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 pack_loose_unreachable;
48static int local;
49static int incremental;
50static int ignore_packed_keep;
51static int allow_ofs_delta;
52static struct pack_idx_option pack_idx_opts;
53static const char *base_name;
54static int progress = 1;
55static int window = 10;
56static unsigned long pack_size_limit;
57static int depth = 50;
58static int delta_search_threads;
59static int pack_to_stdout;
60static int num_preferred_base;
61static struct progress *progress_state;
62static int pack_compression_level = Z_DEFAULT_COMPRESSION;
63static int pack_compression_seen;
64
65static struct packed_git *reuse_packfile;
66static uint32_t reuse_packfile_objects;
67static off_t reuse_packfile_offset;
68
69static int use_bitmap_index = 1;
70static int write_bitmap_index;
71static uint16_t write_bitmap_options;
72
73static unsigned long delta_cache_size = 0;
74static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
75static unsigned long cache_max_small_delta_size = 1000;
76
77static unsigned long window_memory_limit = 0;
78
79/*
80 * stats
81 */
82static uint32_t written, written_delta;
83static uint32_t reused, reused_delta;
84
85/*
86 * Indexed commits
87 */
88static struct commit **indexed_commits;
89static unsigned int indexed_commits_nr;
90static unsigned int indexed_commits_alloc;
91
92static void index_commit_for_bitmap(struct commit *commit)
93{
94 if (indexed_commits_nr >= indexed_commits_alloc) {
95 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
96 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
97 }
98
99 indexed_commits[indexed_commits_nr++] = commit;
100}
101
102static void *get_delta(struct object_entry *entry)
103{
104 unsigned long size, base_size, delta_size;
105 void *buf, *base_buf, *delta_buf;
106 enum object_type type;
107
108 buf = read_sha1_file(entry->idx.sha1, &type, &size);
109 if (!buf)
110 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
111 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
112 if (!base_buf)
113 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
114 delta_buf = diff_delta(base_buf, base_size,
115 buf, size, &delta_size, 0);
116 if (!delta_buf || delta_size != entry->delta_size)
117 die("delta size changed");
118 free(buf);
119 free(base_buf);
120 return delta_buf;
121}
122
123static unsigned long do_compress(void **pptr, unsigned long size)
124{
125 git_zstream stream;
126 void *in, *out;
127 unsigned long maxsize;
128
129 git_deflate_init(&stream, pack_compression_level);
130 maxsize = git_deflate_bound(&stream, size);
131
132 in = *pptr;
133 out = xmalloc(maxsize);
134 *pptr = out;
135
136 stream.next_in = in;
137 stream.avail_in = size;
138 stream.next_out = out;
139 stream.avail_out = maxsize;
140 while (git_deflate(&stream, Z_FINISH) == Z_OK)
141 ; /* nothing */
142 git_deflate_end(&stream);
143
144 free(in);
145 return stream.total_out;
146}
147
148static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
149 const unsigned char *sha1)
150{
151 git_zstream stream;
152 unsigned char ibuf[1024 * 16];
153 unsigned char obuf[1024 * 16];
154 unsigned long olen = 0;
155
156 git_deflate_init(&stream, pack_compression_level);
157
158 for (;;) {
159 ssize_t readlen;
160 int zret = Z_OK;
161 readlen = read_istream(st, ibuf, sizeof(ibuf));
162 if (readlen == -1)
163 die(_("unable to read %s"), sha1_to_hex(sha1));
164
165 stream.next_in = ibuf;
166 stream.avail_in = readlen;
167 while ((stream.avail_in || readlen == 0) &&
168 (zret == Z_OK || zret == Z_BUF_ERROR)) {
169 stream.next_out = obuf;
170 stream.avail_out = sizeof(obuf);
171 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
172 sha1write(f, obuf, stream.next_out - obuf);
173 olen += stream.next_out - obuf;
174 }
175 if (stream.avail_in)
176 die(_("deflate error (%d)"), zret);
177 if (readlen == 0) {
178 if (zret != Z_STREAM_END)
179 die(_("deflate error (%d)"), zret);
180 break;
181 }
182 }
183 git_deflate_end(&stream);
184 return olen;
185}
186
187/*
188 * we are going to reuse the existing object data as is. make
189 * sure it is not corrupt.
190 */
191static int check_pack_inflate(struct packed_git *p,
192 struct pack_window **w_curs,
193 off_t offset,
194 off_t len,
195 unsigned long expect)
196{
197 git_zstream stream;
198 unsigned char fakebuf[4096], *in;
199 int st;
200
201 memset(&stream, 0, sizeof(stream));
202 git_inflate_init(&stream);
203 do {
204 in = use_pack(p, w_curs, offset, &stream.avail_in);
205 stream.next_in = in;
206 stream.next_out = fakebuf;
207 stream.avail_out = sizeof(fakebuf);
208 st = git_inflate(&stream, Z_FINISH);
209 offset += stream.next_in - in;
210 } while (st == Z_OK || st == Z_BUF_ERROR);
211 git_inflate_end(&stream);
212 return (st == Z_STREAM_END &&
213 stream.total_out == expect &&
214 stream.total_in == len) ? 0 : -1;
215}
216
217static void copy_pack_data(struct sha1file *f,
218 struct packed_git *p,
219 struct pack_window **w_curs,
220 off_t offset,
221 off_t len)
222{
223 unsigned char *in;
224 unsigned long avail;
225
226 while (len) {
227 in = use_pack(p, w_curs, offset, &avail);
228 if (avail > len)
229 avail = (unsigned long)len;
230 sha1write(f, in, avail);
231 offset += avail;
232 len -= avail;
233 }
234}
235
236/* Return 0 if we will bust the pack-size limit */
237static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
238 unsigned long limit, int usable_delta)
239{
240 unsigned long size, datalen;
241 unsigned char header[10], dheader[10];
242 unsigned hdrlen;
243 enum object_type type;
244 void *buf;
245 struct git_istream *st = NULL;
246
247 if (!usable_delta) {
248 if (entry->type == OBJ_BLOB &&
249 entry->size > big_file_threshold &&
250 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
251 buf = NULL;
252 else {
253 buf = read_sha1_file(entry->idx.sha1, &type, &size);
254 if (!buf)
255 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
256 }
257 /*
258 * make sure no cached delta data remains from a
259 * previous attempt before a pack split occurred.
260 */
261 free(entry->delta_data);
262 entry->delta_data = NULL;
263 entry->z_delta_size = 0;
264 } else if (entry->delta_data) {
265 size = entry->delta_size;
266 buf = entry->delta_data;
267 entry->delta_data = NULL;
268 type = (allow_ofs_delta && entry->delta->idx.offset) ?
269 OBJ_OFS_DELTA : OBJ_REF_DELTA;
270 } else {
271 buf = get_delta(entry);
272 size = entry->delta_size;
273 type = (allow_ofs_delta && entry->delta->idx.offset) ?
274 OBJ_OFS_DELTA : OBJ_REF_DELTA;
275 }
276
277 if (st) /* large blob case, just assume we don't compress well */
278 datalen = size;
279 else if (entry->z_delta_size)
280 datalen = entry->z_delta_size;
281 else
282 datalen = do_compress(&buf, size);
283
284 /*
285 * The object header is a byte of 'type' followed by zero or
286 * more bytes of length.
287 */
288 hdrlen = encode_in_pack_object_header(type, size, header);
289
290 if (type == OBJ_OFS_DELTA) {
291 /*
292 * Deltas with relative base contain an additional
293 * encoding of the relative offset for the delta
294 * base from this object's position in the pack.
295 */
296 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
297 unsigned pos = sizeof(dheader) - 1;
298 dheader[pos] = ofs & 127;
299 while (ofs >>= 7)
300 dheader[--pos] = 128 | (--ofs & 127);
301 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
302 if (st)
303 close_istream(st);
304 free(buf);
305 return 0;
306 }
307 sha1write(f, header, hdrlen);
308 sha1write(f, dheader + pos, sizeof(dheader) - pos);
309 hdrlen += sizeof(dheader) - pos;
310 } else if (type == OBJ_REF_DELTA) {
311 /*
312 * Deltas with a base reference contain
313 * an additional 20 bytes for the base sha1.
314 */
315 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
316 if (st)
317 close_istream(st);
318 free(buf);
319 return 0;
320 }
321 sha1write(f, header, hdrlen);
322 sha1write(f, entry->delta->idx.sha1, 20);
323 hdrlen += 20;
324 } else {
325 if (limit && hdrlen + datalen + 20 >= limit) {
326 if (st)
327 close_istream(st);
328 free(buf);
329 return 0;
330 }
331 sha1write(f, header, hdrlen);
332 }
333 if (st) {
334 datalen = write_large_blob_data(st, f, entry->idx.sha1);
335 close_istream(st);
336 } else {
337 sha1write(f, buf, datalen);
338 free(buf);
339 }
340
341 return hdrlen + datalen;
342}
343
344/* Return 0 if we will bust the pack-size limit */
345static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
346 unsigned long limit, int usable_delta)
347{
348 struct packed_git *p = entry->in_pack;
349 struct pack_window *w_curs = NULL;
350 struct revindex_entry *revidx;
351 off_t offset;
352 enum object_type type = entry->type;
353 unsigned long datalen;
354 unsigned char header[10], dheader[10];
355 unsigned hdrlen;
356
357 if (entry->delta)
358 type = (allow_ofs_delta && entry->delta->idx.offset) ?
359 OBJ_OFS_DELTA : OBJ_REF_DELTA;
360 hdrlen = encode_in_pack_object_header(type, entry->size, header);
361
362 offset = entry->in_pack_offset;
363 revidx = find_pack_revindex(p, offset);
364 datalen = revidx[1].offset - offset;
365 if (!pack_to_stdout && p->index_version > 1 &&
366 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
367 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
368 unuse_pack(&w_curs);
369 return write_no_reuse_object(f, entry, limit, usable_delta);
370 }
371
372 offset += entry->in_pack_header_size;
373 datalen -= entry->in_pack_header_size;
374
375 if (!pack_to_stdout && p->index_version == 1 &&
376 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
377 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
378 unuse_pack(&w_curs);
379 return write_no_reuse_object(f, entry, limit, usable_delta);
380 }
381
382 if (type == OBJ_OFS_DELTA) {
383 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
384 unsigned pos = sizeof(dheader) - 1;
385 dheader[pos] = ofs & 127;
386 while (ofs >>= 7)
387 dheader[--pos] = 128 | (--ofs & 127);
388 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
389 unuse_pack(&w_curs);
390 return 0;
391 }
392 sha1write(f, header, hdrlen);
393 sha1write(f, dheader + pos, sizeof(dheader) - pos);
394 hdrlen += sizeof(dheader) - pos;
395 reused_delta++;
396 } else if (type == OBJ_REF_DELTA) {
397 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
398 unuse_pack(&w_curs);
399 return 0;
400 }
401 sha1write(f, header, hdrlen);
402 sha1write(f, entry->delta->idx.sha1, 20);
403 hdrlen += 20;
404 reused_delta++;
405 } else {
406 if (limit && hdrlen + datalen + 20 >= limit) {
407 unuse_pack(&w_curs);
408 return 0;
409 }
410 sha1write(f, header, hdrlen);
411 }
412 copy_pack_data(f, p, &w_curs, offset, datalen);
413 unuse_pack(&w_curs);
414 reused++;
415 return hdrlen + datalen;
416}
417
418/* Return 0 if we will bust the pack-size limit */
419static unsigned long write_object(struct sha1file *f,
420 struct object_entry *entry,
421 off_t write_offset)
422{
423 unsigned long limit, len;
424 int usable_delta, to_reuse;
425
426 if (!pack_to_stdout)
427 crc32_begin(f);
428
429 /* apply size limit if limited packsize and not first object */
430 if (!pack_size_limit || !nr_written)
431 limit = 0;
432 else if (pack_size_limit <= write_offset)
433 /*
434 * the earlier object did not fit the limit; avoid
435 * mistaking this with unlimited (i.e. limit = 0).
