (level == -1)
2205 level = Z_DEFAULT_COMPRESSION;
2206 else if (level < 0 || level > Z_BEST_COMPRESSION)
2207 die("bad pack compression level %d", level);
2208 pack_compression_level = level;
2209 pack_compression_seen = 1;
2210 return 0;
2211 }
2212 if (!strcmp(k, "pack.deltacachesize")) {
2213 max_delta_cache_size = git_config_int(k, v);
2214 return 0;
2215 }
2216 if (!strcmp(k, "pack.deltacachelimit")) {
2217 cache_max_small_delta_size = git_config_int(k, v);
2218 return 0;
2219 }
2220 if (!strcmp(k, "pack.writebitmaphashcache")) {
2221 if (git_config_bool(k, v))
2222 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2223 else
2224 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2225 }
2226 if (!strcmp(k, "pack.usebitmaps")) {
2227 use_bitmap_index = git_config_bool(k, v);
2228 return 0;
2229 }
2230 if (!strcmp(k, "pack.threads")) {
2231 delta_search_threads = git_config_int(k, v);
2232 if (delta_search_threads < 0)
2233 die("invalid number of threads specified (%d)",
2234 delta_search_threads);
2235#ifdef NO_PTHREADS
2236 if (delta_search_threads != 1)
2237 warning("no threads support, ignoring %s", k);
2238#endif
2239 return 0;
2240 }
2241 if (!strcmp(k, "pack.indexversion")) {
2242 pack_idx_opts.version = git_config_int(k, v);
2243 if (pack_idx_opts.version > 2)
2244 die("bad pack.indexversion=%"PRIu32,
2245 pack_idx_opts.version);
2246 return 0;
2247 }
2248 return git_default_config(k, v, cb);
2249}
2250
2251static void read_object_list_from_stdin(void)
2252{
2253 char line[40 + 1 + PATH_MAX + 2];
2254 unsigned char sha1[20];
2255
2256 for (;;) {
2257 if (!fgets(line, sizeof(line), stdin)) {
2258 if (feof(stdin))
2259 break;
2260 if (!ferror(stdin))
2261 die("fgets returned NULL, not EOF, not error!");
2262 if (errno != EINTR)
2263 die_errno("fgets");
2264 clearerr(stdin);
2265 continue;
2266 }
2267 if (line[0] == '-') {
2268 if (get_sha1_hex(line+1, sha1))
2269 die("expected edge sha1, got garbage:\n %s",
2270 line);
2271 add_preferred_base(sha1);
2272 continue;
2273 }
2274 if (get_sha1_hex(line, sha1))
2275 die("expected sha1, got garbage:\n %s", line);
2276
2277 add_preferred_base_object(line+41);
2278 add_object_entry(sha1, 0, line+41, 0);
2279 }
2280}
2281
2282#define OBJECT_ADDED (1u<<20)
2283
2284static void show_commit(struct commit *commit, void *data)
2285{
2286 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2287 commit->object.flags |= OBJECT_ADDED;
2288
2289 if (write_bitmap_index)
2290 index_commit_for_bitmap(commit);
2291}
2292
2293static void show_object(struct object *obj, const char *name, void *data)
2294{
2295 add_preferred_base_object(name);
2296 add_object_entry(obj->oid.hash, obj->type, name, 0);
2297 obj->flags |= OBJECT_ADDED;
2298}
2299
2300static void show_edge(struct commit *commit)
2301{
2302 add_preferred_base(commit->object.oid.hash);
2303}
2304
2305struct in_pack_object {
2306 off_t offset;
2307 struct object *object;
2308};
2309
2310struct in_pack {
2311 int alloc;
2312 int nr;
2313 struct in_pack_object *array;
2314};
2315
2316static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2317{
2318 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2319 in_pack->array[in_pack->nr].object = object;
2320 in_pack->nr++;
2321}
2322
2323/*
2324 * Compare the objects in the offset order, in order to emulate the
2325 * "git rev-list --objects" output that produced the pack originally.
