f0474db9914e5d5a28085feed68cf979601eb5c1
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
24static const char *pack_usage[] = {
25 N_("git pack-objects --stdout [options...] [< ref-list | < object-list]"),
26 N_("git pack-objects [options...] base-name [< ref-list | < object-list]"),
27 NULL
28};
29
30/*
31 * Objects we are going to pack are collected in the `to_pack` structure.
32 * It contains an array (dynamically expanded) of the object data, and a map
33 * that can resolve SHA1s to their position in the array.
34 */
35static struct packing_data to_pack;
36
37static struct pack_idx_entry **written_list;
38static uint32_t nr_result, nr_written;
39
40static int non_empty;
41static int reuse_delta = 1, reuse_object = 1;
42static int keep_unreachable, unpack_unreachable, include_tag;
43static unsigned long unpack_unreachable_expiration;
44static int local;
45static int incremental;
46static int ignore_packed_keep;
47static int allow_ofs_delta;
48static struct pack_idx_option pack_idx_opts;
49static const char *base_name;
50static int progress = 1;
51static int window = 10;
52static unsigned long pack_size_limit;
53static int depth = 50;
54static int delta_search_threads;
55static int pack_to_stdout;
56static int num_preferred_base;
57static struct progress *progress_state;
58static int pack_compression_level = Z_DEFAULT_COMPRESSION;
59static int pack_compression_seen;
60
61static struct packed_git *reuse_packfile;
62static uint32_t reuse_packfile_objects;
63static off_t reuse_packfile_offset;
64
65static int use_bitmap_index = 1;
66static int write_bitmap_index;
67static uint16_t write_bitmap_options;
68
69static unsigned long delta_cache_size = 0;
70static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
71static unsigned long cache_max_small_delta_size = 1000;
72
73static unsigned long window_memory_limit = 0;
74
75/*
76 * stats
77 */
78static uint32_t written, written_delta;
79static uint32_t reused, reused_delta;
80
81/*
82 * Indexed commits
83 */
84static struct commit **indexed_commits;
85static unsigned int indexed_commits_nr;
86static unsigned int indexed_commits_alloc;
87
88static void index_commit_for_bitmap(struct commit *commit)
89{
90 if (indexed_commits_nr >= indexed_commits_alloc) {
91 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
92 indexed_commits = xrealloc(indexed_commits,
93 indexed_commits_alloc * sizeof(struct commit *));
94 }
95
96 indexed_commits[indexed_commits_nr++] = commit;
97}
98
99static void *get_delta(struct object_entry *entry)
100{
101 unsigned long size, base_size, delta_size;
102 void *buf, *base_buf, *delta_buf;
103 enum object_type type;
104
105 buf = read_sha1_file(entry->idx.sha1, &type, &size);
106 if (!buf)
107 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
108 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
109 if (!base_buf)
110 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
111 delta_buf = diff_delta(base_buf, base_size,
112 buf, size, &delta_size, 0);
113 if (!delta_buf || delta_size != entry->delta_size)
114 die("delta size changed");
115 free(buf);
116 free(base_buf);
117 return delta_buf;
118}
119
120static unsigned long do_compress(void **pptr, unsigned long size)
121{
122 git_zstream stream;
123 void *in, *out;
124 unsigned long maxsize;
125
126 memset(&stream, 0, sizeof(stream));
127 git_deflate_init(&stream, pack_compression_level);
128 maxsize = git_deflate_bound(&stream, size);
129
130 in = *pptr;
131 out = xmalloc(maxsize);
132 *pptr = out;
133
134 stream.next_in = in;
135 stream.avail_in = size;
136 stream.next_out = out;
137 stream.avail_out = maxsize;
138 while (git_deflate(&stream, Z_FINISH) == Z_OK)
139 ; /* nothing */
140 git_deflate_end(&stream);
141
142 free(in);
143 return stream.total_out;
144}
145
146static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
147 const unsigned char *sha1)
148{
149 git_zstream stream;
150 unsigned char ibuf[1024 * 16];
151 unsigned char obuf[1024 * 16];
152 unsigned long olen = 0;
153
154 memset(&stream, 0, sizeof(stream));
155 git_deflate_init(&stream, pack_compression_level);
156
157 for (;;) {
158 ssize_t readlen;
159 int zret = Z_OK;
160 readlen = read_istream(st, ibuf, sizeof(ibuf));
161 if (readlen == -1)
162 die(_("unable to read %s"), sha1_to_hex(sha1));
163
164 stream.next_in = ibuf;
165 stream.avail_in = readlen;
166 while ((stream.avail_in || readlen == 0) &&
167 (zret == Z_OK || zret == Z_BUF_ERROR)) {
168 stream.next_out = obuf;
169 stream.avail_out = sizeof(obuf);
170 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
171 sha1write(f, obuf, stream.next_out - obuf);
172 olen += stream.next_out - obuf;
173 }
174 if (stream.avail_in)
175 die(_("deflate error (%d)"), zret);
176 if (readlen == 0) {
177 if (zret != Z_STREAM_END)
178 die(_("deflate error (%d)"), zret);
179 break;
180 }
181 }
182 git_deflate_end(&stream);
183 return olen;
184}
185
186/*
187 * we are going to reuse the existing object data as is. make
188 * sure it is not corrupt.
189 */
190static int check_pack_inflate(struct packed_git *p,
191 struct pack_window **w_curs,
192 off_t offset,
193 off_t len,
194 unsigned long expect)
195{
196 git_zstream stream;
197 unsigned char fakebuf[4096], *in;
198 int st;
199
200 memset(&stream, 0, sizeof(stream));
201 git_inflate_init(&stream);
202 do {
203 in = use_pack(p, w_curs, offset, &stream.avail_in);
204 stream.next_in = in;
205 stream.next_out = fakebuf;
206 stream.avail_out = sizeof(fakebuf);
207 st = git_inflate(&stream, Z_FINISH);
208 offset += stream.next_in - in;
209 } while (st == Z_OK || st == Z_BUF_ERROR);
210 git_inflate_end(&stream);
211 return (st == Z_STREAM_END &&
212 stream.total_out == expect &&
213 stream.total_in == len) ? 0 : -1;
214}
215
216static void copy_pack_data(struct sha1file *f,
217 struct packed_git *p,
218 struct pack_window **w_curs,
219 off_t offset,
220 off_t len)
221{
222 unsigned char *in;
223 unsigned long avail;
224
225 while (len) {
226 in = use_pack(p, w_curs, offset, &avail);
227 if (avail > len)
228 avail = (unsigned long)len;
229 sha1write(f, in, avail);
230 offset += avail;
231 len -= avail;
232 }
233}
234
235/* Return 0 if we will bust the pack-size limit */
236static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
237 unsigned long limit, int usable_delta)
238{
239 unsigned long size, datalen;
240 unsigned char header[10], dheader[10];
241 unsigned hdrlen;
242 enum object_type type;
243 void *buf;
244 struct git_istream *st = NULL;
245
246 if (!usable_delta) {
247 if (entry->type == OBJ_BLOB &&
248 entry->size > big_file_threshold &&
249 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
250 buf = NULL;
251 else {
252 buf = read_sha1_file(entry->idx.sha1, &type, &size);
253 if (!buf)
254 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
255 }
256 /*
257 * make sure no cached delta data remains from a
258 * previous attempt before a pack split occurred.
259 */
260 free(entry->delta_data);
261 entry->delta_data = NULL;
262 entry->z_delta_size = 0;
263 } else if (entry->delta_data) {
264 size = entry->delta_size;
265 buf = entry->delta_data;
266 entry->delta_data = NULL;
267 type = (allow_ofs_delta && entry->delta->idx.offset) ?
268 OBJ_OFS_DELTA : OBJ_REF_DELTA;
269 } else {
270 buf = get_delta(entry);
271 size = entry->delta_size;
272 type = (allow_ofs_delta && entry->delta->idx.offset) ?
273 OBJ_OFS_DELTA : OBJ_REF_DELTA;
274 }
275
276 if (st) /* large blob case, just assume we don't compress well */
277 datalen = size;
278 else if (entry->z_delta_size)
279 datalen = entry->z_delta_size;
280 else
281 datalen = do_compress(&buf, size);
282
283 /*
284 * The object header is a byte of 'type' followed by zero or
285 * more bytes of length.
286 */
287 hdrlen = encode_in_pack_object_header(type, size, header);
288
289 if (type == OBJ_OFS_DELTA) {
290 /*
291 * Deltas with relative base contain an additional
292 * encoding of the relative offset for the delta
293 * base from this object's position in the pack.
294 */
295 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
296 unsigned pos = sizeof(dheader) - 1;
297 dheader[pos] = ofs & 127;
298 while (ofs >>= 7)
299 dheader[--pos] = 128 | (--ofs & 127);
300 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
301 if (st)
302 close_istream(st);
303 free(buf);
304 return 0;
305 }
306 sha1write(f, header, hdrlen);
307 sha1write(f, dheader + pos, sizeof(dheader) - pos);
308 hdrlen += sizeof(dheader) - pos;
309 } else if (type == OBJ_REF_DELTA) {
310 /*
311 * Deltas with a base reference contain
312 * an additional 20 bytes for the base sha1.
313 */
314 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
315 if (st)
316 close_istream(st);
317 free(buf);
318 return 0;
319 }
320 sha1write(f, header, hdrlen);
321 sha1write(f, entry->delta->idx.sha1, 20);
322 hdrlen += 20;
323 } else {
324 if (limit && hdrlen + datalen + 20 >= limit) {
325 if (st)
326 close_istream(st);
327 free(buf);
328 return 0;
329 }
330 sha1write(f, header, hdrlen);
331 }
332 if (st) {
333 datalen = write_large_blob_data(st, f, entry->idx.sha1);
334 close_istream(st);
335 } else {
336 sha1write(f, buf, datalen);
337 free(buf);
338 }
339
340 return hdrlen + datalen;
341}
342
343/* Return 0 if we will bust the pack-size limit */
344static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
345 unsigned long limit, int usable_delta)
346{
347 struct packed_git *p = entry->in_pack;
348 struct pack_window *w_curs = NULL;
349 struct revindex_entry *revidx;
350 off_t offset;
351 enum object_type type = entry->type;
352 unsigned long datalen;
353 unsigned char header[10], dheader[10];
354 unsigned hdrlen;
355
356 if (entry->delta)
357 type = (allow_ofs_delta && entry->delta->idx.offset) ?