436 */
437 limit = 1;
438 else
439 limit = pack_size_limit - write_offset;
440
441 if (!entry->delta)
442 usable_delta = 0; /* no delta */
443 else if (!pack_size_limit)
444 usable_delta = 1; /* unlimited packfile */
445 else if (entry->delta->idx.offset == (off_t)-1)
446 usable_delta = 0; /* base was written to another pack */
447 else if (entry->delta->idx.offset)
448 usable_delta = 1; /* base already exists in this pack */
449 else
450 usable_delta = 0; /* base could end up in another pack */
451
452 if (!reuse_object)
453 to_reuse = 0; /* explicit */
454 else if (!entry->in_pack)
455 to_reuse = 0; /* can't reuse what we don't have */
456 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
457 /* check_object() decided it for us ... */
458 to_reuse = usable_delta;
459 /* ... but pack split may override that */
460 else if (entry->type != entry->in_pack_type)
461 to_reuse = 0; /* pack has delta which is unusable */
462 else if (entry->delta)
463 to_reuse = 0; /* we want to pack afresh */
464 else
465 to_reuse = 1; /* we have it in-pack undeltified,
466 * and we do not need to deltify it.
467 */
468
469 if (!to_reuse)
470 len = write_no_reuse_object(f, entry, limit, usable_delta);
471 else
472 len = write_reuse_object(f, entry, limit, usable_delta);
473 if (!len)
474 return 0;
475
476 if (usable_delta)
477 written_delta++;
478 written++;
479 if (!pack_to_stdout)
480 entry->idx.crc32 = crc32_end(f);
481 return len;
482}
483
484enum write_one_status {
485 WRITE_ONE_SKIP = -1, /* already written */
486 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
487 WRITE_ONE_WRITTEN = 1, /* normal */
488 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
489};
490
491static enum write_one_status write_one(struct sha1file *f,
492 struct object_entry *e,
493 off_t *offset)
494{
495 unsigned long size;
496 int recursing;
497
498 /*
499 * we set offset to 1 (which is an impossible value) to mark
500 * the fact that this object is involved in "write its base
501 * first before writing a deltified object" recursion.
502 */
503 recursing = (e->idx.offset == 1);
504 if (recursing) {
505 warning("recursive delta detected for object %s",
506 sha1_to_hex(e->idx.sha1));
507 return WRITE_ONE_RECURSIVE;
508 } else if (e->idx.offset || e->preferred_base) {
509 /* offset is non zero if object is written already. */
510 return WRITE_ONE_SKIP;
511 }
512
513 /* if we are deltified, write out base object first. */
514 if (e->delta) {
515 e->idx.offset = 1; /* now recurse */
516 switch (write_one(f, e->delta, offset)) {
517 case WRITE_ONE_RECURSIVE:
518 /* we cannot depend on this one */
519 e->delta = NULL;
520 break;
521 default:
522 break;
523 case WRITE_ONE_BREAK:
524 e->idx.offset = recursing;
525 return WRITE_ONE_BREAK;
526 }
527 }
528
529 e->idx.offset = *offset;
530 size = write_object(f, e, *offset);
531 if (!size) {
532 e->idx.offset = recursing;
533 return WRITE_ONE_BREAK;
534 }
535 written_list[nr_written++] = &e->idx;
536
537 /* make sure off_t is sufficiently large not to wrap */
538 if (signed_add_overflows(*offset, size))
539 die("pack too large for current definition of off_t");
540 *offset += size;
541 return WRITE_ONE_WRITTEN;
542}
543
544static int mark_tagged(const char *path, const struct object_id *oid, int flag,
545 void *cb_data)
546{
547 unsigned char peeled[20];
548 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
549
550 if (entry)
551 entry->tagged = 1;
552 if (!peel_ref(path, peeled)) {
553 entry = packlist_find(&to_pack, peeled, NULL);
554 if (entry)
555 entry->tagged = 1;
556 }
557 return 0;
558}
559
560static inline void add_to_write_order(struct object_entry **wo,
561 unsigned int *endp,
562 struct object_entry *e)
563{
564 if (e->filled)
565 return;
566 wo[(*endp)++] = e;
567 e->filled = 1;
568}
569
570static void add_descendants_to_write_order(struct object_entry **wo,
571 unsigned int *endp,
572 struct object_entry *e)
573{
574 int add_to_order = 1;
575 while (e) {
576 if (add_to_order) {
577 struct object_entry *s;
578 /* add this node... */
579 add_to_write_order(wo, endp, e);
580 /* all its siblings... */
581 for (s = e->delta_sibling; s; s = s->delta_sibling) {
582 add_to_write_order(wo, endp, s);
583 }
584 }
585 /* drop down a level to add left subtree nodes if possible */
586 if (e->delta_child) {
587 add_to_order = 1;
588 e = e->delta_child;
589 } else {
590 add_to_order = 0;
591 /* our sibling might have some children, it is next */
592 if (e->delta_sibling) {
593 e = e->delta_sibling;
594 continue;
595 }
596 /* go back to our parent node */
597 e = e->delta;
598 while (e && !e->delta_sibling) {
599 /* we're on the right side of a subtree, keep
600 * going up until we can go right again */
601 e = e->delta;
602 }
603 if (!e) {
604 /* done- we hit our original root node */
605 return;
606 }
607 /* pass it off to sibling at this level */
608 e = e->delta_sibling;
609 }
610 };
611}
612
613static void add_family_to_write_order(struct object_entry **wo,
614 unsigned int *endp,
615 struct object_entry *e)
616{
617 struct object_entry *root;
618
619 for (root = e; root->delta; root = root->delta)
620 ; /* nothing */
621 add_descendants_to_write_order(wo, endp, root);
622}
623
624static struct object_entry **compute_write_order(void)
625{
626 unsigned int i, wo_end, last_untagged;
627
628 struct object_entry **wo;
629 struct object_entry *objects = to_pack.objects;
630
631 for (i = 0; i < to_pack.nr_objects; i++) {
632 objects[i].tagged = 0;
633 objects[i].filled = 0;
634 objects[i].delta_child = NULL;
635 objects[i].delta_sibling = NULL;
636 }
637
638 /*
639 * Fully connect delta_child/delta_sibling network.
640 * Make sure delta_sibling is sorted in the original
641 * recency order.
642 */
643 for (i = to_pack.nr_objects; i > 0;) {
644 struct object_entry *e = &objects[--i];
645 if (!e->delta)
646 continue;
647 /* Mark me as the first child */
648 e->delta_sibling = e->delta->delta_child;
649 e->delta->delta_child = e;
650 }
651
652 /*
653 * Mark objects that are at the tip of tags.
654 */
655 for_each_tag_ref(mark_tagged, NULL);
656
657 /*
658 * Give the objects in the original recency order until
659 * we see a tagged tip.
660 */
661 ALLOC_ARRAY(wo, to_pack.nr_objects);
662 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
663 if (objects[i].tagged)
664 break;
665 add_to_write_order(wo, &wo_end, &objects[i]);
666 }
667 last_untagged = i;
668
669 /*
670 * Then fill all the tagged tips.
671 */
672 for (; i < to_pack.nr_objects; i++) {
673 if (objects[i].tagged)
674 add_to_write_order(wo, &wo_end, &objects[i]);
675 }
676
677 /*
678 * And then all remaining commits and tags.
679 */
680 for (i = last_untagged; i < to_pack.nr_objects; i++) {
681 if (objects[i].type != OBJ_COMMIT &&
682 objects[i].type != OBJ_TAG)
683 continue;
684 add_to_write_order(wo, &wo_end, &objects[i]);
685 }
686
687 /*
688 * And then all the trees.
689 */
690 for (i = last_untagged; i < to_pack.nr_objects; i++) {
691 if (objects[i].type != OBJ_TREE)
692 continue;
693 add_to_write_order(wo, &wo_end, &objects[i]);
694 }
695
696 /*
697 * Finally all the rest in really tight order
698 */
699 for (i = last_untagged; i < to_pack.nr_objects; i++) {
700 if (!objects[i].filled)
701 add_family_to_write_order(wo, &wo_end, &objects[i]);
702 }
703
704 if (wo_end != to_pack.nr_objects)
705 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
706
707 return wo;
708}
709
710static off_t write_reused_pack(struct sha1file *f)
711{
712 unsigned char buffer[8192];
713 off_t to_write, total;
714 int fd;
715
716 if (!is_pack_valid(reuse_packfile))
717 die("packfile is invalid: %s", reuse_packfile->pack_name);
718
719 fd = git_open_noatime(reuse_packfile->pack_name);
720 if (fd < 0)
721 die_errno("unable to open packfile for reuse: %s",
722 reuse_packfile->pack_name);
723
724 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
725 die_errno("unable to seek in reused packfile");
726
727 if (reuse_packfile_offset < 0)
728 reuse_packfile_offset = reuse_packfile->pack_size - 20;
729
730 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
731
732 while (to_write) {
733 int read_pack = xread(fd, buffer, sizeof(buffer));
734
735 if (read_pack <= 0)
736 die_errno("unable to read from reused packfile");
737
738 if (read_pack > to_write)
739 read_pack = to_write;
740
741 sha1write(f, buffer, read_pack);
742 to_write -= read_pack;
743
744 /*
745 * We don't know the actual number of objects written,
746 * only how many bytes written, how many bytes total, and
747 * how many objects total. So we can fake it by pretending all
748 * objects we are writing are the same size. This gives us a
749 * smooth progress meter, and at the end it matches the true
750 * answer.
