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