358 OBJ_OFS_DELTA : OBJ_REF_DELTA;
359 hdrlen = encode_in_pack_object_header(type, entry->size, header);
360
361 offset = entry->in_pack_offset;
362 revidx = find_pack_revindex(p, offset);
363 datalen = revidx[1].offset - offset;
364 if (!pack_to_stdout && p->index_version > 1 &&
365 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
366 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
367 unuse_pack(&w_curs);
368 return write_no_reuse_object(f, entry, limit, usable_delta);
369 }
370
371 offset += entry->in_pack_header_size;
372 datalen -= entry->in_pack_header_size;
373
374 if (!pack_to_stdout && p->index_version == 1 &&
375 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
376 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
377 unuse_pack(&w_curs);
378 return write_no_reuse_object(f, entry, limit, usable_delta);
379 }
380
381 if (type == OBJ_OFS_DELTA) {
382 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
383 unsigned pos = sizeof(dheader) - 1;
384 dheader[pos] = ofs & 127;
385 while (ofs >>= 7)
386 dheader[--pos] = 128 | (--ofs & 127);
387 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
388 unuse_pack(&w_curs);
389 return 0;
390 }
391 sha1write(f, header, hdrlen);
392 sha1write(f, dheader + pos, sizeof(dheader) - pos);
393 hdrlen += sizeof(dheader) - pos;
394 reused_delta++;
395 } else if (type == OBJ_REF_DELTA) {
396 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
397 unuse_pack(&w_curs);
398 return 0;
399 }
400 sha1write(f, header, hdrlen);
401 sha1write(f, entry->delta->idx.sha1, 20);
402 hdrlen += 20;
403 reused_delta++;
404 } else {
405 if (limit && hdrlen + datalen + 20 >= limit) {
406 unuse_pack(&w_curs);
407 return 0;
408 }
409 sha1write(f, header, hdrlen);
410 }
411 copy_pack_data(f, p, &w_curs, offset, datalen);
412 unuse_pack(&w_curs);
413 reused++;
414 return hdrlen + datalen;
415}
416
417/* Return 0 if we will bust the pack-size limit */
418static unsigned long write_object(struct sha1file *f,
419 struct object_entry *entry,
420 off_t write_offset)
421{
422 unsigned long limit, len;
423 int usable_delta, to_reuse;
424
425 if (!pack_to_stdout)
426 crc32_begin(f);
427
428 /* apply size limit if limited packsize and not first object */
429 if (!pack_size_limit || !nr_written)
430 limit = 0;
431 else if (pack_size_limit <= write_offset)
432 /*
433 * the earlier object did not fit the limit; avoid
434 * mistaking this with unlimited (i.e. limit = 0).
435 */
436 limit = 1;
437 else
438 limit = pack_size_limit - write_offset;
439
440 if (!entry->delta)
441 usable_delta = 0; /* no delta */
442 else if (!pack_size_limit)
443 usable_delta = 1; /* unlimited packfile */
444 else if (entry->delta->idx.offset == (off_t)-1)
445 usable_delta = 0; /* base was written to another pack */
446 else if (entry->delta->idx.offset)
447 usable_delta = 1; /* base already exists in this pack */
448 else
449 usable_delta = 0; /* base could end up in another pack */
450
451 if (!reuse_object)
452 to_reuse = 0; /* explicit */
453 else if (!entry->in_pack)
454 to_reuse = 0; /* can't reuse what we don't have */
455 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
456 /* check_object() decided it for us ... */
457 to_reuse = usable_delta;
458 /* ... but pack split may override that */
459 else if (entry->type != entry->in_pack_type)
460 to_reuse = 0; /* pack has delta which is unusable */
461 else if (entry->delta)
462 to_reuse = 0; /* we want to pack afresh */
463 else
464 to_reuse = 1; /* we have it in-pack undeltified,
465 * and we do not need to deltify it.
466 */
467
468 if (!to_reuse)
469 len = write_no_reuse_object(f, entry, limit, usable_delta);
470 else
471 len = write_reuse_object(f, entry, limit, usable_delta);
472 if (!len)
473 return 0;
474
475 if (usable_delta)
476 written_delta++;
477 written++;
478 if (!pack_to_stdout)
479 entry->idx.crc32 = crc32_end(f);
480 return len;
481}
482
483enum write_one_status {
484 WRITE_ONE_SKIP = -1, /* already written */
485 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
486 WRITE_ONE_WRITTEN = 1, /* normal */
487 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
488};
489
490static enum write_one_status write_one(struct sha1file *f,
491 struct object_entry *e,
492 off_t *offset)
493{
494 unsigned long size;
495 int recursing;
496
497 /*
498 * we set offset to 1 (which is an impossible value) to mark
499 * the fact that this object is involved in "write its base
500 * first before writing a deltified object" recursion.
501 */
502 recursing = (e->idx.offset == 1);
503 if (recursing) {
504 warning("recursive delta detected for object %s",
505 sha1_to_hex(e->idx.sha1));
506 return WRITE_ONE_RECURSIVE;
507 } else if (e->idx.offset || e->preferred_base) {
508 /* offset is non zero if object is written already. */
509 return WRITE_ONE_SKIP;
510 }
511
512 /* if we are deltified, write out base object first. */
513 if (e->delta) {
514 e->idx.offset = 1; /* now recurse */
515 switch (write_one(f, e->delta, offset)) {
516 case WRITE_ONE_RECURSIVE:
517 /* we cannot depend on this one */
518 e->delta = NULL;
519 break;
520 default:
521 break;
522 case WRITE_ONE_BREAK:
523 e->idx.offset = recursing;
524 return WRITE_ONE_BREAK;
525 }
526 }
527
528 e->idx.offset = *offset;
529 size = write_object(f, e, *offset);
530 if (!size) {
531 e->idx.offset = recursing;
532 return WRITE_ONE_BREAK;
533 }
534 written_list[nr_written++] = &e->idx;
535
536 /* make sure off_t is sufficiently large not to wrap */
537 if (signed_add_overflows(*offset, size))
538 die("pack too large for current definition of off_t");
539 *offset += size;
540 return WRITE_ONE_WRITTEN;
541}
542
543static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
544 void *cb_data)
545{
546 unsigned char peeled[20];
547 struct object_entry *entry = packlist_find(&to_pack, sha1, NULL);
548
549 if (entry)
550 entry->tagged = 1;
551 if (!peel_ref(path, peeled)) {
552 entry = packlist_find(&to_pack, peeled, NULL);
553 if (entry)
554 entry->tagged = 1;
555 }
556 return 0;
557}
558
559static inline void add_to_write_order(struct object_entry **wo,
560 unsigned int *endp,
561 struct object_entry *e)
562{
563 if (e->filled)
564 return;
565 wo[(*endp)++] = e;
566 e->filled = 1;
567}
568
569static void add_descendants_to_write_order(struct object_entry **wo,
570 unsigned int *endp,
571 struct object_entry *e)
572{
573 int add_to_order = 1;
574 while (e) {
575 if (add_to_order) {
576 struct object_entry *s;
577 /* add this node... */
578 add_to_write_order(wo, endp, e);
579 /* all its siblings... */
580 for (s = e->delta_sibling; s; s = s->delta_sibling) {
581 add_to_write_order(wo, endp, s);
582 }
583 }
584 /* drop down a level to add left subtree nodes if possible */
585 if (e->delta_child) {
586 add_to_order = 1;
587 e = e->delta_child;
588 } else {
589 add_to_order = 0;
590 /* our sibling might have some children, it is next */
591 if (e->delta_sibling) {
592 e = e->delta_sibling;
593 continue;
594 }
595 /* go back to our parent node */
596 e = e->delta;
597 while (e && !e->delta_sibling) {
598 /* we're on the right side of a subtree, keep
599 * going up until we can go right again */
600 e = e->delta;
601 }
602 if (!e) {
603 /* done- we hit our original root node */
604 return;
605 }
606 /* pass it off to sibling at this level */
607 e = e->delta_sibling;
608 }
609 };
610}
611
612static void add_family_to_write_order(struct object_entry **wo,
613 unsigned int *endp,
614 struct object_entry *e)
615{
616 struct object_entry *root;
617
618 for (root = e; root->delta; root = root->delta)
619 ; /* nothing */
620 add_descendants_to_write_order(wo, endp, root);
621}
622
623static struct object_entry **compute_write_order(void)
624{
625 unsigned int i, wo_end, last_untagged;
626
627 struct object_entry **wo = xmalloc(to_pack.nr_objects * sizeof(*wo));
628 struct object_entry *objects = to_pack.objects;
629
630 for (i = 0; i < to_pack.nr_objects; i++) {
631 objects[i].tagged = 0;
632 objects[i].filled = 0;
633 objects[i].delta_child = NULL;
634 objects[i].delta_sibling = NULL;
635 }
636
637 /*
638 * Fully connect delta_child/delta_sibling network.
639 * Make sure delta_sibling is sorted in the original
640 * recency order.
641 */
642 for (i = to_pack.nr_objects; i > 0;) {
643 struct object_entry *e = &objects[--i];
644 if (!e->delta)
645 continue;
646 /* Mark me as the first child */
647 e->delta_sibling = e->delta->delta_child;
648 e->delta->delta_child = e;
649 }
650
651 /*
652 * Mark objects that are at the tip of tags.
653 */
654 for_each_tag_ref(mark_tagged, NULL);
655
656 /*
657 * Give the objects in the original recency order until
658 * we see a tagged tip.
659 */
660 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
661 if (objects[i].tagged)
662 break;
663 add_to_write_order(wo, &wo_end, &objects[i]);
664 }
665 last_untagged = i;
666
667 /*
668 * Then fill all the tagged tips.