751 */
752 written = reuse_packfile_objects *
753 (((double)(total - to_write)) / total);
754 display_progress(progress_state, written);
755 }
756
757 close(fd);
758 written = reuse_packfile_objects;
759 display_progress(progress_state, written);
760 return reuse_packfile_offset - sizeof(struct pack_header);
761}
762
763static const char no_split_warning[] = N_(
764"disabling bitmap writing, packs are split due to pack.packSizeLimit"
765);
766
767static void write_pack_file(void)
768{
769 uint32_t i = 0, j;
770 struct sha1file *f;
771 off_t offset;
772 uint32_t nr_remaining = nr_result;
773 time_t last_mtime = 0;
774 struct object_entry **write_order;
775
776 if (progress > pack_to_stdout)
777 progress_state = start_progress(_("Writing objects"), nr_result);
778 ALLOC_ARRAY(written_list, to_pack.nr_objects);
779 write_order = compute_write_order();
780
781 do {
782 unsigned char sha1[20];
783 char *pack_tmp_name = NULL;
784
785 if (pack_to_stdout)
786 f = sha1fd_throughput(1, "<stdout>", progress_state);
787 else
788 f = create_tmp_packfile(&pack_tmp_name);
789
790 offset = write_pack_header(f, nr_remaining);
791
792 if (reuse_packfile) {
793 off_t packfile_size;
794 assert(pack_to_stdout);
795
796 packfile_size = write_reused_pack(f);
797 offset += packfile_size;
798 }
799
800 nr_written = 0;
801 for (; i < to_pack.nr_objects; i++) {
802 struct object_entry *e = write_order[i];
803 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
804 break;
805 display_progress(progress_state, written);
806 }
807
808 /*
809 * Did we write the wrong # entries in the header?
810 * If so, rewrite it like in fast-import
811 */
812 if (pack_to_stdout) {
813 sha1close(f, sha1, CSUM_CLOSE);
814 } else if (nr_written == nr_remaining) {
815 sha1close(f, sha1, CSUM_FSYNC);
816 } else {
817 int fd = sha1close(f, sha1, 0);
818 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
819 nr_written, sha1, offset);
820 close(fd);
821 if (write_bitmap_index) {
822 warning(_(no_split_warning));
823 write_bitmap_index = 0;
824 }
825 }
826
827 if (!pack_to_stdout) {
828 struct stat st;
829 struct strbuf tmpname = STRBUF_INIT;
830
831 /*
832 * Packs are runtime accessed in their mtime
833 * order since newer packs are more likely to contain
834 * younger objects. So if we are creating multiple
835 * packs then we should modify the mtime of later ones
836 * to preserve this property.
837 */
838 if (stat(pack_tmp_name, &st) < 0) {
839 warning_errno("failed to stat %s", pack_tmp_name);
840 } else if (!last_mtime) {
841 last_mtime = st.st_mtime;
842 } else {
843 struct utimbuf utb;
844 utb.actime = st.st_atime;
845 utb.modtime = --last_mtime;
846 if (utime(pack_tmp_name, &utb) < 0)
847 warning_errno("failed utime() on %s", pack_tmp_name);
848 }
849
850 strbuf_addf(&tmpname, "%s-", base_name);
851
852 if (write_bitmap_index) {
853 bitmap_writer_set_checksum(sha1);
854 bitmap_writer_build_type_index(written_list, nr_written);
855 }
856
857 finish_tmp_packfile(&tmpname, pack_tmp_name,
858 written_list, nr_written,
859 &pack_idx_opts, sha1);
860
861 if (write_bitmap_index) {
862 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
863
864 stop_progress(&progress_state);
865
866 bitmap_writer_show_progress(progress);
867 bitmap_writer_reuse_bitmaps(&to_pack);
868 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
869 bitmap_writer_build(&to_pack);
870 bitmap_writer_finish(written_list, nr_written,
871 tmpname.buf, write_bitmap_options);
872 write_bitmap_index = 0;
873 }
874
875 strbuf_release(&tmpname);
876 free(pack_tmp_name);
877 puts(sha1_to_hex(sha1));
878 }
879
880 /* mark written objects as written to previous pack */
881 for (j = 0; j < nr_written; j++) {
882 written_list[j]->offset = (off_t)-1;
883 }
884 nr_remaining -= nr_written;
885 } while (nr_remaining && i < to_pack.nr_objects);
886
887 free(written_list);
888 free(write_order);
889 stop_progress(&progress_state);
890 if (written != nr_result)
891 die("wrote %"PRIu32" objects while expecting %"PRIu32,
892 written, nr_result);
893}
894
895static void setup_delta_attr_check(struct git_attr_check *check)
896{
897 static struct git_attr *attr_delta;
898
899 if (!attr_delta)
900 attr_delta = git_attr("delta");
901
902 check[0].attr = attr_delta;
903}
904
905static int no_try_delta(const char *path)
906{
907 struct git_attr_check check[1];
908
909 setup_delta_attr_check(check);
910 if (git_check_attr(path, ARRAY_SIZE(check), check))
911 return 0;
912 if (ATTR_FALSE(check->value))
913 return 1;
914 return 0;
915}
916
917/*
918 * When adding an object, check whether we have already added it
919 * to our packing list. If so, we can skip. However, if we are
920 * being asked to excludei t, but the previous mention was to include
921 * it, make sure to adjust its flags and tweak our numbers accordingly.
922 *
923 * As an optimization, we pass out the index position where we would have
924 * found the item, since that saves us from having to look it up again a
925 * few lines later when we want to add the new entry.
926 */
927static int have_duplicate_entry(const unsigned char *sha1,
928 int exclude,
929 uint32_t *index_pos)
930{
931 struct object_entry *entry;
932
933 entry = packlist_find(&to_pack, sha1, index_pos);
934 if (!entry)
935 return 0;
936
937 if (exclude) {
938 if (!entry->preferred_base)
939 nr_result--;
940 entry->preferred_base = 1;
941 }
942
943 return 1;
944}
945
946/*
947 * Check whether we want the object in the pack (e.g., we do not want
948 * objects found in non-local stores if the "--local" option was used).
949 *
950 * As a side effect of this check, we will find the packed version of this
951 * object, if any. We therefore pass out the pack information to avoid having
952 * to look it up again later.
953 */
954static int want_object_in_pack(const unsigned char *sha1,
955 int exclude,
956 struct packed_git **found_pack,
957 off_t *found_offset)
958{
959 struct packed_git *p;
960
961 if (!exclude && local && has_loose_object_nonlocal(sha1))
962 return 0;
963
964 *found_pack = NULL;
965 *found_offset = 0;
966
967 for (p = packed_git; p; p = p->next) {
968 off_t offset = find_pack_entry_one(sha1, p);
969 if (offset) {
970 if (!*found_pack) {
971 if (!is_pack_valid(p))
972 continue;
973 *found_offset = offset;
974 *found_pack = p;
975 }
976 if (exclude)
977 return 1;
978 if (incremental)
979 return 0;
980 if (local && !p->pack_local)
981 return 0;
982 if (ignore_packed_keep && p->pack_local && p->pack_keep)
983 return 0;
984 }
985 }
986
987 return 1;
988}
989
990static void create_object_entry(const unsigned char *sha1,
991 enum object_type type,
992 uint32_t hash,
993 int exclude,
994 int no_try_delta,
995 uint32_t index_pos,
996 struct packed_git *found_pack,
997 off_t found_offset)
998{
999 struct object_entry *entry;
1000
1001 entry = packlist_alloc(&to_pack, sha1, index_pos);
1002 entry->hash = hash;
1003 if (type)
1004 entry->type = type;
1005 if (exclude)
1006 entry->preferred_base = 1;
1007 else
1008 nr_result++;
1009 if (found_pack) {
1010 entry->in_pack = found_pack;
1011 entry->in_pack_offset = found_offset;
1012 }
1013
1014 entry->no_try_delta = no_try_delta;
1015}
1016
1017static const char no_closure_warning[] = N_(
1018"disabling bitmap writing, as some objects are not being packed"
1019);
1020
1021static int add_object_entry(const unsigned char *sha1, enum object_type type,
1022 const char *name, int exclude)
1023{
1024 struct packed_git *found_pack;
1025 off_t found_offset;
1026 uint32_t index_pos;
1027
1028 if (have_duplicate_entry(sha1, exclude, &index_pos))
1029 return 0;
1030
1031 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1032 /* The pack is missing an object, so it will not have closure */
1033 if (write_bitmap_index) {
1034 warning(_(no_closure_warning));
1035 write_bitmap_index = 0;
1036 }
1037 return 0;
1038 }
1039
1040 create_object_entry(sha1, type, pack_name_hash(name),
1041 exclude, name && no_try_delta(name),
1042 index_pos, found_pack, found_offset);
1043
1044 display_progress(progress_state, nr_result);
1045 return 1;
1046}
1047
1048static int add_object_entry_from_bitmap(const unsigned char *sha1,
1049 enum object_type type,
1050 int flags, uint32_t name_hash,
1051 struct packed_git *pack, off_t offset)
1052{
1053 uint32_t index_pos;
1054
1055 if (have_duplicate_entry(sha1, 0, &index_pos))
1056 return 0;
1057
1058 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1059
1060 display_progress(progress_state, nr_result);
1061 return 1;
1062}
1063
1064struct pbase_tree_cache {
1065 unsigned char sha1[20];
1066 int ref;
1067 int temporary;
1068 void *tree_data;
1069 unsigned long tree_size;
1070};
1071
1072static struct pbase_tree_cache *(pbase_tree_cache[256]);
1073static int pbase_tree_cache_ix(const unsigned char *sha1)
1074{
1075 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1076}
1077static int pbase_tree_cache_ix_incr(int ix)
1078{
1079 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1080}
1081
1082static struct pbase_tree {
1083 struct pbase_tree *next;
1084 /* This is a phony "cache" entry; we are not
1085 * going to evict it or find it through _get()
1086 * mechanism -- this is for the toplevel node that
1087 * would almost always change with any commit.
1088 */
1089 struct pbase_tree_cache pcache;
1090} *pbase_tree;
1091
1092static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1093{
1094 struct pbase_tree_cache *ent, *nent;
1095 void *data;
1096 unsigned long size;
1097 enum object_type type;
1098 int neigh;
1099 int my_ix = pbase_tree_cache_ix(sha1);
1100 int available_ix = -1;
1101
1102 /* pbase-tree-cache acts as a limited hashtable.
1103 * your object will be found at your index or within a few
1104 * slots after that slot if it is cached.
1105 */
1106 for (neigh = 0; neigh < 8; neigh++) {
1107 ent = pbase_tree_cache[my_ix];
1108 if (ent && !hashcmp(ent->sha1, sha1)) {
1109 ent->ref++;
1110 return ent;
1111 }
1112 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1113 ((0 <= available_ix) &&
1114 (!ent && pbase_tree_cache[available_ix])))
1115 available_ix = my_ix;
1116 if (!ent)
1117 break;
1118 my_ix = pbase_tree_cache_ix_incr(my_ix);
1119 }
1120
1121 /* Did not find one. Either we got a bogus request or
1122 * we need to read and perhaps cache.