2326 */
2327static int ofscmp(const void *a_, const void *b_)
2328{
2329 struct in_pack_object *a = (struct in_pack_object *)a_;
2330 struct in_pack_object *b = (struct in_pack_object *)b_;
2331
2332 if (a->offset < b->offset)
2333 return -1;
2334 else if (a->offset > b->offset)
2335 return 1;
2336 else
2337 return oidcmp(&a->object->oid, &b->object->oid);
2338}
2339
2340static void add_objects_in_unpacked_packs(struct rev_info *revs)
2341{
2342 struct packed_git *p;
2343 struct in_pack in_pack;
2344 uint32_t i;
2345
2346 memset(&in_pack, 0, sizeof(in_pack));
2347
2348 for (p = packed_git; p; p = p->next) {
2349 const unsigned char *sha1;
2350 struct object *o;
2351
2352 if (!p->pack_local || p->pack_keep)
2353 continue;
2354 if (open_pack_index(p))
2355 die("cannot open pack index");
2356
2357 ALLOC_GROW(in_pack.array,
2358 in_pack.nr + p->num_objects,
2359 in_pack.alloc);
2360
2361 for (i = 0; i < p->num_objects; i++) {
2362 sha1 = nth_packed_object_sha1(p, i);
2363 o = lookup_unknown_object(sha1);
2364 if (!(o->flags & OBJECT_ADDED))
2365 mark_in_pack_object(o, p, &in_pack);
2366 o->flags |= OBJECT_ADDED;
2367 }
2368 }
2369
2370 if (in_pack.nr) {
2371 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2372 ofscmp);
2373 for (i = 0; i < in_pack.nr; i++) {
2374 struct object *o = in_pack.array[i].object;
2375 add_object_entry(o->oid.hash, o->type, "", 0);
2376 }
2377 }
2378 free(in_pack.array);
2379}
2380
2381static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2382{
2383 static struct packed_git *last_found = (void *)1;
2384 struct packed_git *p;
2385
2386 p = (last_found != (void *)1) ? last_found : packed_git;
2387
2388 while (p) {
2389 if ((!p->pack_local || p->pack_keep) &&
2390 find_pack_entry_one(sha1, p)) {
2391 last_found = p;
2392 return 1;
2393 }
2394 if (p == last_found)
2395 p = packed_git;
2396 else
2397 p = p->next;
2398 if (p == last_found)
2399 p = p->next;
2400 }
2401 return 0;
2402}
2403
2404/*
2405 * Store a list of sha1s that are should not be discarded
2406 * because they are either written too recently, or are
2407 * reachable from another object that was.
2408 *
2409 * This is filled by get_object_list.
2410 */
2411static struct sha1_array recent_objects;
2412
2413static int loosened_object_can_be_discarded(const unsigned char *sha1,
2414 unsigned long mtime)
2415{
2416 if (!unpack_unreachable_expiration)
2417 return 0;
2418 if (mtime > unpack_unreachable_expiration)
2419 return 0;
2420 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2421 return 0;
2422 return 1;
2423}
2424
2425static void loosen_unused_packed_objects(struct rev_info *revs)
2426{
2427 struct packed_git *p;
2428 uint32_t i;
2429 const unsigned char *sha1;
2430
2431 for (p = packed_git; p; p = p->next) {
2432 if (!p->pack_local || p->pack_keep)
2433 continue;
2434
2435 if (open_pack_index(p))
2436 die("cannot open pack index");
2437
2438 for (i = 0; i < p->num_objects; i++) {
2439 sha1 = nth_packed_object_sha1(p, i);
2440 if (!packlist_find(&to_pack, sha1, NULL) &&
2441 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2442 !loosened_object_can_be_discarded(sha1, p->mtime))
2443 if (force_object_loose(sha1, p->mtime))
2444 die("unable to force loose object");
2445 }
2446 }
2447}
2448
2449/*
2450 * This tracks any options which a reader of the pack might
2451 * not understand, and which would therefore prevent blind reuse
2452 * of what we have on disk.