669 */
670 for (; i < to_pack.nr_objects; i++) {
671 if (objects[i].tagged)
672 add_to_write_order(wo, &wo_end, &objects[i]);
673 }
674
675 /*
676 * And then all remaining commits and tags.
677 */
678 for (i = last_untagged; i < to_pack.nr_objects; i++) {
679 if (objects[i].type != OBJ_COMMIT &&
680 objects[i].type != OBJ_TAG)
681 continue;
682 add_to_write_order(wo, &wo_end, &objects[i]);
683 }
684
685 /*
686 * And then all the trees.
687 */
688 for (i = last_untagged; i < to_pack.nr_objects; i++) {
689 if (objects[i].type != OBJ_TREE)
690 continue;
691 add_to_write_order(wo, &wo_end, &objects[i]);
692 }
693
694 /*
695 * Finally all the rest in really tight order
696 */
697 for (i = last_untagged; i < to_pack.nr_objects; i++) {
698 if (!objects[i].filled)
699 add_family_to_write_order(wo, &wo_end, &objects[i]);
700 }
701
702 if (wo_end != to_pack.nr_objects)
703 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
704
705 return wo;
706}
707
708static off_t write_reused_pack(struct sha1file *f)
709{
710 unsigned char buffer[8192];
711 off_t to_write, total;
712 int fd;
713
714 if (!is_pack_valid(reuse_packfile))
715 die("packfile is invalid: %s", reuse_packfile->pack_name);
716
717 fd = git_open_noatime(reuse_packfile->pack_name);
718 if (fd < 0)
719 die_errno("unable to open packfile for reuse: %s",
720 reuse_packfile->pack_name);
721
722 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
723 die_errno("unable to seek in reused packfile");
724
725 if (reuse_packfile_offset < 0)
726 reuse_packfile_offset = reuse_packfile->pack_size - 20;
727
728 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
729
730 while (to_write) {
731 int read_pack = xread(fd, buffer, sizeof(buffer));
732
733 if (read_pack <= 0)
734 die_errno("unable to read from reused packfile");
735
736 if (read_pack > to_write)
737 read_pack = to_write;
738
739 sha1write(f, buffer, read_pack);
740 to_write -= read_pack;
741
742 /*
743 * We don't know the actual number of objects written,
744 * only how many bytes written, how many bytes total, and
745 * how many objects total. So we can fake it by pretending all
746 * objects we are writing are the same size. This gives us a
747 * smooth progress meter, and at the end it matches the true
748 * answer.
749 */
750 written = reuse_packfile_objects *
751 (((double)(total - to_write)) / total);
752 display_progress(progress_state, written);
753 }
754
755 close(fd);
756 written = reuse_packfile_objects;
757 display_progress(progress_state, written);
758 return reuse_packfile_offset - sizeof(struct pack_header);
759}
760
761static void write_pack_file(void)
762{
763 uint32_t i = 0, j;
764 struct sha1file *f;
765 off_t offset;
766 uint32_t nr_remaining = nr_result;
767 time_t last_mtime = 0;
768 struct object_entry **write_order;
769
770 if (progress > pack_to_stdout)
771 progress_state = start_progress(_("Writing objects"), nr_result);
772 written_list = xmalloc(to_pack.nr_objects * sizeof(*written_list));
773 write_order = compute_write_order();
774
775 do {
776 unsigned char sha1[20];
777 char *pack_tmp_name = NULL;
778
779 if (pack_to_stdout)
780 f = sha1fd_throughput(1, "<stdout>", progress_state);
781 else
782 f = create_tmp_packfile(&pack_tmp_name);
783
784 offset = write_pack_header(f, nr_remaining);
785
786 if (reuse_packfile) {
787 off_t packfile_size;
788 assert(pack_to_stdout);
789
790 packfile_size = write_reused_pack(f);
791 offset += packfile_size;
792 }
793
794 nr_written = 0;
795 for (; i < to_pack.nr_objects; i++) {
796 struct object_entry *e = write_order[i];
797 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
798 break;
799 display_progress(progress_state, written);
800 }
801
802 /*
803 * Did we write the wrong # entries in the header?
804 * If so, rewrite it like in fast-import
805 */
806 if (pack_to_stdout) {
807 sha1close(f, sha1, CSUM_CLOSE);
808 } else if (nr_written == nr_remaining) {
809 sha1close(f, sha1, CSUM_FSYNC);
810 } else {
811 int fd = sha1close(f, sha1, 0);
812 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
813 nr_written, sha1, offset);
814 close(fd);
815 }
816
817 if (!pack_to_stdout) {
818 struct stat st;
819 char tmpname[PATH_MAX];
820
821 /*
822 * Packs are runtime accessed in their mtime
823 * order since newer packs are more likely to contain
824 * younger objects. So if we are creating multiple
825 * packs then we should modify the mtime of later ones
826 * to preserve this property.
827 */
828 if (stat(pack_tmp_name, &st) < 0) {
829 warning("failed to stat %s: %s",
830 pack_tmp_name, strerror(errno));
831 } else if (!last_mtime) {
832 last_mtime = st.st_mtime;
833 } else {
834 struct utimbuf utb;
835 utb.actime = st.st_atime;
836 utb.modtime = --last_mtime;
837 if (utime(pack_tmp_name, &utb) < 0)
838 warning("failed utime() on %s: %s",
839 pack_tmp_name, strerror(errno));
840 }
841
842 /* Enough space for "-<sha-1>.pack"? */
843 if (sizeof(tmpname) <= strlen(base_name) + 50)
844 die("pack base name '%s' too long", base_name);
845 snprintf(tmpname, sizeof(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 char *end_of_name_prefix = strrchr(tmpname, 0);
858 sprintf(end_of_name_prefix, "%s.bitmap", sha1_to_hex(sha1));
859
860 stop_progress(&progress_state);
861
862 bitmap_writer_show_progress(progress);
863 bitmap_writer_reuse_bitmaps(&to_pack);
864 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
865 bitmap_writer_build(&to_pack);
866 bitmap_writer_finish(written_list, nr_written,
867 tmpname, write_bitmap_options);
868 write_bitmap_index = 0;
869 }
870
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 int add_object_entry(const unsigned char *sha1, enum object_type type,
1015 const char *name, int exclude)
1016{
1017 struct packed_git *found_pack;
1018 off_t found_offset;
1019 uint32_t index_pos;
1020
1021 if (have_duplicate_entry(sha1, exclude, &index_pos))
1022 return 0;
1023
1024 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset))
1025 return 0;
1026
1027 create_object_entry(sha1, type, pack_name_hash(name),
1028 exclude, name && no_try_delta(name),
1029 index_pos, found_pack, found_offset);
1030
1031 display_progress(progress_state, to_pack.nr_objects);
1032 return 1;
1033}
1034
1035static int add_object_entry_from_bitmap(const unsigned char *sha1,
1036 enum object_type type,
1037 int flags, uint32_t name_hash,
1038 struct packed_git *pack, off_t offset)
1039{
1040 uint32_t index_pos;
1041
1042 if (have_duplicate_entry(sha1, 0, &index_pos))
1043 return 0;
1044
1045 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1046
1047 display_progress(progress_state, to_pack.nr_objects);
1048 return 1;
1049}
1050
1051struct pbase_tree_cache {
1052 unsigned char sha1[20];
1053 int ref;
1054 int temporary;
1055 void *tree_data;
1056 unsigned long tree_size;
1057};
1058
1059static struct pbase_tree_cache *(pbase_tree_cache[256]);
1060static int pbase_tree_cache_ix(const unsigned char *sha1)
1061{
1062 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1063}
1064static int pbase_tree_cache_ix_incr(int ix)
1065{
1066 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1067}
1068
1069static struct pbase_tree {
1070 struct pbase_tree *next;
1071 /* This is a phony "cache" entry; we are not
1072 * going to evict it nor find it through _get()
1073 * mechanism -- this is for the toplevel node that
1074 * would almost always change with any commit.
1075 */
1076 struct pbase_tree_cache pcache;
1077} *pbase_tree;
1078
1079static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1080{
1081 struct pbase_tree_cache *ent, *nent;
1082 void *data;
1083 unsigned long size;
1084 enum object_type type;
1085 int neigh;
1086 int my_ix = pbase_tree_cache_ix(sha1);
1087 int available_ix = -1;
1088
1089 /* pbase-tree-cache acts as a limited hashtable.
1090 * your object will be found at your index or within a few
1091 * slots after that slot if it is cached.
1092 */
1093 for (neigh = 0; neigh < 8; neigh++) {
1094 ent = pbase_tree_cache[my_ix];
1095 if (ent && !hashcmp(ent->sha1, sha1)) {
1096 ent->ref++;
1097 return ent;
1098 }
1099 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1100 ((0 <= available_ix) &&
1101 (!ent && pbase_tree_cache[available_ix])))
1102 available_ix = my_ix;
1103 if (!ent)
1104 break;
1105 my_ix = pbase_tree_cache_ix_incr(my_ix);
1106 }
1107
1108 /* Did not find one. Either we got a bogus request or
1109 * we need to read and perhaps cache.