1123 */
1124 data = read_sha1_file(sha1, &type, &size);
1125 if (!data)
1126 return NULL;
1127 if (type != OBJ_TREE) {
1128 free(data);
1129 return NULL;
1130 }
1131
1132 /* We need to either cache or return a throwaway copy */
1133
1134 if (available_ix < 0)
1135 ent = NULL;
1136 else {
1137 ent = pbase_tree_cache[available_ix];
1138 my_ix = available_ix;
1139 }
1140
1141 if (!ent) {
1142 nent = xmalloc(sizeof(*nent));
1143 nent->temporary = (available_ix < 0);
1144 }
1145 else {
1146 /* evict and reuse */
1147 free(ent->tree_data);
1148 nent = ent;
1149 }
1150 hashcpy(nent->sha1, sha1);
1151 nent->tree_data = data;
1152 nent->tree_size = size;
1153 nent->ref = 1;
1154 if (!nent->temporary)
1155 pbase_tree_cache[my_ix] = nent;
1156 return nent;
1157}
1158
1159static void pbase_tree_put(struct pbase_tree_cache *cache)
1160{
1161 if (!cache->temporary) {
1162 cache->ref--;
1163 return;
1164 }
1165 free(cache->tree_data);
1166 free(cache);
1167}
1168
1169static int name_cmp_len(const char *name)
1170{
1171 int i;
1172 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1173 ;
1174 return i;
1175}
1176
1177static void add_pbase_object(struct tree_desc *tree,
1178 const char *name,
1179 int cmplen,
1180 const char *fullname)
1181{
1182 struct name_entry entry;
1183 int cmp;
1184
1185 while (tree_entry(tree,&entry)) {
1186 if (S_ISGITLINK(entry.mode))
1187 continue;
1188 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1189 memcmp(name, entry.path, cmplen);
1190 if (cmp > 0)
1191 continue;
1192 if (cmp < 0)
1193 return;
1194 if (name[cmplen] != '/') {
1195 add_object_entry(entry.oid->hash,
1196 object_type(entry.mode),
1197 fullname, 1);
1198 return;
1199 }
1200 if (S_ISDIR(entry.mode)) {
1201 struct tree_desc sub;
1202 struct pbase_tree_cache *tree;
1203 const char *down = name+cmplen+1;
1204 int downlen = name_cmp_len(down);
1205
1206 tree = pbase_tree_get(entry.oid->hash);
1207 if (!tree)
1208 return;
1209 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1210
1211 add_pbase_object(&sub, down, downlen, fullname);
1212 pbase_tree_put(tree);
1213 }
1214 }
1215}
1216
1217static unsigned *done_pbase_paths;
1218static int done_pbase_paths_num;
1219static int done_pbase_paths_alloc;
1220static int done_pbase_path_pos(unsigned hash)
1221{
1222 int lo = 0;
1223 int hi = done_pbase_paths_num;
1224 while (lo < hi) {
1225 int mi = (hi + lo) / 2;
1226 if (done_pbase_paths[mi] == hash)
1227 return mi;
1228 if (done_pbase_paths[mi] < hash)
1229 hi = mi;
1230 else
1231 lo = mi + 1;
1232 }
1233 return -lo-1;
1234}
1235
1236static int check_pbase_path(unsigned hash)
1237{
1238 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1239 if (0 <= pos)
1240 return 1;
1241 pos = -pos - 1;
1242 ALLOC_GROW(done_pbase_paths,
1243 done_pbase_paths_num + 1,
1244 done_pbase_paths_alloc);
1245 done_pbase_paths_num++;
1246 if (pos < done_pbase_paths_num)
1247 memmove(done_pbase_paths + pos + 1,
1248 done_pbase_paths + pos,
1249 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1250 done_pbase_paths[pos] = hash;
1251 return 0;
1252}
1253
1254static void add_preferred_base_object(const char *name)
1255{
1256 struct pbase_tree *it;
1257 int cmplen;
1258 unsigned hash = pack_name_hash(name);
1259
1260 if (!num_preferred_base || check_pbase_path(hash))
1261 return;
1262
1263 cmplen = name_cmp_len(name);
1264 for (it = pbase_tree; it; it = it->next) {
1265 if (cmplen == 0) {
1266 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1267 }
1268 else {
1269 struct tree_desc tree;
1270 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1271 add_pbase_object(&tree, name, cmplen, name);
1272 }
1273 }
1274}
1275
1276static void add_preferred_base(unsigned char *sha1)
1277{
1278 struct pbase_tree *it;
1279 void *data;
1280 unsigned long size;
1281 unsigned char tree_sha1[20];
1282
1283 if (window <= num_preferred_base++)
1284 return;
1285
1286 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1287 if (!data)
1288 return;
1289
1290 for (it = pbase_tree; it; it = it->next) {
1291 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1292 free(data);
1293 return;
1294 }
1295 }
1296
1297 it = xcalloc(1, sizeof(*it));
1298 it->next = pbase_tree;
1299 pbase_tree = it;
1300
1301 hashcpy(it->pcache.sha1, tree_sha1);
1302 it->pcache.tree_data = data;
1303 it->pcache.tree_size = size;
1304}
1305
1306static void cleanup_preferred_base(void)
1307{
1308 struct pbase_tree *it;
1309 unsigned i;
1310
1311 it = pbase_tree;
1312 pbase_tree = NULL;
1313 while (it) {
1314 struct pbase_tree *this = it;
1315 it = this->next;
1316 free(this->pcache.tree_data);
1317 free(this);
1318 }
1319
1320 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1321 if (!pbase_tree_cache[i])
1322 continue;
1323 free(pbase_tree_cache[i]->tree_data);
1324 free(pbase_tree_cache[i]);
1325 pbase_tree_cache[i] = NULL;
1326 }
1327
1328 free(done_pbase_paths);
1329 done_pbase_paths = NULL;
1330 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1331}
1332
1333static void check_object(struct object_entry *entry)
1334{
1335 if (entry->in_pack) {
1336 struct packed_git *p = entry->in_pack;
1337 struct pack_window *w_curs = NULL;
1338 const unsigned char *base_ref = NULL;
1339 struct object_entry *base_entry;
1340 unsigned long used, used_0;
1341 unsigned long avail;
1342 off_t ofs;
1343 unsigned char *buf, c;
1344
1345 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1346
1347 /*
1348 * We want in_pack_type even if we do not reuse delta
1349 * since non-delta representations could still be reused.
1350 */
1351 used = unpack_object_header_buffer(buf, avail,
1352 &entry->in_pack_type,
1353 &entry->size);
1354 if (used == 0)
1355 goto give_up;
1356
1357 /*
1358 * Determine if this is a delta and if so whether we can
1359 * reuse it or not. Otherwise let's find out as cheaply as
1360 * possible what the actual type and size for this object is.
1361 */
1362 switch (entry->in_pack_type) {
1363 default:
1364 /* Not a delta hence we've already got all we need. */
1365 entry->type = entry->in_pack_type;
1366 entry->in_pack_header_size = used;
1367 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1368 goto give_up;
1369 unuse_pack(&w_curs);
1370 return;
1371 case OBJ_REF_DELTA:
1372 if (reuse_delta && !entry->preferred_base)
1373 base_ref = use_pack(p, &w_curs,
1374 entry->in_pack_offset + used, NULL);
1375 entry->in_pack_header_size = used + 20;
1376 break;
1377 case OBJ_OFS_DELTA:
1378 buf = use_pack(p, &w_curs,
1379 entry->in_pack_offset + used, NULL);
1380 used_0 = 0;
1381 c = buf[used_0++];
1382 ofs = c & 127;
1383 while (c & 128) {
1384 ofs += 1;
1385 if (!ofs || MSB(ofs, 7)) {
1386 error("delta base offset overflow in pack for %s",
1387 sha1_to_hex(entry->idx.sha1));
1388 goto give_up;
1389 }
1390 c = buf[used_0++];
1391 ofs = (ofs << 7) + (c & 127);
1392 }
1393 ofs = entry->in_pack_offset - ofs;
1394 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1395 error("delta base offset out of bound for %s",
1396 sha1_to_hex(entry->idx.sha1));
1397 goto give_up;
1398 }
1399 if (reuse_delta && !entry->preferred_base) {
1400 struct revindex_entry *revidx;
1401 revidx = find_pack_revindex(p, ofs);
1402 if (!revidx)
1403 goto give_up;
1404 base_ref = nth_packed_object_sha1(p, revidx->nr);
1405 }
1406 entry->in_pack_header_size = used + used_0;
1407 break;
1408 }
1409
1410 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1411 /*
1412 * If base_ref was set above that means we wish to
1413 * reuse delta data, and we even found that base
1414 * in the list of objects we want to pack. Goodie!
1415 *
1416 * Depth value does not matter - find_deltas() will
1417 * never consider reused delta as the base object to
1418 * deltify other objects against, in order to avoid
1419 * circular deltas.
1420 */
1421 entry->type = entry->in_pack_type;
1422 entry->delta = base_entry;
1423 entry->delta_size = entry->size;
1424 entry->delta_sibling = base_entry->delta_child;
1425 base_entry->delta_child = entry;
1426 unuse_pack(&w_curs);
1427 return;
1428 }
1429
1430 if (entry->type) {
1431 /*
1432 * This must be a delta and we already know what the
1433 * final object type is. Let's extract the actual
1434 * object size from the delta header.
1435 */
1436 entry->size = get_size_from_delta(p, &w_curs,
1437 entry->in_pack_offset + entry->in_pack_header_size);
1438 if (entry->size == 0)
1439 goto give_up;
1440 unuse_pack(&w_curs);
1441 return;
1442 }
1443
1444 /*
1445 * No choice but to fall back to the recursive delta walk
1446 * with sha1_object_info() to find about the object type
1447 * at this point...
1448 */
1449 give_up:
1450 unuse_pack(&w_curs);
1451 }
1452
1453 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1454 /*
1455 * The error condition is checked in prepare_pack(). This is
1456 * to permit a missing preferred base object to be ignored
1457 * as a preferred base. Doing so can result in a larger
1458 * pack file, but the transfer will still take place.