2453 */
2454static int pack_options_allow_reuse(void)
2455{
2456 return allow_ofs_delta;
2457}
2458
2459static int get_object_list_from_bitmap(struct rev_info *revs)
2460{
2461 if (prepare_bitmap_walk(revs) < 0)
2462 return -1;
2463
2464 if (pack_options_allow_reuse() &&
2465 !reuse_partial_packfile_from_bitmap(
2466 &reuse_packfile,
2467 &reuse_packfile_objects,
2468 &reuse_packfile_offset)) {
2469 assert(reuse_packfile_objects);
2470 nr_result += reuse_packfile_objects;
2471 display_progress(progress_state, nr_result);
2472 }
2473
2474 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2475 return 0;
2476}
2477
2478static void record_recent_object(struct object *obj,
2479 const char *name,
2480 void *data)
2481{
2482 sha1_array_append(&recent_objects, obj->oid.hash);
2483}
2484
2485static void record_recent_commit(struct commit *commit, void *data)
2486{
2487 sha1_array_append(&recent_objects, commit->object.oid.hash);
2488}
2489
2490static void get_object_list(int ac, const char **av)
2491{
2492 struct rev_info revs;
2493 char line[1000];
2494 int flags = 0;
2495
2496 init_revisions(&revs, NULL);
2497 save_commit_buffer = 0;
2498 setup_revisions(ac, av, &revs, NULL);
2499
2500 /* make sure shallows are read */
2501 is_repository_shallow();
2502
2503 while (fgets(line, sizeof(line), stdin) != NULL) {
2504 int len = strlen(line);
2505 if (len && line[len - 1] == '\n')
2506 line[--len] = 0;
2507 if (!len)
2508 break;
2509 if (*line == '-') {
2510 if (!strcmp(line, "--not")) {
2511 flags ^= UNINTERESTING;
2512 write_bitmap_index = 0;
2513 continue;
2514 }
2515 if (starts_with(line, "--shallow ")) {
2516 unsigned char sha1[20];
2517 if (get_sha1_hex(line + 10, sha1))
2518 die("not an SHA-1 '%s'", line + 10);
2519 register_shallow(sha1);
2520 use_bitmap_index = 0;
2521 continue;
2522 }
2523 die("not a rev '%s'", line);
2524 }
2525 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2526 die("bad revision '%s'", line);
2527 }
2528
2529 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2530 return;
2531
2532 if (prepare_revision_walk(&revs))
2533 die("revision walk setup failed");
2534 mark_edges_uninteresting(&revs, show_edge);
2535 traverse_commit_list(&revs, show_commit, show_object, NULL);
2536
2537 if (unpack_unreachable_expiration) {
2538 revs.ignore_missing_links = 1;
2539 if (add_unseen_recent_objects_to_traversal(&revs,
2540 unpack_unreachable_expiration))
2541 die("unable to add recent objects");
2542 if (prepare_revision_walk(&revs))
2543 die("revision walk setup failed");
2544 traverse_commit_list(&revs, record_recent_commit,
2545 record_recent_object, NULL);
2546 }
2547
2548 if (keep_unreachable)
2549 add_objects_in_unpacked_packs(&revs);
2550 if (unpack_unreachable)
2551 loosen_unused_packed_objects(&revs);
2552
2553 sha1_array_clear(&recent_objects);
2554}
2555
2556static int option_parse_index_version(const struct option *opt,
2557 const char *arg, int unset)
2558{
2559 char *c;
2560 const char *val = arg;
2561 pack_idx_opts.version = strtoul(val, &c, 10);
2562 if (pack_idx_opts.version > 2)
2563 die(_("unsupported index version %s"), val);
2564 if (*c == ',' && c[1])
2565 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2566 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2567 die(_("bad index version '%s'"), val);
2568 return 0;
2569}
2570
2571static int option_parse_unpack_unreachable(const struct option *opt,
2572 const char *arg, int unset)
2573{
2574 if (unset) {
2575 unpack_unreachable = 0;
2576 unpack_unreachable_expiration = 0;
2577 }
2578 else {
2579 unpack_unreachable = 1;
2580 if (arg)
2581 unpack_unreachable_expiration = approxidate(arg);
2582 }
2583 return 0;
2584}
2585