1110 */
1111 data = read_sha1_file(sha1, &type, &size);
1112 if (!data)
1113 return NULL;
1114 if (type != OBJ_TREE) {
1115 free(data);
1116 return NULL;
1117 }
1118
1119 /* We need to either cache or return a throwaway copy */
1120
1121 if (available_ix < 0)
1122 ent = NULL;
1123 else {
1124 ent = pbase_tree_cache[available_ix];
1125 my_ix = available_ix;
1126 }
1127
1128 if (!ent) {
1129 nent = xmalloc(sizeof(*nent));
1130 nent->temporary = (available_ix < 0);
1131 }
1132 else {
1133 /* evict and reuse */
1134 free(ent->tree_data);
1135 nent = ent;
1136 }
1137 hashcpy(nent->sha1, sha1);
1138 nent->tree_data = data;
1139 nent->tree_size = size;
1140 nent->ref = 1;
1141 if (!nent->temporary)
1142 pbase_tree_cache[my_ix] = nent;
1143 return nent;
1144}
1145
1146static void pbase_tree_put(struct pbase_tree_cache *cache)
1147{
1148 if (!cache->temporary) {
1149 cache->ref--;
1150 return;
1151 }
1152 free(cache->tree_data);
1153 free(cache);
1154}
1155
1156static int name_cmp_len(const char *name)
1157{
1158 int i;
1159 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1160 ;
1161 return i;
1162}
1163
1164static void add_pbase_object(struct tree_desc *tree,
1165 const char *name,
1166 int cmplen,
1167 const char *fullname)
1168{
1169 struct name_entry entry;
1170 int cmp;
1171
1172 while (tree_entry(tree,&entry)) {
1173 if (S_ISGITLINK(entry.mode))
1174 continue;
1175 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1176 memcmp(name, entry.path, cmplen);
1177 if (cmp > 0)
1178 continue;
1179 if (cmp < 0)
1180 return;
1181 if (name[cmplen] != '/') {
1182 add_object_entry(entry.sha1,
1183 object_type(entry.mode),
1184 fullname, 1);
1185 return;
1186 }
1187 if (S_ISDIR(entry.mode)) {
1188 struct tree_desc sub;
1189 struct pbase_tree_cache *tree;
1190 const char *down = name+cmplen+1;
1191 int downlen = name_cmp_len(down);
1192
1193 tree = pbase_tree_get(entry.sha1);
1194 if (!tree)
1195 return;
1196 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1197
1198 add_pbase_object(&sub, down, downlen, fullname);
1199 pbase_tree_put(tree);
1200 }
1201 }
1202}
1203
1204static unsigned *done_pbase_paths;
1205static int done_pbase_paths_num;
1206static int done_pbase_paths_alloc;
1207static int done_pbase_path_pos(unsigned hash)
1208{
1209 int lo = 0;
1210 int hi = done_pbase_paths_num;
1211 while (lo < hi) {
1212 int mi = (hi + lo) / 2;
1213 if (done_pbase_paths[mi] == hash)
1214 return mi;
1215 if (done_pbase_paths[mi] < hash)
1216 hi = mi;
1217 else
1218 lo = mi + 1;
1219 }
1220 return -lo-1;
1221}
1222
1223static int check_pbase_path(unsigned hash)
1224{
1225 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1226 if (0 <= pos)
1227 return 1;
1228 pos = -pos - 1;
1229 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1230 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1231 done_pbase_paths = xrealloc(done_pbase_paths,
1232 done_pbase_paths_alloc *
1233 sizeof(unsigned));
1234 }
1235 done_pbase_paths_num++;
1236 if (pos < done_pbase_paths_num)
1237 memmove(done_pbase_paths + pos + 1,
1238 done_pbase_paths + pos,
1239 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1240 done_pbase_paths[pos] = hash;
1241 return 0;
1242}
1243
1244static void add_preferred_base_object(const char *name)
1245{
1246 struct pbase_tree *it;
1247 int cmplen;
1248 unsigned hash = pack_name_hash(name);
1249
1250 if (!num_preferred_base || check_pbase_path(hash))
1251 return;
1252
1253 cmplen = name_cmp_len(name);
1254 for (it = pbase_tree; it; it = it->next) {
1255 if (cmplen == 0) {
1256 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1257 }
1258 else {
1259 struct tree_desc tree;
1260 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1261 add_pbase_object(&tree, name, cmplen, name);
1262 }
1263 }
1264}
1265
1266static void add_preferred_base(unsigned char *sha1)
1267{
1268 struct pbase_tree *it;
1269 void *data;
1270 unsigned long size;
1271 unsigned char tree_sha1[20];
1272
1273 if (window <= num_preferred_base++)
1274 return;
1275
1276 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1277 if (!data)
1278 return;
1279
1280 for (it = pbase_tree; it; it = it->next) {
1281 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1282 free(data);
1283 return;
1284 }
1285 }
1286
1287 it = xcalloc(1, sizeof(*it));
1288 it->next = pbase_tree;
1289 pbase_tree = it;
1290
1291 hashcpy(it->pcache.sha1, tree_sha1);
1292 it->pcache.tree_data = data;
1293 it->pcache.tree_size = size;
1294}
1295
1296static void cleanup_preferred_base(void)
1297{
1298 struct pbase_tree *it;
1299 unsigned i;
1300
1301 it = pbase_tree;
1302 pbase_tree = NULL;
1303 while (it) {
1304 struct pbase_tree *this = it;
1305 it = this->next;
1306 free(this->pcache.tree_data);
1307 free(this);
1308 }
1309
1310 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1311 if (!pbase_tree_cache[i])
1312 continue;
1313 free(pbase_tree_cache[i]->tree_data);
1314 free(pbase_tree_cache[i]);
1315 pbase_tree_cache[i] = NULL;
1316 }
1317
1318 free(done_pbase_paths);
1319 done_pbase_paths = NULL;
1320 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1321}
1322
1323static void check_object(struct object_entry *entry)
1324{
1325 if (entry->in_pack) {
1326 struct packed_git *p = entry->in_pack;
1327 struct pack_window *w_curs = NULL;
1328 const unsigned char *base_ref = NULL;
1329 struct object_entry *base_entry;
1330 unsigned long used, used_0;
1331 unsigned long avail;
1332 off_t ofs;
1333 unsigned char *buf, c;
1334
1335 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1336
1337 /*
1338 * We want in_pack_type even if we do not reuse delta
1339 * since non-delta representations could still be reused.
1340 */
1341 used = unpack_object_header_buffer(buf, avail,
1342 &entry->in_pack_type,
1343 &entry->size);
1344 if (used == 0)
1345 goto give_up;
1346
1347 /*
1348 * Determine if this is a delta and if so whether we can
1349 * reuse it or not. Otherwise let's find out as cheaply as
1350 * possible what the actual type and size for this object is.
1351 */
1352 switch (entry->in_pack_type) {
1353 default:
1354 /* Not a delta hence we've already got all we need. */
1355 entry->type = entry->in_pack_type;
1356 entry->in_pack_header_size = used;
1357 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1358 goto give_up;
1359 unuse_pack(&w_curs);
1360 return;
1361 case OBJ_REF_DELTA:
1362 if (reuse_delta && !entry->preferred_base)
1363 base_ref = use_pack(p, &w_curs,
1364 entry->in_pack_offset + used, NULL);
1365 entry->in_pack_header_size = used + 20;
1366 break;
1367 case OBJ_OFS_DELTA:
1368 buf = use_pack(p, &w_curs,
1369 entry->in_pack_offset + used, NULL);
1370 used_0 = 0;
1371 c = buf[used_0++];
1372 ofs = c & 127;
1373 while (c & 128) {
1374 ofs += 1;
1375 if (!ofs || MSB(ofs, 7)) {
1376 error("delta base offset overflow in pack for %s",
1377 sha1_to_hex(entry->idx.sha1));
1378 goto give_up;
1379 }
1380 c = buf[used_0++];
1381 ofs = (ofs << 7) + (c & 127);
1382 }
1383 ofs = entry->in_pack_offset - ofs;
1384 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1385 error("delta base offset out of bound for %s",
1386 sha1_to_hex(entry->idx.sha1));
1387 goto give_up;
1388 }
1389 if (reuse_delta && !entry->preferred_base) {
1390 struct revindex_entry *revidx;
1391 revidx = find_pack_revindex(p, ofs);
1392 if (!revidx)
1393 goto give_up;
1394 base_ref = nth_packed_object_sha1(p, revidx->nr);
1395 }
1396 entry->in_pack_header_size = used + used_0;
1397 break;
1398 }
1399
1400 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1401 /*
1402 * If base_ref was set above that means we wish to
1403 * reuse delta data, and we even found that base
1404 * in the list of objects we want to pack. Goodie!
1405 *
1406 * Depth value does not matter - find_deltas() will
1407 * never consider reused delta as the base object to
1408 * deltify other objects against, in order to avoid
1409 * circular deltas.
1410 */
1411 entry->type = entry->in_pack_type;
1412 entry->delta = base_entry;
1413 entry->delta_size = entry->size;
1414 entry->delta_sibling = base_entry->delta_child;
1415 base_entry->delta_child = entry;
1416 unuse_pack(&w_curs);
1417 return;
1418 }
1419
1420 if (entry->type) {
1421 /*
1422 * This must be a delta and we already know what the
1423 * final object type is. Let's extract the actual
1424 * object size from the delta header.
1425 */
1426 entry->size = get_size_from_delta(p, &w_curs,
1427 entry->in_pack_offset + entry->in_pack_header_size);
1428 if (entry->size == 0)
1429 goto give_up;
1430 unuse_pack(&w_curs);
1431 return;
1432 }
1433
1434 /*
1435 * No choice but to fall back to the recursive delta walk
1436 * with sha1_object_info() to find about the object type
1437 * at this point...
1438 */
1439 give_up:
1440 unuse_pack(&w_curs);
1441 }
1442
1443 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1444 /*
1445 * The error condition is checked in prepare_pack(). This is
1446 * to permit a missing preferred base object to be ignored
1447 * as a preferred base. Doing so can result in a larger
1448 * pack file, but the transfer will still take place.
1449 */
1450}
1451
1452static int pack_offset_sort(const void *_a, const void *_b)
1453{
1454 const struct object_entry *a = *(struct object_entry **)_a;
1455 const struct object_entry *b = *(struct object_entry **)_b;
1456
1457 /* avoid filesystem trashing with loose objects */
1458 if (!a->in_pack && !b->in_pack)
1459 return hashcmp(a->idx.sha1, b->idx.sha1);
1460
1461 if (a->in_pack < b->in_pack)
1462 return -1;
1463 if (a->in_pack > b->in_pack)
1464 return 1;
1465 return a->in_pack_offset < b->in_pack_offset ? -1 :
1466 (a->in_pack_offset > b->in_pack_offset);
1467}
1468
1469static void get_object_details(void)
1470{
1471 uint32_t i;
1472 struct object_entry **sorted_by_offset;
1473
1474 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1475 for (i = 0; i < to_pack.nr_objects; i++)
1476 sorted_by_offset[i] = to_pack.objects + i;
1477 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1478
1479 for (i = 0; i < to_pack.nr_objects; i++) {
1480 struct object_entry *entry = sorted_by_offset[i];
1481 check_object(entry);
1482 if (big_file_threshold < entry->size)
1483 entry->no_try_delta = 1;
1484 }
1485
1486 free(sorted_by_offset);
1487}
1488
1489/*
1490 * We search for deltas in a list sorted by type, by filename hash, and then
1491 * by size, so that we see progressively smaller and smaller files.