1459 */
1460}
1461
1462static int pack_offset_sort(const void *_a, const void *_b)
1463{
1464 const struct object_entry *a = *(struct object_entry **)_a;
1465 const struct object_entry *b = *(struct object_entry **)_b;
1466
1467 /* avoid filesystem trashing with loose objects */
1468 if (!a->in_pack && !b->in_pack)
1469 return hashcmp(a->idx.sha1, b->idx.sha1);
1470
1471 if (a->in_pack < b->in_pack)
1472 return -1;
1473 if (a->in_pack > b->in_pack)
1474 return 1;
1475 return a->in_pack_offset < b->in_pack_offset ? -1 :
1476 (a->in_pack_offset > b->in_pack_offset);
1477}
1478
1479static void get_object_details(void)
1480{
1481 uint32_t i;
1482 struct object_entry **sorted_by_offset;
1483
1484 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1485 for (i = 0; i < to_pack.nr_objects; i++)
1486 sorted_by_offset[i] = to_pack.objects + i;
1487 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1488
1489 for (i = 0; i < to_pack.nr_objects; i++) {
1490 struct object_entry *entry = sorted_by_offset[i];
1491 check_object(entry);
1492 if (big_file_threshold < entry->size)
1493 entry->no_try_delta = 1;
1494 }
1495
1496 free(sorted_by_offset);
1497}
1498
1499/*
1500 * We search for deltas in a list sorted by type, by filename hash, and then
1501 * by size, so that we see progressively smaller and smaller files.
1502 * That's because we prefer deltas to be from the bigger file
1503 * to the smaller -- deletes are potentially cheaper, but perhaps
1504 * more importantly, the bigger file is likely the more recent
1505 * one. The deepest deltas are therefore the oldest objects which are
1506 * less susceptible to be accessed often.
1507 */
1508static int type_size_sort(const void *_a, const void *_b)
1509{
1510 const struct object_entry *a = *(struct object_entry **)_a;
1511 const struct object_entry *b = *(struct object_entry **)_b;
1512
1513 if (a->type > b->type)
1514 return -1;
1515 if (a->type < b->type)
1516 return 1;
1517 if (a->hash > b->hash)
1518 return -1;
1519 if (a->hash < b->hash)
1520 return 1;
1521 if (a->preferred_base > b->preferred_base)
1522 return -1;
1523 if (a->preferred_base < b->preferred_base)
1524 return 1;
1525 if (a->size > b->size)
1526 return -1;
1527 if (a->size < b->size)
1528 return 1;
1529 return a < b ? -1 : (a > b); /* newest first */
1530}
1531
1532struct unpacked {
1533 struct object_entry *entry;
1534 void *data;
1535 struct delta_index *index;
1536 unsigned depth;
1537};
1538
1539static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1540 unsigned long delta_size)
1541{
1542 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1543 return 0;
1544
1545 if (delta_size < cache_max_small_delta_size)
1546 return 1;
1547
1548 /* cache delta, if objects are large enough compared to delta size */
1549 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1550 return 1;
1551
1552 return 0;
1553}
1554
1555#ifndef NO_PTHREADS
1556
1557static pthread_mutex_t read_mutex;
1558#define read_lock() pthread_mutex_lock(&read_mutex)
1559#define read_unlock() pthread_mutex_unlock(&read_mutex)
1560
1561static pthread_mutex_t cache_mutex;
1562#define cache_lock() pthread_mutex_lock(&cache_mutex)
1563#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1564
1565static pthread_mutex_t progress_mutex;
1566#define progress_lock() pthread_mutex_lock(&progress_mutex)
1567#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1568
1569#else
1570
1571#define read_lock() (void)0
1572#define read_unlock() (void)0
1573#define cache_lock() (void)0
1574#define cache_unlock() (void)0
1575#define progress_lock() (void)0
1576#define progress_unlock() (void)0
1577
1578#endif
1579
1580static int try_delta(struct unpacked *trg, struct unpacked *src,
1581 unsigned max_depth, unsigned long *mem_usage)
1582{
1583 struct object_entry *trg_entry = trg->entry;
1584 struct object_entry *src_entry = src->entry;
1585 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1586 unsigned ref_depth;
1587 enum object_type type;
1588 void *delta_buf;
1589
1590 /* Don't bother doing diffs between different types */
1591 if (trg_entry->type != src_entry->type)
1592 return -1;
1593
1594 /*
1595 * We do not bother to try a delta that we discarded on an
1596 * earlier try, but only when reusing delta data. Note that
1597 * src_entry that is marked as the preferred_base should always
1598 * be considered, as even if we produce a suboptimal delta against
1599 * it, we will still save the transfer cost, as we already know
1600 * the other side has it and we won't send src_entry at all.
1601 */
1602 if (reuse_delta && trg_entry->in_pack &&
1603 trg_entry->in_pack == src_entry->in_pack &&
1604 !src_entry->preferred_base &&
1605 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1606 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1607 return 0;
1608
1609 /* Let's not bust the allowed depth. */
1610 if (src->depth >= max_depth)
1611 return 0;
1612
1613 /* Now some size filtering heuristics. */
1614 trg_size = trg_entry->size;
1615 if (!trg_entry->delta) {
1616 max_size = trg_size/2 - 20;
1617 ref_depth = 1;
1618 } else {
1619 max_size = trg_entry->delta_size;
1620 ref_depth = trg->depth;
1621 }
1622 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1623 (max_depth - ref_depth + 1);
1624 if (max_size == 0)
1625 return 0;
1626 src_size = src_entry->size;
1627 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1628 if (sizediff >= max_size)
1629 return 0;
1630 if (trg_size < src_size / 32)
1631 return 0;
1632
1633 /* Load data if not already done */
1634 if (!trg->data) {
1635 read_lock();
1636 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1637 read_unlock();
1638 if (!trg->data)
1639 die("object %s cannot be read",
1640 sha1_to_hex(trg_entry->idx.sha1));
1641 if (sz != trg_size)
1642 die("object %s inconsistent object length (%lu vs %lu)",
1643 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1644 *mem_usage += sz;
1645 }
1646 if (!src->data) {
1647 read_lock();
1648 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1649 read_unlock();
1650 if (!src->data) {
1651 if (src_entry->preferred_base) {
1652 static int warned = 0;
1653 if (!warned++)
1654 warning("object %s cannot be read",
1655 sha1_to_hex(src_entry->idx.sha1));
1656 /*
1657 * Those objects are not included in the
1658 * resulting pack. Be resilient and ignore
1659 * them if they can't be read, in case the
1660 * pack could be created nevertheless.
1661 */
1662 return 0;
1663 }
1664 die("object %s cannot be read",
1665 sha1_to_hex(src_entry->idx.sha1));
1666 }
1667 if (sz != src_size)
1668 die("object %s inconsistent object length (%lu vs %lu)",
1669 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1670 *mem_usage += sz;
1671 }
1672 if (!src->index) {
1673 src->index = create_delta_index(src->data, src_size);
1674 if (!src->index) {
1675 static int warned = 0;
1676 if (!warned++)
1677 warning("suboptimal pack - out of memory");
1678 return 0;
1679 }
1680 *mem_usage += sizeof_delta_index(src->index);
1681 }
1682
1683 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1684 if (!delta_buf)
1685 return 0;
1686
1687 if (trg_entry->delta) {
1688 /* Prefer only shallower same-sized deltas. */
1689 if (delta_size == trg_entry->delta_size &&
1690 src->depth + 1 >= trg->depth) {
1691 free(delta_buf);
1692 return 0;
1693 }
1694 }
1695
1696 /*
1697 * Handle memory allocation outside of the cache
1698 * accounting lock. Compiler will optimize the strangeness
1699 * away when NO_PTHREADS is defined.
1700 */
1701 free(trg_entry->delta_data);
1702 cache_lock();
1703 if (trg_entry->delta_data) {
1704 delta_cache_size -= trg_entry->delta_size;
1705 trg_entry->delta_data = NULL;
1706 }
1707 if (delta_cacheable(src_size, trg_size, delta_size)) {
1708 delta_cache_size += delta_size;
1709 cache_unlock();
1710 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1711 } else {
1712 cache_unlock();
1713 free(delta_buf);
1714 }
1715
1716 trg_entry->delta = src_entry;
1717 trg_entry->delta_size = delta_size;
1718 trg->depth = src->depth + 1;
1719
1720 return 1;
1721}
1722
1723static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1724{
1725 struct object_entry *child = me->delta_child;
1726 unsigned int m = n;
1727 while (child) {
1728 unsigned int c = check_delta_limit(child, n + 1);
1729 if (m < c)
1730 m = c;
1731 child = child->delta_sibling;
1732 }
1733 return m;
1734}
1735
1736static unsigned long free_unpacked(struct unpacked *n)
1737{
1738 unsigned long freed_mem = sizeof_delta_index(n->index);
1739 free_delta_index(n->index);
1740 n->index = NULL;
1741 if (n->data) {
1742 freed_mem += n->entry->size;
1743 free(n->data);
1744 n->data = NULL;
1745 }
1746 n->entry = NULL;
1747 n->depth = 0;
1748 return freed_mem;
1749}
1750
1751static void find_deltas(struct object_entry **list, unsigned *list_size,
1752 int window, int depth, unsigned *processed)
1753{
1754 uint32_t i, idx = 0, count = 0;
1755 struct unpacked *array;
1756 unsigned long mem_usage = 0;
1757
1758 array = xcalloc(window, sizeof(struct unpacked));
1759
1760 for (;;) {
1761 struct object_entry *entry;
1762 struct unpacked *n = array + idx;
1763 int j, max_depth, best_base = -1;
1764
1765 progress_lock();
1766 if (!*list_size) {
1767 progress_unlock();
1768 break;
1769 }
1770 entry = *list++;
1771 (*list_size)--;
1772 if (!entry->preferred_base) {
1773 (*processed)++;
1774 display_progress(progress_state, *processed);
1775 }
1776 progress_unlock();
1777
1778 mem_usage -= free_unpacked(n);
1779 n->entry = entry;
1780
1781 while (window_memory_limit &&
1782 mem_usage > window_memory_limit &&
1783 count > 1) {
1784 uint32_t tail = (idx + window - count) % window;
1785 mem_usage -= free_unpacked(array + tail);
1786 count--;
1787 }
1788
1789 /* We do not compute delta to *create* objects we are not
1790 * going to pack.
1791 */
1792 if (entry->preferred_base)
1793 goto next;
1794
1795 /*
1796 * If the current object is at pack edge, take the depth the
1797 * objects that depend on the current object into account
1798 * otherwise they would become too deep.
1799 */
1800 max_depth = depth;
1801 if (entry->delta_child) {
1802 max_depth -= check_delta_limit(entry, 0);
1803 if (max_depth <= 0)
1804 goto next;
1805 }
1806
1807 j = window;
1808 while (--j > 0) {
1809 int ret;
1810 uint32_t other_idx = idx + j;
1811 struct unpacked *m;
1812 if (other_idx >= window)
1813 other_idx -= window;
1814 m = array + other_idx;
1815 if (!m->entry)
1816 break;
1817 ret = try_delta(n, m, max_depth, &mem_usage);
1818 if (ret < 0)
1819 break;
1820 else if (ret > 0)
1821 best_base = other_idx;
1822 }
1823
1824 /*
1825 * If we decided to cache the delta data, then it is best
1826 * to compress it right away. First because we have to do
1827 * it anyway, and doing it here while we're threaded will
1828 * save a lot of time in the non threaded write phase,
1829 * as well as allow for caching more deltas within
1830 * the same cache size limit.