2586int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2587{
2588 int use_internal_rev_list = 0;
2589 int thin = 0;
2590 int shallow = 0;
2591 int all_progress_implied = 0;
2592 struct argv_array rp = ARGV_ARRAY_INIT;
2593 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2594 int rev_list_index = 0;
2595 struct option pack_objects_options[] = {
2596 OPT_SET_INT('q', "quiet", &progress,
2597 N_("do not show progress meter"), 0),
2598 OPT_SET_INT(0, "progress", &progress,
2599 N_("show progress meter"), 1),
2600 OPT_SET_INT(0, "all-progress", &progress,
2601 N_("show progress meter during object writing phase"), 2),
2602 OPT_BOOL(0, "all-progress-implied",
2603 &all_progress_implied,
2604 N_("similar to --all-progress when progress meter is shown")),
2605 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2606 N_("write the pack index file in the specified idx format version"),
2607 0, option_parse_index_version },
2608 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2609 N_("maximum size of each output pack file")),
2610 OPT_BOOL(0, "local", &local,
2611 N_("ignore borrowed objects from alternate object store")),
2612 OPT_BOOL(0, "incremental", &incremental,
2613 N_("ignore packed objects")),
2614 OPT_INTEGER(0, "window", &window,
2615 N_("limit pack window by objects")),
2616 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2617 N_("limit pack window by memory in addition to object limit")),
2618 OPT_INTEGER(0, "depth", &depth,
2619 N_("maximum length of delta chain allowed in the resulting pack")),
2620 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2621 N_("reuse existing deltas")),
2622 OPT_BOOL(0, "reuse-object", &reuse_object,
2623 N_("reuse existing objects")),
2624 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2625 N_("use OFS_DELTA objects")),
2626 OPT_INTEGER(0, "threads", &delta_search_threads,
2627 N_("use threads when searching for best delta matches")),
2628 OPT_BOOL(0, "non-empty", &non_empty,
2629 N_("do not create an empty pack output")),
2630 OPT_BOOL(0, "revs", &use_internal_rev_list,
2631 N_("read revision arguments from standard input")),
2632 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2633 N_("limit the objects to those that are not yet packed"),
2634 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2635 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2636 N_("include objects reachable from any reference"),
2637 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2638 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2639 N_("include objects referred by reflog entries"),
2640 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2641 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2642 N_("include objects referred to by the index"),
2643 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2644 OPT_BOOL(0, "stdout", &pack_to_stdout,
2645 N_("output pack to stdout")),
2646 OPT_BOOL(0, "include-tag", &include_tag,
2647 N_("include tag objects that refer to objects to be packed")),
2648 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2649 N_("keep unreachable objects")),
2650 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2651 N_("unpack unreachable objects newer than <time>"),
2652 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2653 OPT_BOOL(0, "thin", &thin,
2654 N_("create thin packs")),
2655 OPT_BOOL(0, "shallow", &shallow,
2656 N_("create packs suitable for shallow fetches")),
2657 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2658 N_("ignore packs that have companion .keep file")),
2659 OPT_INTEGER(0, "compression", &pack_compression_level,
2660 N_("pack compression level")),
2661 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2662 N_("do not hide commits by grafts"), 0),