1492 * That's because we prefer deltas to be from the bigger file
1493 * to the smaller -- deletes are potentially cheaper, but perhaps
1494 * more importantly, the bigger file is likely the more recent
1495 * one. The deepest deltas are therefore the oldest objects which are
1496 * less susceptible to be accessed often.
1497 */
1498static int type_size_sort(const void *_a, const void *_b)
1499{
1500 const struct object_entry *a = *(struct object_entry **)_a;
1501 const struct object_entry *b = *(struct object_entry **)_b;
1502
1503 if (a->type > b->type)
1504 return -1;
1505 if (a->type < b->type)
1506 return 1;
1507 if (a->hash > b->hash)
1508 return -1;
1509 if (a->hash < b->hash)
1510 return 1;
1511 if (a->preferred_base > b->preferred_base)
1512 return -1;
1513 if (a->preferred_base < b->preferred_base)
1514 return 1;
1515 if (a->size > b->size)
1516 return -1;
1517 if (a->size < b->size)
1518 return 1;
1519 return a < b ? -1 : (a > b); /* newest first */
1520}
1521
1522struct unpacked {
1523 struct object_entry *entry;
1524 void *data;
1525 struct delta_index *index;
1526 unsigned depth;
1527};
1528
1529static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1530 unsigned long delta_size)
1531{
1532 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1533 return 0;
1534
1535 if (delta_size < cache_max_small_delta_size)
1536 return 1;
1537
1538 /* cache delta, if objects are large enough compared to delta size */
1539 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1540 return 1;
1541
1542 return 0;
1543}
1544
1545#ifndef NO_PTHREADS
1546
1547static pthread_mutex_t read_mutex;
1548#define read_lock() pthread_mutex_lock(&read_mutex)
1549#define read_unlock() pthread_mutex_unlock(&read_mutex)
1550
1551static pthread_mutex_t cache_mutex;
1552#define cache_lock() pthread_mutex_lock(&cache_mutex)
1553#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1554
1555static pthread_mutex_t progress_mutex;
1556#define progress_lock() pthread_mutex_lock(&progress_mutex)
1557#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1558
1559#else
1560
1561#define read_lock() (void)0
1562#define read_unlock() (void)0
1563#define cache_lock() (void)0
1564#define cache_unlock() (void)0
1565#define progress_lock() (void)0
1566#define progress_unlock() (void)0
1567
1568#endif
1569
1570static int try_delta(struct unpacked *trg, struct unpacked *src,
1571 unsigned max_depth, unsigned long *mem_usage)
1572{
1573 struct object_entry *trg_entry = trg->entry;
1574 struct object_entry *src_entry = src->entry;
1575 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1576 unsigned ref_depth;
1577 enum object_type type;
1578 void *delta_buf;
1579
1580 /* Don't bother doing diffs between different types */
1581 if (trg_entry->type != src_entry->type)
1582 return -1;
1583
1584 /*
1585 * We do not bother to try a delta that we discarded on an
1586 * earlier try, but only when reusing delta data. Note that
1587 * src_entry that is marked as the preferred_base should always
1588 * be considered, as even if we produce a suboptimal delta against
1589 * it, we will still save the transfer cost, as we already know
1590 * the other side has it and we won't send src_entry at all.
1591 */
1592 if (reuse_delta && trg_entry->in_pack &&
1593 trg_entry->in_pack == src_entry->in_pack &&
1594 !src_entry->preferred_base &&
1595 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1596 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1597 return 0;
1598
1599 /* Let's not bust the allowed depth. */
1600 if (src->depth >= max_depth)
1601 return 0;
1602
1603 /* Now some size filtering heuristics. */
1604 trg_size = trg_entry->size;
1605 if (!trg_entry->delta) {
1606 max_size = trg_size/2 - 20;
1607 ref_depth = 1;
1608 } else {
1609 max_size = trg_entry->delta_size;
1610 ref_depth = trg->depth;
1611 }
1612 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1613 (max_depth - ref_depth + 1);
1614 if (max_size == 0)
1615 return 0;
1616 src_size = src_entry->size;
1617 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1618 if (sizediff >= max_size)
1619 return 0;
1620 if (trg_size < src_size / 32)
1621 return 0;
1622
1623 /* Load data if not already done */
1624 if (!trg->data) {
1625 read_lock();
1626 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1627 read_unlock();
1628 if (!trg->data)
1629 die("object %s cannot be read",
1630 sha1_to_hex(trg_entry->idx.sha1));
1631 if (sz != trg_size)
1632 die("object %s inconsistent object length (%lu vs %lu)",
1633 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1634 *mem_usage += sz;
1635 }
1636 if (!src->data) {
1637 read_lock();
1638 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1639 read_unlock();
1640 if (!src->data) {
1641 if (src_entry->preferred_base) {
1642 static int warned = 0;
1643 if (!warned++)
1644 warning("object %s cannot be read",
1645 sha1_to_hex(src_entry->idx.sha1));
1646 /*
1647 * Those objects are not included in the
1648 * resulting pack. Be resilient and ignore
1649 * them if they can't be read, in case the
1650 * pack could be created nevertheless.
1651 */
1652 return 0;
1653 }
1654 die("object %s cannot be read",
1655 sha1_to_hex(src_entry->idx.sha1));
1656 }
1657 if (sz != src_size)
1658 die("object %s inconsistent object length (%lu vs %lu)",
1659 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1660 *mem_usage += sz;
1661 }
1662 if (!src->index) {
1663 src->index = create_delta_index(src->data, src_size);
1664 if (!src->index) {
1665 static int warned = 0;
1666 if (!warned++)
1667 warning("suboptimal pack - out of memory");
1668 return 0;
1669 }
1670 *mem_usage += sizeof_delta_index(src->index);
1671 }
1672
1673 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1674 if (!delta_buf)
1675 return 0;
1676
1677 if (trg_entry->delta) {
1678 /* Prefer only shallower same-sized deltas. */
1679 if (delta_size == trg_entry->delta_size &&
1680 src->depth + 1 >= trg->depth) {
1681 free(delta_buf);
1682 return 0;
1683 }
1684 }
1685
1686 /*
1687 * Handle memory allocation outside of the cache
1688 * accounting lock. Compiler will optimize the strangeness
1689 * away when NO_PTHREADS is defined.
1690 */
1691 free(trg_entry->delta_data);
1692 cache_lock();
1693 if (trg_entry->delta_data) {
1694 delta_cache_size -= trg_entry->delta_size;
1695 trg_entry->delta_data = NULL;
1696 }
1697 if (delta_cacheable(src_size, trg_size, delta_size)) {
1698 delta_cache_size += delta_size;
1699 cache_unlock();
1700 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1701 } else {
1702 cache_unlock();
1703 free(delta_buf);
1704 }
1705
1706 trg_entry->delta = src_entry;
1707 trg_entry->delta_size = delta_size;
1708 trg->depth = src->depth + 1;
1709
1710 return 1;
1711}
1712
1713static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1714{
1715 struct object_entry *child = me->delta_child;
1716 unsigned int m = n;
1717 while (child) {
1718 unsigned int c = check_delta_limit(child, n + 1);
1719 if (m < c)
1720 m = c;
1721 child = child->delta_sibling;
1722 }
1723 return m;
1724}
1725
1726static unsigned long free_unpacked(struct unpacked *n)
1727{
1728 unsigned long freed_mem = sizeof_delta_index(n->index);
1729 free_delta_index(n->index);
1730 n->index = NULL;
1731 if (n->data) {
1732 freed_mem += n->entry->size;
1733 free(n->data);
1734 n->data = NULL;
1735 }
1736 n->entry = NULL;
1737 n->depth = 0;
1738 return freed_mem;
1739}
1740
1741static void find_deltas(struct object_entry **list, unsigned *list_size,
1742 int window, int depth, unsigned *processed)
1743{
1744 uint32_t i, idx = 0, count = 0;
1745 struct unpacked *array;
1746 unsigned long mem_usage = 0;
1747
1748 array = xcalloc(window, sizeof(struct unpacked));
1749
1750 for (;;) {
1751 struct object_entry *entry;
1752 struct unpacked *n = array + idx;
1753 int j, max_depth, best_base = -1;
1754
1755 progress_lock();
1756 if (!*list_size) {
1757 progress_unlock();
1758 break;
1759 }
1760 entry = *list++;
1761 (*list_size)--;
1762 if (!entry->preferred_base) {
1763 (*processed)++;
1764 display_progress(progress_state, *processed);
1765 }
1766 progress_unlock();
1767
1768 mem_usage -= free_unpacked(n);
1769 n->entry = entry;
1770
1771 while (window_memory_limit &&
1772 mem_usage > window_memory_limit &&
1773 count > 1) {
1774 uint32_t tail = (idx + window - count) % window;
1775 mem_usage -= free_unpacked(array + tail);
1776 count--;
1777 }
1778
1779 /* We do not compute delta to *create* objects we are not
1780 * going to pack.
1781 */
1782 if (entry->preferred_base)
1783 goto next;
1784
1785 /*
1786 * If the current object is at pack edge, take the depth the
1787 * objects that depend on the current object into account
1788 * otherwise they would become too deep.
1789 */
1790 max_depth = depth;
1791 if (entry->delta_child) {
1792 max_depth -= check_delta_limit(entry, 0);
1793 if (max_depth <= 0)
1794 goto next;
1795 }
1796
1797 j = window;
1798 while (--j > 0) {
1799 int ret;
1800 uint32_t other_idx = idx + j;
1801 struct unpacked *m;
1802 if (other_idx >= window)
1803 other_idx -= window;
1804 m = array + other_idx;
1805 if (!m->entry)
1806 break;
1807 ret = try_delta(n, m, max_depth, &mem_usage);
1808 if (ret < 0)
1809 break;
1810 else if (ret > 0)
1811 best_base = other_idx;
1812 }
1813
1814 /*
1815 * If we decided to cache the delta data, then it is best
1816 * to compress it right away. First because we have to do
1817 * it anyway, and doing it here while we're threaded will
1818 * save a lot of time in the non threaded write phase,
1819 * as well as allow for caching more deltas within
1820 * the same cache size limit.