1831 * ...
1832 * But only if not writing to stdout, since in that case
1833 * the network is most likely throttling writes anyway,
1834 * and therefore it is best to go to the write phase ASAP
1835 * instead, as we can afford spending more time compressing
1836 * between writes at that moment.
1837 */
1838 if (entry->delta_data && !pack_to_stdout) {
1839 entry->z_delta_size = do_compress(&entry->delta_data,
1840 entry->delta_size);
1841 cache_lock();
1842 delta_cache_size -= entry->delta_size;
1843 delta_cache_size += entry->z_delta_size;
1844 cache_unlock();
1845 }
1846
1847 /* if we made n a delta, and if n is already at max
1848 * depth, leaving it in the window is pointless. we
1849 * should evict it first.
1850 */
1851 if (entry->delta && max_depth <= n->depth)
1852 continue;
1853
1854 /*
1855 * Move the best delta base up in the window, after the
1856 * currently deltified object, to keep it longer. It will
1857 * be the first base object to be attempted next.
1858 */
1859 if (entry->delta) {
1860 struct unpacked swap = array[best_base];
1861 int dist = (window + idx - best_base) % window;
1862 int dst = best_base;
1863 while (dist--) {
1864 int src = (dst + 1) % window;
1865 array[dst] = array[src];
1866 dst = src;
1867 }
1868 array[dst] = swap;
1869 }
1870
1871 next:
1872 idx++;
1873 if (count + 1 < window)
1874 count++;
1875 if (idx >= window)
1876 idx = 0;
1877 }
1878
1879 for (i = 0; i < window; ++i) {
1880 free_delta_index(array[i].index);
1881 free(array[i].data);
1882 }
1883 free(array);
1884}
1885
1886#ifndef NO_PTHREADS
1887
1888static void try_to_free_from_threads(size_t size)
1889{
1890 read_lock();
1891 release_pack_memory(size);
1892 read_unlock();
1893}
1894
1895static try_to_free_t old_try_to_free_routine;
1896
1897/*
1898 * The main thread waits on the condition that (at least) one of the workers
1899 * has stopped working (which is indicated in the .working member of
1900 * struct thread_params).
1901 * When a work thread has completed its work, it sets .working to 0 and
1902 * signals the main thread and waits on the condition that .data_ready
1903 * becomes 1.
1904 */
1905
1906struct thread_params {
1907 pthread_t thread;
1908 struct object_entry **list;
1909 unsigned list_size;
1910 unsigned remaining;
1911 int window;
1912 int depth;
1913 int working;
1914 int data_ready;
1915 pthread_mutex_t mutex;
1916 pthread_cond_t cond;
1917 unsigned *processed;
1918};
1919
1920static pthread_cond_t progress_cond;
1921
1922/*
1923 * Mutex and conditional variable can't be statically-initialized on Windows.
1924 */
1925static void init_threaded_search(void)
1926{
1927 init_recursive_mutex(&read_mutex);
1928 pthread_mutex_init(&cache_mutex, NULL);
1929 pthread_mutex_init(&progress_mutex, NULL);
1930 pthread_cond_init(&progress_cond, NULL);
1931 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1932}
1933
1934static void cleanup_threaded_search(void)
1935{
1936 set_try_to_free_routine(old_try_to_free_routine);
1937 pthread_cond_destroy(&progress_cond);
1938 pthread_mutex_destroy(&read_mutex);
1939 pthread_mutex_destroy(&cache_mutex);
1940 pthread_mutex_destroy(&progress_mutex);
1941}
1942
1943static void *threaded_find_deltas(void *arg)
1944{
1945 struct thread_params *me = arg;
1946
1947 while (me->remaining) {
1948 find_deltas(me->list, &me->remaining,
1949 me->window, me->depth, me->processed);
1950
1951 progress_lock();
1952 me->working = 0;
1953 pthread_cond_signal(&progress_cond);
1954 progress_unlock();
1955
1956 /*
1957 * We must not set ->data_ready before we wait on the
1958 * condition because the main thread may have set it to 1
1959 * before we get here. In order to be sure that new
1960 * work is available if we see 1 in ->data_ready, it
1961 * was initialized to 0 before this thread was spawned
1962 * and we reset it to 0 right away.
1963 */
1964 pthread_mutex_lock(&me->mutex);
1965 while (!me->data_ready)
1966 pthread_cond_wait(&me->cond, &me->mutex);
1967 me->data_ready = 0;
1968 pthread_mutex_unlock(&me->mutex);
1969 }
1970 /* leave ->working 1 so that this doesn't get more work assigned */
1971 return NULL;
1972}
1973
1974static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1975 int window, int depth, unsigned *processed)
1976{
1977 struct thread_params *p;
1978 int i, ret, active_threads = 0;
1979
1980 init_threaded_search();
1981
1982 if (delta_search_threads <= 1) {
1983 find_deltas(list, &list_size, window, depth, processed);
1984 cleanup_threaded_search();
1985 return;
1986 }
1987 if (progress > pack_to_stdout)
1988 fprintf(stderr, "Delta compression using up to %d threads.\n",
1989 delta_search_threads);
1990 p = xcalloc(delta_search_threads, sizeof(*p));
1991
1992 /* Partition the work amongst work threads. */
1993 for (i = 0; i < delta_search_threads; i++) {
1994 unsigned sub_size = list_size / (delta_search_threads - i);
1995
1996 /* don't use too small segments or no deltas will be found */
1997 if (sub_size < 2*window && i+1 < delta_search_threads)
1998 sub_size = 0;
1999
2000 p[i].window = window;
2001 p[i].depth = depth;
2002 p[i].processed = processed;
2003 p[i].working = 1;
2004 p[i].data_ready = 0;
2005
2006 /* try to split chunks on "path" boundaries */
2007 while (sub_size && sub_size < list_size &&
2008 list[sub_size]->hash &&
2009 list[sub_size]->hash == list[sub_size-1]->hash)
2010 sub_size++;
2011
2012 p[i].list = list;
2013 p[i].list_size = sub_size;
2014 p[i].remaining = sub_size;
2015
2016 list += sub_size;
2017 list_size -= sub_size;
2018 }
2019
2020 /* Start work threads. */
2021 for (i = 0; i < delta_search_threads; i++) {
2022 if (!p[i].list_size)
2023 continue;
2024 pthread_mutex_init(&p[i].mutex, NULL);
2025 pthread_cond_init(&p[i].cond, NULL);
2026 ret = pthread_create(&p[i].thread, NULL,
2027 threaded_find_deltas, &p[i]);
2028 if (ret)
2029 die("unable to create thread: %s", strerror(ret));
2030 active_threads++;
2031 }
2032
2033 /*
2034 * Now let's wait for work completion. Each time a thread is done
2035 * with its work, we steal half of the remaining work from the
2036 * thread with the largest number of unprocessed objects and give
2037 * it to that newly idle thread. This ensure good load balancing
2038 * until the remaining object list segments are simply too short
2039 * to be worth splitting anymore.
2040 */
2041 while (active_threads) {
2042 struct thread_params *target = NULL;
2043 struct thread_params *victim = NULL;
2044 unsigned sub_size = 0;
2045
2046 progress_lock();
2047 for (;;) {
2048 for (i = 0; !target && i < delta_search_threads; i++)
2049 if (!p[i].working)
2050 target = &p[i];
2051 if (target)
2052 break;
2053 pthread_cond_wait(&progress_cond, &progress_mutex);
2054 }
2055
2056 for (i = 0; i < delta_search_threads; i++)
2057 if (p[i].remaining > 2*window &&
2058 (!victim || victim->remaining < p[i].remaining))
2059 victim = &p[i];
2060 if (victim) {
2061 sub_size = victim->remaining / 2;
2062 list = victim->list + victim->list_size - sub_size;
2063 while (sub_size && list[0]->hash &&
2064 list[0]->hash == list[-1]->hash) {
2065 list++;
2066 sub_size--;
2067 }
2068 if (!sub_size) {
2069 /*
2070 * It is possible for some "paths" to have
2071 * so many objects that no hash boundary
2072 * might be found. Let's just steal the
2073 * exact half in that case.
2074 */
2075 sub_size = victim->remaining / 2;
2076 list -= sub_size;
2077 }
2078 target->list = list;
2079 victim->list_size -= sub_size;
2080 victim->remaining -= sub_size;
2081 }
2082 target->list_size = sub_size;
2083 target->remaining = sub_size;
2084 target->working = 1;
2085 progress_unlock();
2086
2087 pthread_mutex_lock(&target->mutex);
2088 target->data_ready = 1;
2089 pthread_cond_signal(&target->cond);
2090 pthread_mutex_unlock(&target->mutex);
2091
2092 if (!sub_size) {
2093 pthread_join(target->thread, NULL);
2094 pthread_cond_destroy(&target->cond);
2095 pthread_mutex_destroy(&target->mutex);
2096 active_threads--;
2097 }
2098 }
2099 cleanup_threaded_search();
2100 free(p);
2101}
2102
2103#else
2104#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2105#endif
2106
2107static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2108{
2109 struct object_id peeled;
2110
2111 if (starts_with(path, "refs/tags/") && /* is a tag? */
2112 !peel_ref(path, peeled.hash) && /* peelable? */
2113 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2114 add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2115 return 0;
2116}
2117
2118static void prepare_pack(int window, int depth)
2119{
2120 struct object_entry **delta_list;
2121 uint32_t i, nr_deltas;
2122 unsigned n;
2123
2124 get_object_details();
2125
2126 /*
2127 * If we're locally repacking then we need to be doubly careful
2128 * from now on in order to make sure no stealth corruption gets
2129 * propagated to the new pack. Clients receiving streamed packs
2130 * should validate everything they get anyway so no need to incur
2131 * the additional cost here in that case.
2132 */
2133 if (!pack_to_stdout)
2134 do_check_packed_object_crc = 1;
2135
2136 if (!to_pack.nr_objects || !window || !depth)
2137 return;
2138
2139 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2140 nr_deltas = n = 0;
2141
2142 for (i = 0; i < to_pack.nr_objects; i++) {
2143 struct object_entry *entry = to_pack.objects + i;
2144
2145 if (entry->delta)
2146 /* This happens if we decided to reuse existing
2147 * delta from a pack. "reuse_delta &&" is implied.