2663 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2664 N_("use a bitmap index if available to speed up counting objects")),
2665 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2666 N_("write a bitmap index together with the pack index")),
2667 OPT_END(),
2668 };
2669
2670 check_replace_refs = 0;
2671
2672 reset_pack_idx_option(&pack_idx_opts);
2673 git_config(git_pack_config, NULL);
2674 if (!pack_compression_seen && core_compression_seen)
2675 pack_compression_level = core_compression_level;
2676
2677 progress = isatty(2);
2678 argc = parse_options(argc, argv, prefix, pack_objects_options,
2679 pack_usage, 0);
2680
2681 if (argc) {
2682 base_name = argv[0];
2683 argc--;
2684 }
2685 if (pack_to_stdout != !base_name || argc)
2686 usage_with_options(pack_usage, pack_objects_options);
2687
2688 argv_array_push(&rp, "pack-objects");
2689 if (thin) {
2690 use_internal_rev_list = 1;
2691 argv_array_push(&rp, shallow
2692 ? "--objects-edge-aggressive"
2693 : "--objects-edge");
2694 } else
2695 argv_array_push(&rp, "--objects");
2696
2697 if (rev_list_all) {
2698 use_internal_rev_list = 1;
2699 argv_array_push(&rp, "--all");
2700 }
2701 if (rev_list_reflog) {
2702 use_internal_rev_list = 1;
2703 argv_array_push(&rp, "--reflog");
2704 }
2705 if (rev_list_index) {
2706 use_internal_rev_list = 1;
2707 argv_array_push(&rp, "--indexed-objects");
2708 }
2709 if (rev_list_unpacked) {
2710 use_internal_rev_list = 1;
2711 argv_array_push(&rp, "--unpacked");
2712 }
2713
2714 if (!reuse_object)
2715 reuse_delta = 0;
2716 if (pack_compression_level == -1)
2717 pack_compression_level = Z_DEFAULT_COMPRESSION;
2718 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2719 die("bad pack compression level %d", pack_compression_level);
2720
2721 if (!delta_search_threads) /* --threads=0 means autodetect */
2722 delta_search_threads = online_cpus();
2723
2724#ifdef NO_PTHREADS
2725 if (delta_search_threads != 1)
2726 warning("no threads support, ignoring --threads");
2727#endif
2728 if (!pack_to_stdout && !pack_size_limit)
2729 pack_size_limit = pack_size_limit_cfg;
2730 if (pack_to_stdout && pack_size_limit)
2731 die("--max-pack-size cannot be used to build a pack for transfer.");
2732 if (pack_size_limit && pack_size_limit < 1024*1024) {
2733 warning("minimum pack size limit is 1 MiB");
2734 pack_size_limit = 1024*1024;
2735 }
2736
2737 if (!pack_to_stdout && thin)
2738 die("--thin cannot be used to build an indexable pack.");
2739
2740 if (keep_unreachable && unpack_unreachable)
2741 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2742 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2743 unpack_unreachable_expiration = 0;
2744
2745 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2746 use_bitmap_index = 0;
2747
2748 if (pack_to_stdout || !rev_list_all)
2749 write_bitmap_index = 0;
2750
2751 if (progress && all_progress_implied)
2752 progress = 2;
2753
2754 prepare_packed_git();
2755
2756 if (progress)
2757 progress_state = start_progress(_("Counting objects"), 0);
2758 if (!use_internal_rev_list)
2759 read_object_list_from_stdin();
2760 else {
2761 get_object_list(rp.argc, rp.argv);
2762 argv_array_clear(&rp);
2763 }
2764 cleanup_preferred_base();
2765 if (include_tag && nr_result)
2766 for_each_ref(add_ref_tag, NULL);
2767 stop_progress(&progress_state);
2768
2769 if (non_empty && !nr_result)
2770 return 0;
2771 if (nr_result)
2772 prepare_pack(window, depth);
2773 write_pack_file();
2774 if (progress)
2775 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2776 " reused %"PRIu32" (delta %"PRIu32")\n",
2777 written, written_delta, reused, reused_delta);
2778 return 0;
2779}