1821 * ...
1822 * But only if not writing to stdout, since in that case
1823 * the network is most likely throttling writes anyway,
1824 * and therefore it is best to go to the write phase ASAP
1825 * instead, as we can afford spending more time compressing
1826 * between writes at that moment.
1827 */
1828 if (entry->delta_data && !pack_to_stdout) {
1829 entry->z_delta_size = do_compress(&entry->delta_data,
1830 entry->delta_size);
1831 cache_lock();
1832 delta_cache_size -= entry->delta_size;
1833 delta_cache_size += entry->z_delta_size;
1834 cache_unlock();
1835 }
1836
1837 /* if we made n a delta, and if n is already at max
1838 * depth, leaving it in the window is pointless. we
1839 * should evict it first.
1840 */
1841 if (entry->delta && max_depth <= n->depth)
1842 continue;
1843
1844 /*
1845 * Move the best delta base up in the window, after the
1846 * currently deltified object, to keep it longer. It will
1847 * be the first base object to be attempted next.
1848 */
1849 if (entry->delta) {
1850 struct unpacked swap = array[best_base];
1851 int dist = (window + idx - best_base) % window;
1852 int dst = best_base;
1853 while (dist--) {
1854 int src = (dst + 1) % window;
1855 array[dst] = array[src];
1856 dst = src;
1857 }
1858 array[dst] = swap;
1859 }
1860
1861 next:
1862 idx++;
1863 if (count + 1 < window)
1864 count++;
1865 if (idx >= window)
1866 idx = 0;
1867 }
1868
1869 for (i = 0; i < window; ++i) {
1870 free_delta_index(array[i].index);
1871 free(array[i].data);
1872 }
1873 free(array);
1874}
1875
1876#ifndef NO_PTHREADS
1877
1878static void try_to_free_from_threads(size_t size)
1879{
1880 read_lock();
1881 release_pack_memory(size);
1882 read_unlock();
1883}
1884
1885static try_to_free_t old_try_to_free_routine;
1886
1887/*
1888 * The main thread waits on the condition that (at least) one of the workers
1889 * has stopped working (which is indicated in the .working member of
1890 * struct thread_params).
1891 * When a work thread has completed its work, it sets .working to 0 and
1892 * signals the main thread and waits on the condition that .data_ready
1893 * becomes 1.
1894 */
1895
1896struct thread_params {
1897 pthread_t thread;
1898 struct object_entry **list;
1899 unsigned list_size;
1900 unsigned remaining;
1901 int window;
1902 int depth;
1903 int working;
1904 int data_ready;
1905 pthread_mutex_t mutex;
1906 pthread_cond_t cond;
1907 unsigned *processed;
1908};
1909
1910static pthread_cond_t progress_cond;
1911
1912/*
1913 * Mutex and conditional variable can't be statically-initialized on Windows.
1914 */
1915static void init_threaded_search(void)
1916{
1917 init_recursive_mutex(&read_mutex);
1918 pthread_mutex_init(&cache_mutex, NULL);
1919 pthread_mutex_init(&progress_mutex, NULL);
1920 pthread_cond_init(&progress_cond, NULL);
1921 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1922}
1923
1924static void cleanup_threaded_search(void)
1925{
1926 set_try_to_free_routine(old_try_to_free_routine);
1927 pthread_cond_destroy(&progress_cond);
1928 pthread_mutex_destroy(&read_mutex);
1929 pthread_mutex_destroy(&cache_mutex);
1930 pthread_mutex_destroy(&progress_mutex);
1931}
1932
1933static void *threaded_find_deltas(void *arg)
1934{
1935 struct thread_params *me = arg;
1936
1937 while (me->remaining) {
1938 find_deltas(me->list, &me->remaining,
1939 me->window, me->depth, me->processed);
1940
1941 progress_lock();
1942 me->working = 0;
1943 pthread_cond_signal(&progress_cond);
1944 progress_unlock();
1945
1946 /*
1947 * We must not set ->data_ready before we wait on the
1948 * condition because the main thread may have set it to 1
1949 * before we get here. In order to be sure that new
1950 * work is available if we see 1 in ->data_ready, it
1951 * was initialized to 0 before this thread was spawned
1952 * and we reset it to 0 right away.
1953 */
1954 pthread_mutex_lock(&me->mutex);
1955 while (!me->data_ready)
1956 pthread_cond_wait(&me->cond, &me->mutex);
1957 me->data_ready = 0;
1958 pthread_mutex_unlock(&me->mutex);
1959 }
1960 /* leave ->working 1 so that this doesn't get more work assigned */
1961 return NULL;
1962}
1963
1964static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1965 int window, int depth, unsigned *processed)
1966{
1967 struct thread_params *p;
1968 int i, ret, active_threads = 0;
1969
1970 init_threaded_search();
1971
1972 if (!delta_search_threads) /* --threads=0 means autodetect */
1973 delta_search_threads = online_cpus();
1974 if (delta_search_threads <= 1) {
1975 find_deltas(list, &list_size, window, depth, processed);
1976 cleanup_threaded_search();
1977 return;
1978 }
1979 if (progress > pack_to_stdout)
1980 fprintf(stderr, "Delta compression using up to %d threads.\n",
1981 delta_search_threads);
1982 p = xcalloc(delta_search_threads, sizeof(*p));
1983
1984 /* Partition the work amongst work threads. */
1985 for (i = 0; i < delta_search_threads; i++) {
1986 unsigned sub_size = list_size / (delta_search_threads - i);
1987
1988 /* don't use too small segments or no deltas will be found */
1989 if (sub_size < 2*window && i+1 < delta_search_threads)
1990 sub_size = 0;
1991
1992 p[i].window = window;
1993 p[i].depth = depth;
1994 p[i].processed = processed;
1995 p[i].working = 1;
1996 p[i].data_ready = 0;
1997
1998 /* try to split chunks on "path" boundaries */
1999 while (sub_size && sub_size < list_size &&
2000 list[sub_size]->hash &&
2001 list[sub_size]->hash == list[sub_size-1]->hash)
2002 sub_size++;
2003
2004 p[i].list = list;
2005 p[i].list_size = sub_size;
2006 p[i].remaining = sub_size;
2007
2008 list += sub_size;
2009 list_size -= sub_size;
2010 }
2011
2012 /* Start work threads. */
2013 for (i = 0; i < delta_search_threads; i++) {
2014 if (!p[i].list_size)
2015 continue;
2016 pthread_mutex_init(&p[i].mutex, NULL);
2017 pthread_cond_init(&p[i].cond, NULL);
2018 ret = pthread_create(&p[i].thread, NULL,
2019 threaded_find_deltas, &p[i]);
2020 if (ret)
2021 die("unable to create thread: %s", strerror(ret));
2022 active_threads++;
2023 }
2024
2025 /*
2026 * Now let's wait for work completion. Each time a thread is done
2027 * with its work, we steal half of the remaining work from the
2028 * thread with the largest number of unprocessed objects and give
2029 * it to that newly idle thread. This ensure good load balancing
2030 * until the remaining object list segments are simply too short
2031 * to be worth splitting anymore.
2032 */
2033 while (active_threads) {
2034 struct thread_params *target = NULL;
2035 struct thread_params *victim = NULL;
2036 unsigned sub_size = 0;
2037
2038 progress_lock();
2039 for (;;) {
2040 for (i = 0; !target && i < delta_search_threads; i++)
2041 if (!p[i].working)
2042 target = &p[i];
2043 if (target)
2044 break;
2045 pthread_cond_wait(&progress_cond, &progress_mutex);
2046 }
2047
2048 for (i = 0; i < delta_search_threads; i++)
2049 if (p[i].remaining > 2*window &&
2050 (!victim || victim->remaining < p[i].remaining))
2051 victim = &p[i];
2052 if (victim) {
2053 sub_size = victim->remaining / 2;
2054 list = victim->list + victim->list_size - sub_size;
2055 while (sub_size && list[0]->hash &&
2056 list[0]->hash == list[-1]->hash) {
2057 list++;
2058 sub_size--;
2059 }
2060 if (!sub_size) {
2061 /*
2062 * It is possible for some "paths" to have
2063 * so many objects that no hash boundary
2064 * might be found. Let's just steal the
2065 * exact half in that case.
2066 */
2067 sub_size = victim->remaining / 2;
2068 list -= sub_size;
2069 }
2070 target->list = list;
2071 victim->list_size -= sub_size;
2072 victim->remaining -= sub_size;
2073 }
2074 target->list_size = sub_size;
2075 target->remaining = sub_size;
2076 target->working = 1;
2077 progress_unlock();
2078
2079 pthread_mutex_lock(&target->mutex);
2080 target->data_ready = 1;
2081 pthread_cond_signal(&target->cond);
2082 pthread_mutex_unlock(&target->mutex);
2083
2084 if (!sub_size) {
2085 pthread_join(target->thread, NULL);
2086 pthread_cond_destroy(&target->cond);
2087 pthread_mutex_destroy(&target->mutex);
2088 active_threads--;
2089 }
2090 }
2091 cleanup_threaded_search();
2092 free(p);
2093}
2094
2095#else
2096#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2097#endif
2098
2099static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
2100{
2101 unsigned char peeled[20];
2102
2103 if (starts_with(path, "refs/tags/") && /* is a tag? */
2104 !peel_ref(path, peeled) && /* peelable? */
2105 packlist_find(&to_pack, peeled, NULL)) /* object packed? */
2106 add_object_entry(sha1, OBJ_TAG, NULL, 0);
2107 return 0;
2108}
2109
2110static void prepare_pack(int window, int depth)
2111{
2112 struct object_entry **delta_list;
2113 uint32_t i, nr_deltas;
2114 unsigned n;
2115
2116 get_object_details();
2117
2118 /*
2119 * If we're locally repacking then we need to be doubly careful
2120 * from now on in order to make sure no stealth corruption gets
2121 * propagated to the new pack. Clients receiving streamed packs
2122 * should validate everything they get anyway so no need to incur
2123 * the additional cost here in that case.