2148 */
2149 continue;
2150
2151 if (entry->size < 50)
2152 continue;
2153
2154 if (entry->no_try_delta)
2155 continue;
2156
2157 if (!entry->preferred_base) {
2158 nr_deltas++;
2159 if (entry->type < 0)
2160 die("unable to get type of object %s",
2161 sha1_to_hex(entry->idx.sha1));
2162 } else {
2163 if (entry->type < 0) {
2164 /*
2165 * This object is not found, but we
2166 * don't have to include it anyway.
2167 */
2168 continue;
2169 }
2170 }
2171
2172 delta_list[n++] = entry;
2173 }
2174
2175 if (nr_deltas && n > 1) {
2176 unsigned nr_done = 0;
2177 if (progress)
2178 progress_state = start_progress(_("Compressing objects"),
2179 nr_deltas);
2180 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2181 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2182 stop_progress(&progress_state);
2183 if (nr_done != nr_deltas)
2184 die("inconsistency with delta count");
2185 }
2186 free(delta_list);
2187}
2188
2189static int git_pack_config(const char *k, const char *v, void *cb)
2190{
2191 if (!strcmp(k, "pack.window")) {
2192 window = git_config_int(k, v);
2193 return 0;
2194 }
2195 if (!strcmp(k, "pack.windowmemory")) {
2196 window_memory_limit = git_config_ulong(k, v);
2197 return 0;
2198 }
2199 if (!strcmp(k, "pack.depth")) {
2200 depth = git_config_int(k, v);
2201 return 0;
2202 }
2203 if (!strcmp(k, "pack.compression")) {
2204 int level = git_config_int(k, v);
2205 if (level == -1)
2206 level = Z_DEFAULT_COMPRESSION;
2207 else if (level < 0 || level > Z_BEST_COMPRESSION)
2208 die("bad pack compression level %d", level);
2209 pack_compression_level = level;
2210 pack_compression_seen = 1;
2211 return 0;
2212 }
2213 if (!strcmp(k, "pack.deltacachesize")) {
2214 max_delta_cache_size = git_config_int(k, v);
2215 return 0;
2216 }
2217 if (!strcmp(k, "pack.deltacachelimit")) {
2218 cache_max_small_delta_size = git_config_int(k, v);
2219 return 0;
2220 }
2221 if (!strcmp(k, "pack.writebitmaphashcache")) {
2222 if (git_config_bool(k, v))
2223 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2224 else
2225 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2226 }
2227 if (!strcmp(k, "pack.usebitmaps")) {
2228 use_bitmap_index = git_config_bool(k, v);
2229 return 0;
2230 }
2231 if (!strcmp(k, "pack.threads")) {
2232 delta_search_threads = git_config_int(k, v);
2233 if (delta_search_threads < 0)
2234 die("invalid number of threads specified (%d)",
2235 delta_search_threads);
2236#ifdef NO_PTHREADS
2237 if (delta_search_threads != 1)
2238 warning("no threads support, ignoring %s", k);
2239#endif
2240 return 0;
2241 }
2242 if (!strcmp(k, "pack.indexversion")) {
2243 pack_idx_opts.version = git_config_int(k, v);
2244 if (pack_idx_opts.version > 2)
2245 die("bad pack.indexversion=%"PRIu32,
2246 pack_idx_opts.version);
2247 return 0;
2248 }
2249 return git_default_config(k, v, cb);
2250}
2251
2252static void read_object_list_from_stdin(void)
2253{
2254 char line[40 + 1 + PATH_MAX + 2];
2255 unsigned char sha1[20];
2256
2257 for (;;) {
2258 if (!fgets(line, sizeof(line), stdin)) {
2259 if (feof(stdin))
2260 break;
2261 if (!ferror(stdin))
2262 die("fgets returned NULL, not EOF, not error!");
2263 if (errno != EINTR)
2264 die_errno("fgets");
2265 clearerr(stdin);
2266 continue;
2267 }
2268 if (line[0] == '-') {
2269 if (get_sha1_hex(line+1, sha1))
2270 die("expected edge sha1, got garbage:\n %s",
2271 line);
2272 add_preferred_base(sha1);
2273 continue;
2274 }
2275 if (get_sha1_hex(line, sha1))
2276 die("expected sha1, got garbage:\n %s", line);
2277
2278 add_preferred_base_object(line+41);
2279 add_object_entry(sha1, 0, line+41, 0);
2280 }
2281}
2282
2283#define OBJECT_ADDED (1u<<20)
2284
2285static void show_commit(struct commit *commit, void *data)
2286{
2287 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2288 commit->object.flags |= OBJECT_ADDED;
2289
2290 if (write_bitmap_index)
2291 index_commit_for_bitmap(commit);
2292}
2293
2294static void show_object(struct object *obj, const char *name, void *data)
2295{
2296 add_preferred_base_object(name);
2297 add_object_entry(obj->oid.hash, obj->type, name, 0);
2298 obj->flags |= OBJECT_ADDED;
2299}
2300
2301static void show_edge(struct commit *commit)
2302{
2303 add_preferred_base(commit->object.oid.hash);
2304}
2305
2306struct in_pack_object {
2307 off_t offset;
2308 struct object *object;
2309};
2310
2311struct in_pack {
2312 int alloc;
2313 int nr;
2314 struct in_pack_object *array;
2315};
2316
2317static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2318{
2319 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2320 in_pack->array[in_pack->nr].object = object;
2321 in_pack->nr++;
2322}
2323
2324/*
2325 * Compare the objects in the offset order, in order to emulate the
2326 * "git rev-list --objects" output that produced the pack originally.
2327 */
2328static int ofscmp(const void *a_, const void *b_)
2329{
2330 struct in_pack_object *a = (struct in_pack_object *)a_;
2331 struct in_pack_object *b = (struct in_pack_object *)b_;
2332
2333 if (a->offset < b->offset)
2334 return -1;
2335 else if (a->offset > b->offset)
2336 return 1;
2337 else
2338 return oidcmp(&a->object->oid, &b->object->oid);
2339}
2340
2341static void add_objects_in_unpacked_packs(struct rev_info *revs)
2342{
2343 struct packed_git *p;
2344 struct in_pack in_pack;
2345 uint32_t i;
2346
2347 memset(&in_pack, 0, sizeof(in_pack));
2348
2349 for (p = packed_git; p; p = p->next) {
2350 const unsigned char *sha1;
2351 struct object *o;
2352
2353 if (!p->pack_local || p->pack_keep)
2354 continue;
2355 if (open_pack_index(p))
2356 die("cannot open pack index");
2357
2358 ALLOC_GROW(in_pack.array,
2359 in_pack.nr + p->num_objects,
2360 in_pack.alloc);
2361
2362 for (i = 0; i < p->num_objects; i++) {
2363 sha1 = nth_packed_object_sha1(p, i);
2364 o = lookup_unknown_object(sha1);
2365 if (!(o->flags & OBJECT_ADDED))
2366 mark_in_pack_object(o, p, &in_pack);
2367 o->flags |= OBJECT_ADDED;
2368 }
2369 }
2370
2371 if (in_pack.nr) {
2372 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2373 ofscmp);
2374 for (i = 0; i < in_pack.nr; i++) {
2375 struct object *o = in_pack.array[i].object;
2376 add_object_entry(o->oid.hash, o->type, "", 0);
2377 }
2378 }
2379 free(in_pack.array);
2380}
2381
2382static int add_loose_object(const unsigned char *sha1, const char *path,
2383 void *data)
2384{
2385 enum object_type type = sha1_object_info(sha1, NULL);
2386
2387 if (type < 0) {
2388 warning("loose object at %s could not be examined", path);
2389 return 0;
2390 }
2391
2392 add_object_entry(sha1, type, "", 0);
2393 return 0;
2394}
2395
2396/*
2397 * We actually don't even have to worry about reachability here.
2398 * add_object_entry will weed out duplicates, so we just add every
2399 * loose object we find.
2400 */
2401static void add_unreachable_loose_objects(void)
2402{
2403 for_each_loose_file_in_objdir(get_object_directory(),
2404 add_loose_object,
2405 NULL, NULL, NULL);
2406}
2407
2408static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2409{
2410 static struct packed_git *last_found = (void *)1;
2411 struct packed_git *p;
2412
2413 p = (last_found != (void *)1) ? last_found : packed_git;
2414
2415 while (p) {
2416 if ((!p->pack_local || p->pack_keep) &&
2417 find_pack_entry_one(sha1, p)) {
2418 last_found = p;
2419 return 1;
2420 }
2421 if (p == last_found)
2422 p = packed_git;
2423 else
2424 p = p->next;
2425 if (p == last_found)
2426 p = p->next;
2427 }
2428 return 0;
2429}
2430
2431/*
2432 * Store a list of sha1s that are should not be discarded
2433 * because they are either written too recently, or are
2434 * reachable from another object that was.
2435 *
2436 * This is filled by get_object_list.
2437 */
2438static struct sha1_array recent_objects;
2439
2440static int loosened_object_can_be_discarded(const unsigned char *sha1,
2441 unsigned long mtime)
2442{
2443 if (!unpack_unreachable_expiration)
2444 return 0;
2445 if (mtime > unpack_unreachable_expiration)
2446 return 0;
2447 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2448 return 0;
2449 return 1;
2450}
2451
2452static void loosen_unused_packed_objects(struct rev_info *revs)
2453{
2454 struct packed_git *p;
2455 uint32_t i;
2456 const unsigned char *sha1;
2457
2458 for (p = packed_git; p; p = p->next) {
2459 if (!p->pack_local || p->pack_keep)
2460 continue;
2461
2462 if (open_pack_index(p))
2463 die("cannot open pack index");
2464
2465 for (i = 0; i < p->num_objects; i++) {
2466 sha1 = nth_packed_object_sha1(p, i);
2467 if (!packlist_find(&to_pack, sha1, NULL) &&
2468 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2469 !loosened_object_can_be_discarded(sha1, p->mtime))
2470 if (force_object_loose(sha1, p->mtime))
2471 die("unable to force loose object");
2472 }
2473 }
2474}
2475
2476/*
2477 * This tracks any options which a reader of the pack might
2478 * not understand, and which would therefore prevent blind reuse
2479 * of what we have on disk.