2124 */
2125 if (!pack_to_stdout)
2126 do_check_packed_object_crc = 1;
2127
2128 if (!to_pack.nr_objects || !window || !depth)
2129 return;
2130
2131 delta_list = xmalloc(to_pack.nr_objects * sizeof(*delta_list));
2132 nr_deltas = n = 0;
2133
2134 for (i = 0; i < to_pack.nr_objects; i++) {
2135 struct object_entry *entry = to_pack.objects + i;
2136
2137 if (entry->delta)
2138 /* This happens if we decided to reuse existing
2139 * delta from a pack. "reuse_delta &&" is implied.
2140 */
2141 continue;
2142
2143 if (entry->size < 50)
2144 continue;
2145
2146 if (entry->no_try_delta)
2147 continue;
2148
2149 if (!entry->preferred_base) {
2150 nr_deltas++;
2151 if (entry->type < 0)
2152 die("unable to get type of object %s",
2153 sha1_to_hex(entry->idx.sha1));
2154 } else {
2155 if (entry->type < 0) {
2156 /*
2157 * This object is not found, but we
2158 * don't have to include it anyway.
2159 */
2160 continue;
2161 }
2162 }
2163
2164 delta_list[n++] = entry;
2165 }
2166
2167 if (nr_deltas && n > 1) {
2168 unsigned nr_done = 0;
2169 if (progress)
2170 progress_state = start_progress(_("Compressing objects"),
2171 nr_deltas);
2172 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2173 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2174 stop_progress(&progress_state);
2175 if (nr_done != nr_deltas)
2176 die("inconsistency with delta count");
2177 }
2178 free(delta_list);
2179}
2180
2181static int git_pack_config(const char *k, const char *v, void *cb)
2182{
2183 if (!strcmp(k, "pack.window")) {
2184 window = git_config_int(k, v);
2185 return 0;
2186 }
2187 if (!strcmp(k, "pack.windowmemory")) {
2188 window_memory_limit = git_config_ulong(k, v);
2189 return 0;
2190 }
2191 if (!strcmp(k, "pack.depth")) {
2192 depth = git_config_int(k, v);
2193 return 0;
2194 }
2195 if (!strcmp(k, "pack.compression")) {
2196 int level = git_config_int(k, v);
2197 if (level == -1)
2198 level = Z_DEFAULT_COMPRESSION;
2199 else if (level < 0 || level > Z_BEST_COMPRESSION)
2200 die("bad pack compression level %d", level);
2201 pack_compression_level = level;
2202 pack_compression_seen = 1;
2203 return 0;
2204 }
2205 if (!strcmp(k, "pack.deltacachesize")) {
2206 max_delta_cache_size = git_config_int(k, v);
2207 return 0;
2208 }
2209 if (!strcmp(k, "pack.deltacachelimit")) {
2210 cache_max_small_delta_size = git_config_int(k, v);
2211 return 0;
2212 }
2213 if (!strcmp(k, "pack.writebitmaps")) {
2214 write_bitmap_index = git_config_bool(k, v);
2215 return 0;
2216 }
2217 if (!strcmp(k, "pack.writebitmaphashcache")) {
2218 if (git_config_bool(k, v))
2219 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2220 else
2221 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2222 }
2223 if (!strcmp(k, "pack.usebitmaps")) {
2224 use_bitmap_index = git_config_bool(k, v);
2225 return 0;
2226 }
2227 if (!strcmp(k, "pack.threads")) {
2228 delta_search_threads = git_config_int(k, v);
2229 if (delta_search_threads < 0)
2230 die("invalid number of threads specified (%d)",
2231 delta_search_threads);
2232#ifdef NO_PTHREADS
2233 if (delta_search_threads != 1)
2234 warning("no threads support, ignoring %s", k);
2235#endif
2236 return 0;
2237 }
2238 if (!strcmp(k, "pack.indexversion")) {
2239 pack_idx_opts.version = git_config_int(k, v);
2240 if (pack_idx_opts.version > 2)
2241 die("bad pack.indexversion=%"PRIu32,
2242 pack_idx_opts.version);
2243 return 0;
2244 }
2245 return git_default_config(k, v, cb);
2246}
2247
2248static void read_object_list_from_stdin(void)
2249{
2250 char line[40 + 1 + PATH_MAX + 2];
2251 unsigned char sha1[20];
2252
2253 for (;;) {
2254 if (!fgets(line, sizeof(line), stdin)) {
2255 if (feof(stdin))
2256 break;
2257 if (!ferror(stdin))
2258 die("fgets returned NULL, not EOF, not error!");
2259 if (errno != EINTR)
2260 die_errno("fgets");
2261 clearerr(stdin);
2262 continue;
2263 }
2264 if (line[0] == '-') {
2265 if (get_sha1_hex(line+1, sha1))
2266 die("expected edge sha1, got garbage:\n %s",
2267 line);
2268 add_preferred_base(sha1);
2269 continue;
2270 }
2271 if (get_sha1_hex(line, sha1))
2272 die("expected sha1, got garbage:\n %s", line);
2273
2274 add_preferred_base_object(line+41);
2275 add_object_entry(sha1, 0, line+41, 0);
2276 }
2277}
2278
2279#define OBJECT_ADDED (1u<<20)
2280
2281static void show_commit(struct commit *commit, void *data)
2282{
2283 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2284 commit->object.flags |= OBJECT_ADDED;
2285
2286 if (write_bitmap_index)
2287 index_commit_for_bitmap(commit);
2288}
2289
2290static void show_object(struct object *obj,
2291 const struct name_path *path, const char *last,
2292 void *data)
2293{
2294 char *name = path_name(path, last);
2295
2296 add_preferred_base_object(name);
2297 add_object_entry(obj->sha1, obj->type, name, 0);
2298 obj->flags |= OBJECT_ADDED;
2299
2300 /*
2301 * We will have generated the hash from the name,
2302 * but not saved a pointer to it - we can free it
2303 */
2304 free((char *)name);
2305}
2306
2307static void show_edge(struct commit *commit)
2308{
2309 add_preferred_base(commit->object.sha1);
2310}
2311
2312struct in_pack_object {
2313 off_t offset;
2314 struct object *object;
2315};
2316
2317struct in_pack {
2318 int alloc;
2319 int nr;
2320 struct in_pack_object *array;
2321};
2322
2323static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2324{
2325 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2326 in_pack->array[in_pack->nr].object = object;
2327 in_pack->nr++;
2328}
2329
2330/*
2331 * Compare the objects in the offset order, in order to emulate the
2332 * "git rev-list --objects" output that produced the pack originally.
2333 */
2334static int ofscmp(const void *a_, const void *b_)
2335{
2336 struct in_pack_object *a = (struct in_pack_object *)a_;
2337 struct in_pack_object *b = (struct in_pack_object *)b_;
2338
2339 if (a->offset < b->offset)
2340 return -1;
2341 else if (a->offset > b->offset)
2342 return 1;
2343 else
2344 return hashcmp(a->object->sha1, b->object->sha1);
2345}
2346
2347static void add_objects_in_unpacked_packs(struct rev_info *revs)
2348{
2349 struct packed_git *p;
2350 struct in_pack in_pack;
2351 uint32_t i;
2352
2353 memset(&in_pack, 0, sizeof(in_pack));
2354
2355 for (p = packed_git; p; p = p->next) {
2356 const unsigned char *sha1;
2357 struct object *o;
2358
2359 if (!p->pack_local || p->pack_keep)
2360 continue;
2361 if (open_pack_index(p))
2362 die("cannot open pack index");
2363
2364 ALLOC_GROW(in_pack.array,
2365 in_pack.nr + p->num_objects,
2366 in_pack.alloc);
2367
2368 for (i = 0; i < p->num_objects; i++) {
2369 sha1 = nth_packed_object_sha1(p, i);
2370 o = lookup_unknown_object(sha1);
2371 if (!(o->flags & OBJECT_ADDED))
2372 mark_in_pack_object(o, p, &in_pack);
2373 o->flags |= OBJECT_ADDED;
2374 }
2375 }
2376
2377 if (in_pack.nr) {
2378 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2379 ofscmp);
2380 for (i = 0; i < in_pack.nr; i++) {
2381 struct object *o = in_pack.array[i].object;
2382 add_object_entry(o->sha1, o->type, "", 0);
2383 }
2384 }
2385 free(in_pack.array);
2386}
2387
2388static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2389{
2390 static struct packed_git *last_found = (void *)1;
2391 struct packed_git *p;
2392
2393 p = (last_found != (void *)1) ? last_found : packed_git;
2394
2395 while (p) {
2396 if ((!p->pack_local || p->pack_keep) &&
2397 find_pack_entry_one(sha1, p)) {
2398 last_found = p;
2399 return 1;
2400 }
2401 if (p == last_found)
2402 p = packed_git;
2403 else
2404 p = p->next;
2405 if (p == last_found)
2406 p = p->next;
2407 }
2408 return 0;
2409}
2410
2411static void loosen_unused_packed_objects(struct rev_info *revs)
2412{
2413 struct packed_git *p;
2414 uint32_t i;
2415 const unsigned char *sha1;
2416
2417 for (p = packed_git; p; p = p->next) {
2418 if (!p->pack_local || p->pack_keep)
2419 continue;
2420
2421 if (unpack_unreachable_expiration &&
2422 p->mtime < unpack_unreachable_expiration)
2423 continue;
2424
2425 if (open_pack_index(p))
2426 die("cannot open pack index");
2427
2428 for (i = 0; i < p->num_objects; i++) {
2429 sha1 = nth_packed_object_sha1(p, i);
2430 if (!packlist_find(&to_pack, sha1, NULL) &&
2431 !has_sha1_pack_kept_or_nonlocal(sha1))
2432 if (force_object_loose(sha1, p->mtime))
2433 die("unable to force loose object");
2434 }
2435 }
2436}
2437
2438static int get_object_list_from_bitmap(struct rev_info *revs)
2439{
2440 if (prepare_bitmap_walk(revs) < 0)
2441 return -1;
2442
2443 if (!reuse_partial_packfile_from_bitmap(
2444 &reuse_packfile,
2445 &reuse_packfile_objects,
2446 &reuse_packfile_offset)) {
2447 assert(reuse_packfile_objects);
2448 nr_result += reuse_packfile_objects;
2449
2450 if (progress) {
2451 fprintf(stderr, "Reusing existing pack: %d, done.\n",
2452 reuse_packfile_objects);
2453 fflush(stderr);
2454 }
2455 }
2456
2457 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2458 return 0;
2459}
2460
2461static void get_object_list(int ac, const char **av)
2462{
2463 struct rev_info revs;
2464 char line[1000];
2465 int flags = 0;
2466
2467 init_revisions(&revs, NULL);
2468 save_commit_buffer = 0;
2469 setup_revisions(ac, av, &revs, NULL);
2470
2471 while (fgets(line, sizeof(line), stdin) != NULL) {
2472 int len = strlen(line);
2473 if (len && line[len - 1] == '\n')
2474 line[--len] = 0;
2475 if (!len)
2476 break;
2477 if (*line == '-') {
2478 if (!strcmp(line, "--not")) {
2479 flags ^= UNINTERESTING;
2480 write_bitmap_index = 0;
2481 continue;
2482 }
2483 die("not a rev '%s'", line);
2484 }
2485 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2486 die("bad revision '%s'", line);
2487 }
2488
2489 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2490 return;
2491
2492 if (prepare_revision_walk(&revs))
2493 die("revision walk setup failed");
2494 mark_edges_uninteresting(&revs, show_edge);
2495 traverse_commit_list(&revs, show_commit, show_object, NULL);
2496
2497 if (keep_unreachable)
2498 add_objects_in_unpacked_packs(&revs);
2499 if (unpack_unreachable)
2500 loosen_unused_packed_objects(&revs);
2501}
2502
2503static int option_parse_index_version(const struct option *opt,
2504 const char *arg, int unset)
2505{
2506 char *c;
2507 const char *val = arg;
2508 pack_idx_opts.version = strtoul(val, &c, 10);
2509 if (pack_idx_opts.version > 2)
2510 die(_("unsupported index version %s"), val);
2511 if (*c == ',' && c[1])
2512 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2513 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2514 die(_("bad index version '%s'"), val);
2515 return 0;
2516}
2517
2518static int option_parse_unpack_unreachable(const struct option *opt,
2519 const char *arg, int unset)
2520{
2521 if (unset) {
2522 unpack_unreachable = 0;
2523 unpack_unreachable_expiration = 0;
2524 }
2525 else {
2526 unpack_unreachable = 1;
2527 if (arg)
2528 unpack_unreachable_expiration = approxidate(arg);
2529 }
2530 return 0;
2531}
2532
2533static int option_parse_ulong(const struct option *opt,
2534 const char *arg, int unset)
2535{
2536 if (unset)
2537 die(_("option %s does not accept negative form"),
2538 opt->long_name);
2539
2540 if (!git_parse_ulong(arg, opt->value))
2541 die(_("unable to parse value '%s' for option %s"),
2542 arg, opt->long_name);
2543 return 0;
2544}
2545
2546#define OPT_ULONG(s, l, v, h) \
2547 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2548 PARSE_OPT_NONEG, option_parse_ulong }
2549
2550int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2551{
2552 int use_internal_rev_list = 0;
2553 int thin = 0;
2554 int all_progress_implied = 0;
2555 const char *rp_av[6];
2556 int rp_ac = 0;
2557 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2558 struct option pack_objects_options[] = {
2559 OPT_SET_INT('q', "quiet", &progress,
2560 N_("do not show progress meter"), 0),
2561 OPT_SET_INT(0, "progress", &progress,
2562 N_("show progress meter"), 1),
2563 OPT_SET_INT(0, "all-progress", &progress,
2564 N_("show progress meter during object writing phase"), 2),
2565 OPT_BOOL(0, "all-progress-implied",
2566 &all_progress_implied,
2567 N_("similar to --all-progress when progress meter is shown")),
2568 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2569 N_("write the pack index file in the specified idx format version"),
2570 0, option_parse_index_version },
2571 OPT_ULONG(0, "max-pack-size", &pack_size_limit,
2572 N_("maximum size of each output pack file")),
2573 OPT_BOOL(0, "local", &local,
2574 N_("ignore borrowed objects from alternate object store")),
2575 OPT_BOOL(0, "incremental", &incremental,
2576 N_("ignore packed objects")),
2577 OPT_INTEGER(0, "window", &window,
2578 N_("limit pack window by objects")),
2579 OPT_ULONG(0, "window-memory", &window_memory_limit,
2580 N_("limit pack window by memory in addition to object limit")),
2581 OPT_INTEGER(0, "depth", &depth,
2582 N_("maximum length of delta chain allowed in the resulting pack")),
2583 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2584 N_("reuse existing deltas")),
2585 OPT_BOOL(0, "reuse-object", &reuse_object,
2586 N_("reuse existing objects")),
2587 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2588 N_("use OFS_DELTA objects")),
2589 OPT_INTEGER(0, "threads", &delta_search_threads,
2590 N_("use threads when searching for best delta matches")),
2591 OPT_BOOL(0, "non-empty", &non_empty,
2592 N_("do not create an empty pack output")),
2593 OPT_BOOL(0, "revs", &use_internal_rev_list,
2594 N_("read revision arguments from standard input")),
2595 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2596 N_("limit the objects to those that are not yet packed"),
2597 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2598 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2599 N_("include objects reachable from any reference"),
2600 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2601 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2602 N_("include objects referred by reflog entries"),
2603 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2604 OPT_BOOL(0, "stdout", &pack_to_stdout,
2605 N_("output pack to stdout")),
2606 OPT_BOOL(0, "include-tag", &include_tag,
2607 N_("include tag objects that refer to objects to be packed")),
2608 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2609 N_("keep unreachable objects")),
2610 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2611 N_("unpack unreachable objects newer than <time>"),
2612 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2613 OPT_BOOL(0, "thin", &thin,
2614 N_("create thin packs")),
2615 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2616 N_("ignore packs that have companion .keep file")),
2617 OPT_INTEGER(0, "compression", &pack_compression_level,
2618 N_("pack compression level")),
2619 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2620 N_("do not hide commits by grafts"), 0),
2621 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2622 N_("use a bitmap index if available to speed up counting objects")),
2623 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2624 N_("write a bitmap index together with the pack index")),
2625 OPT_END(),
2626 };
2627
2628 check_replace_refs = 0;
2629
2630 reset_pack_idx_option(&pack_idx_opts);
2631 git_config(git_pack_config, NULL);
2632 if (!pack_compression_seen && core_compression_seen)
2633 pack_compression_level = core_compression_level;
2634
2635 progress = isatty(2);
2636 argc = parse_options(argc, argv, prefix, pack_objects_options,
2637 pack_usage, 0);
2638
2639 if (argc) {
2640 base_name = argv[0];
2641 argc--;
2642 }
2643 if (pack_to_stdout != !base_name || argc)
2644 usage_with_options(pack_usage, pack_objects_options);
2645
2646 rp_av[rp_ac++] = "pack-objects";
2647 if (thin) {
2648 use_internal_rev_list = 1;
2649 rp_av[rp_ac++] = "--objects-edge";
2650 } else
2651 rp_av[rp_ac++] = "--objects";
2652
2653 if (rev_list_all) {
2654 use_internal_rev_list = 1;
2655 rp_av[rp_ac++] = "--all";
2656 }
2657 if (rev_list_reflog) {
2658 use_internal_rev_list = 1;
2659 rp_av[rp_ac++] = "--reflog";
2660 }
2661 if (rev_list_unpacked) {
2662 use_internal_rev_list = 1;
2663 rp_av[rp_ac++] = "--unpacked";
2664 }
2665
2666 if (!reuse_object)
2667 reuse_delta = 0;
2668 if (pack_compression_level == -1)
2669 pack_compression_level = Z_DEFAULT_COMPRESSION;
2670 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2671 die("bad pack compression level %d", pack_compression_level);
2672#ifdef NO_PTHREADS
2673 if (delta_search_threads != 1)
2674 warning("no threads support, ignoring --threads");
2675#endif
2676 if (!pack_to_stdout && !pack_size_limit)
2677 pack_size_limit = pack_size_limit_cfg;
2678 if (pack_to_stdout && pack_size_limit)
2679 die("--max-pack-size cannot be used to build a pack for transfer.");
2680 if (pack_size_limit && pack_size_limit < 1024*1024) {
2681 warning("minimum pack size limit is 1 MiB");
2682 pack_size_limit = 1024*1024;
2683 }
2684
2685 if (!pack_to_stdout && thin)
2686 die("--thin cannot be used to build an indexable pack.");
2687
2688 if (keep_unreachable && unpack_unreachable)
2689 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2690
2691 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2692 use_bitmap_index = 0;
2693
2694 if (pack_to_stdout || !rev_list_all)
2695 write_bitmap_index = 0;
2696
2697 if (progress && all_progress_implied)
2698 progress = 2;
2699
2700 prepare_packed_git();
2701
2702 if (progress)
2703 progress_state = start_progress(_("Counting objects"), 0);
2704 if (!use_internal_rev_list)
2705 read_object_list_from_stdin();
2706 else {
2707 rp_av[rp_ac] = NULL;
2708 get_object_list(rp_ac, rp_av);
2709 }
2710 cleanup_preferred_base();
2711 if (include_tag && nr_result)
2712 for_each_ref(add_ref_tag, NULL);
2713 stop_progress(&progress_state);
2714
2715 if (non_empty && !nr_result)
2716 return 0;
2717 if (nr_result)
2718 prepare_pack(window, depth);
2719 write_pack_file();
2720 if (progress)
2721 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2722 " reused %"PRIu32" (delta %"PRIu32")\n",
2723 written, written_delta, reused, reused_delta);
2724 return 0;
2725}