2480 */
2481static int pack_options_allow_reuse(void)
2482{
2483 return allow_ofs_delta;
2484}
2485
2486static int get_object_list_from_bitmap(struct rev_info *revs)
2487{
2488 if (prepare_bitmap_walk(revs) < 0)
2489 return -1;
2490
2491 if (pack_options_allow_reuse() &&
2492 !reuse_partial_packfile_from_bitmap(
2493 &reuse_packfile,
2494 &reuse_packfile_objects,
2495 &reuse_packfile_offset)) {
2496 assert(reuse_packfile_objects);
2497 nr_result += reuse_packfile_objects;
2498 display_progress(progress_state, nr_result);
2499 }
2500
2501 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2502 return 0;
2503}
2504
2505static void record_recent_object(struct object *obj,
2506 const char *name,
2507 void *data)
2508{
2509 sha1_array_append(&recent_objects, obj->oid.hash);
2510}
2511
2512static void record_recent_commit(struct commit *commit, void *data)
2513{
2514 sha1_array_append(&recent_objects, commit->object.oid.hash);
2515}
2516
2517static void get_object_list(int ac, const char **av)
2518{
2519 struct rev_info revs;
2520 char line[1000];
2521 int flags = 0;
2522
2523 init_revisions(&revs, NULL);
2524 save_commit_buffer = 0;
2525 setup_revisions(ac, av, &revs, NULL);
2526
2527 /* make sure shallows are read */
2528 is_repository_shallow();
2529
2530 while (fgets(line, sizeof(line), stdin) != NULL) {
2531 int len = strlen(line);
2532 if (len && line[len - 1] == '\n')
2533 line[--len] = 0;
2534 if (!len)
2535 break;
2536 if (*line == '-') {
2537 if (!strcmp(line, "--not")) {
2538 flags ^= UNINTERESTING;
2539 write_bitmap_index = 0;
2540 continue;
2541 }
2542 if (starts_with(line, "--shallow ")) {
2543 unsigned char sha1[20];
2544 if (get_sha1_hex(line + 10, sha1))
2545 die("not an SHA-1 '%s'", line + 10);
2546 register_shallow(sha1);
2547 use_bitmap_index = 0;
2548 continue;
2549 }
2550 die("not a rev '%s'", line);
2551 }
2552 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2553 die("bad revision '%s'", line);
2554 }
2555
2556 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2557 return;
2558
2559 if (prepare_revision_walk(&revs))
2560 die("revision walk setup failed");
2561 mark_edges_uninteresting(&revs, show_edge);
2562 traverse_commit_list(&revs, show_commit, show_object, NULL);
2563
2564 if (unpack_unreachable_expiration) {
2565 revs.ignore_missing_links = 1;
2566 if (add_unseen_recent_objects_to_traversal(&revs,
2567 unpack_unreachable_expiration))
2568 die("unable to add recent objects");
2569 if (prepare_revision_walk(&revs))
2570 die("revision walk setup failed");
2571 traverse_commit_list(&revs, record_recent_commit,
2572 record_recent_object, NULL);
2573 }
2574
2575 if (keep_unreachable)
2576 add_objects_in_unpacked_packs(&revs);
2577 if (pack_loose_unreachable)
2578 add_unreachable_loose_objects();
2579 if (unpack_unreachable)
2580 loosen_unused_packed_objects(&revs);
2581
2582 sha1_array_clear(&recent_objects);
2583}
2584
2585static int option_parse_index_version(const struct option *opt,
2586 const char *arg, int unset)
2587{
2588 char *c;
2589 const char *val = arg;
2590 pack_idx_opts.version = strtoul(val, &c, 10);
2591 if (pack_idx_opts.version > 2)
2592 die(_("unsupported index version %s"), val);
2593 if (*c == ',' && c[1])
2594 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2595 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2596 die(_("bad index version '%s'"), val);
2597 return 0;
2598}
2599
2600static int option_parse_unpack_unreachable(const struct option *opt,
2601 const char *arg, int unset)
2602{
2603 if (unset) {
2604 unpack_unreachable = 0;
2605 unpack_unreachable_expiration = 0;
2606 }
2607 else {
2608 unpack_unreachable = 1;
2609 if (arg)
2610 unpack_unreachable_expiration = approxidate(arg);
2611 }
2612 return 0;
2613}
2614
2615int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2616{
2617 int use_internal_rev_list = 0;
2618 int thin = 0;
2619 int shallow = 0;
2620 int all_progress_implied = 0;
2621 struct argv_array rp = ARGV_ARRAY_INIT;
2622 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2623 int rev_list_index = 0;
2624 struct option pack_objects_options[] = {
2625 OPT_SET_INT('q', "quiet", &progress,
2626 N_("do not show progress meter"), 0),
2627 OPT_SET_INT(0, "progress", &progress,
2628 N_("show progress meter"), 1),
2629 OPT_SET_INT(0, "all-progress", &progress,
2630 N_("show progress meter during object writing phase"), 2),
2631 OPT_BOOL(0, "all-progress-implied",
2632 &all_progress_implied,
2633 N_("similar to --all-progress when progress meter is shown")),
2634 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2635 N_("write the pack index file in the specified idx format version"),
2636 0, option_parse_index_version },
2637 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2638 N_("maximum size of each output pack file")),
2639 OPT_BOOL(0, "local", &local,
2640 N_("ignore borrowed objects from alternate object store")),
2641 OPT_BOOL(0, "incremental", &incremental,
2642 N_("ignore packed objects")),
2643 OPT_INTEGER(0, "window", &window,
2644 N_("limit pack window by objects")),
2645 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2646 N_("limit pack window by memory in addition to object limit")),
2647 OPT_INTEGER(0, "depth", &depth,
2648 N_("maximum length of delta chain allowed in the resulting pack")),
2649 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2650 N_("reuse existing deltas")),
2651 OPT_BOOL(0, "reuse-object", &reuse_object,
2652 N_("reuse existing objects")),
2653 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2654 N_("use OFS_DELTA objects")),
2655 OPT_INTEGER(0, "threads", &delta_search_threads,
2656 N_("use threads when searching for best delta matches")),
2657 OPT_BOOL(0, "non-empty", &non_empty,
2658 N_("do not create an empty pack output")),
2659 OPT_BOOL(0, "revs", &use_internal_rev_list,
2660 N_("read revision arguments from standard input")),
2661 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2662 N_("limit the objects to those that are not yet packed"),
2663 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2664 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2665 N_("include objects reachable from any reference"),
2666 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2667 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2668 N_("include objects referred by reflog entries"),
2669 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2670 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2671 N_("include objects referred to by the index"),
2672 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2673 OPT_BOOL(0, "stdout", &pack_to_stdout,
2674 N_("output pack to stdout")),
2675 OPT_BOOL(0, "include-tag", &include_tag,
2676 N_("include tag objects that refer to objects to be packed")),
2677 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2678 N_("keep unreachable objects")),
2679 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2680 N_("pack loose unreachable objects")),
2681 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2682 N_("unpack unreachable objects newer than <time>"),
2683 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2684 OPT_BOOL(0, "thin", &thin,
2685 N_("create thin packs")),
2686 OPT_BOOL(0, "shallow", &shallow,
2687 N_("create packs suitable for shallow fetches")),
2688 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2689 N_("ignore packs that have companion .keep file")),
2690 OPT_INTEGER(0, "compression", &pack_compression_level,
2691 N_("pack compression level")),
2692 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2693 N_("do not hide commits by grafts"), 0),
2694 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2695 N_("use a bitmap index if available to speed up counting objects")),
2696 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2697 N_("write a bitmap index together with the pack index")),
2698 OPT_END(),
2699 };
2700
2701 check_replace_refs = 0;
2702
2703 reset_pack_idx_option(&pack_idx_opts);
2704 git_config(git_pack_config, NULL);
2705 if (!pack_compression_seen && core_compression_seen)
2706 pack_compression_level = core_compression_level;
2707
2708 progress = isatty(2);
2709 argc = parse_options(argc, argv, prefix, pack_objects_options,
2710 pack_usage, 0);
2711
2712 if (argc) {
2713 base_name = argv[0];
2714 argc--;
2715 }
2716 if (pack_to_stdout != !base_name || argc)
2717 usage_with_options(pack_usage, pack_objects_options);
2718
2719 argv_array_push(&rp, "pack-objects");
2720 if (thin) {
2721 use_internal_rev_list = 1;
2722 argv_array_push(&rp, shallow
2723 ? "--objects-edge-aggressive"
2724 : "--objects-edge");
2725 } else
2726 argv_array_push(&rp, "--objects");
2727
2728 if (rev_list_all) {
2729 use_internal_rev_list = 1;
2730 argv_array_push(&rp, "--all");
2731 }
2732 if (rev_list_reflog) {
2733 use_internal_rev_list = 1;
2734 argv_array_push(&rp, "--reflog");
2735 }
2736 if (rev_list_index) {
2737 use_internal_rev_list = 1;
2738 argv_array_push(&rp, "--indexed-objects");
2739 }
2740 if (rev_list_unpacked) {
2741 use_internal_rev_list = 1;
2742 argv_array_push(&rp, "--unpacked");
2743 }
2744
2745 if (!reuse_object)
2746 reuse_delta = 0;
2747 if (pack_compression_level == -1)
2748 pack_compression_level = Z_DEFAULT_COMPRESSION;
2749 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2750 die("bad pack compression level %d", pack_compression_level);
2751
2752 if (!delta_search_threads) /* --threads=0 means autodetect */
2753 delta_search_threads = online_cpus();
2754
2755#ifdef NO_PTHREADS
2756 if (delta_search_threads != 1)
2757 warning("no threads support, ignoring --threads");
2758#endif
2759 if (!pack_to_stdout && !pack_size_limit)
2760 pack_size_limit = pack_size_limit_cfg;
2761 if (pack_to_stdout && pack_size_limit)
2762 die("--max-pack-size cannot be used to build a pack for transfer.");
2763 if (pack_size_limit && pack_size_limit < 1024*1024) {
2764 warning("minimum pack size limit is 1 MiB");
2765 pack_size_limit = 1024*1024;
2766 }
2767
2768 if (!pack_to_stdout && thin)
2769 die("--thin cannot be used to build an indexable pack.");
2770
2771 if (keep_unreachable && unpack_unreachable)
2772 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2773 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2774 unpack_unreachable_expiration = 0;
2775
2776 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2777 use_bitmap_index = 0;
2778
2779 if (pack_to_stdout || !rev_list_all)
2780 write_bitmap_index = 0;
2781
2782 if (progress && all_progress_implied)
2783 progress = 2;
2784
2785 prepare_packed_git();
2786
2787 if (progress)
2788 progress_state = start_progress(_("Counting objects"), 0);
2789 if (!use_internal_rev_list)
2790 read_object_list_from_stdin();
2791 else {
2792 get_object_list(rp.argc, rp.argv);
2793 argv_array_clear(&rp);
2794 }
2795 cleanup_preferred_base();
2796 if (include_tag && nr_result)
2797 for_each_ref(add_ref_tag, NULL);
2798 stop_progress(&progress_state);
2799
2800 if (non_empty && !nr_result)
2801 return 0;
2802 if (nr_result)
2803 prepare_pack(window, depth);
2804 write_pack_file();
2805 if (progress)
2806 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2807 " reused %"PRIu32" (delta %"PRIu32")\n",
2808 written, written_delta, reused, reused_delta);
2809 return 0;
2810}