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;
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 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 close(fd);
744 written += reuse_packfile_objects;
745 return reuse_packfile_offset - sizeof(struct pack_header);
746}
747
748static void write_pack_file(void)
749{
750 uint32_t i = 0, j;
751 struct sha1file *f;
752 off_t offset;
753 uint32_t nr_remaining = nr_result;
754 time_t last_mtime = 0;
755 struct object_entry **write_order;
756
757 if (progress > pack_to_stdout)
758 progress_state = start_progress(_("Writing objects"), nr_result);
759 written_list = xmalloc(to_pack.nr_objects * sizeof(*written_list));
760 write_order = compute_write_order();
761
762 do {
763 unsigned char sha1[20];
764 char *pack_tmp_name = NULL;
765
766 if (pack_to_stdout)
767 f = sha1fd_throughput(1, "<stdout>", progress_state);
768 else
769 f = create_tmp_packfile(&pack_tmp_name);
770
771 offset = write_pack_header(f, nr_remaining);
772
773 if (reuse_packfile) {
774 off_t packfile_size;
775 assert(pack_to_stdout);
776
777 packfile_size = write_reused_pack(f);
778 offset += packfile_size;
779 }
780
781 nr_written = 0;
782 for (; i < to_pack.nr_objects; i++) {
783 struct object_entry *e = write_order[i];
784 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
785 break;
786 display_progress(progress_state, written);
787 }
788
789 /*
790 * Did we write the wrong # entries in the header?
791 * If so, rewrite it like in fast-import
792 */
793 if (pack_to_stdout) {
794 sha1close(f, sha1, CSUM_CLOSE);
795 } else if (nr_written == nr_remaining) {
796 sha1close(f, sha1, CSUM_FSYNC);
797 } else {
798 int fd = sha1close(f, sha1, 0);
799 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
800 nr_written, sha1, offset);
801 close(fd);
802 }
803
804 if (!pack_to_stdout) {
805 struct stat st;
806 struct strbuf tmpname = STRBUF_INIT;
807
808 /*
809 * Packs are runtime accessed in their mtime
810 * order since newer packs are more likely to contain
811 * younger objects. So if we are creating multiple
812 * packs then we should modify the mtime of later ones
813 * to preserve this property.
814 */
815 if (stat(pack_tmp_name, &st) < 0) {
816 warning("failed to stat %s: %s",
817 pack_tmp_name, strerror(errno));
818 } else if (!last_mtime) {
819 last_mtime = st.st_mtime;
820 } else {
821 struct utimbuf utb;
822 utb.actime = st.st_atime;
823 utb.modtime = --last_mtime;
824 if (utime(pack_tmp_name, &utb) < 0)
825 warning("failed utime() on %s: %s",
826 pack_tmp_name, strerror(errno));
827 }
828
829 strbuf_addf(&tmpname, "%s-", base_name);
830
831 if (write_bitmap_index) {
832 bitmap_writer_set_checksum(sha1);
833 bitmap_writer_build_type_index(written_list, nr_written);
834 }
835
836 finish_tmp_packfile(&tmpname, pack_tmp_name,
837 written_list, nr_written,
838 &pack_idx_opts, sha1);
839
840 if (write_bitmap_index) {
841 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
842
843 stop_progress(&progress_state);
844
845 bitmap_writer_show_progress(progress);
846 bitmap_writer_reuse_bitmaps(&to_pack);
847 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
848 bitmap_writer_build(&to_pack);
849 bitmap_writer_finish(written_list, nr_written,
850 tmpname.buf, write_bitmap_options);
851 write_bitmap_index = 0;
852 }
853
854 strbuf_release(&tmpname);
855 free(pack_tmp_name);
856 puts(sha1_to_hex(sha1));
857 }
858
859 /* mark written objects as written to previous pack */
860 for (j = 0; j < nr_written; j++) {
861 written_list[j]->offset = (off_t)-1;
862 }
863 nr_remaining -= nr_written;
864 } while (nr_remaining && i < to_pack.nr_objects);
865
866 free(written_list);
867 free(write_order);
868 stop_progress(&progress_state);
869 if (written != nr_result)
870 die("wrote %"PRIu32" objects while expecting %"PRIu32,
871 written, nr_result);
872}
873
874static void setup_delta_attr_check(struct git_attr_check *check)
875{
876 static struct git_attr *attr_delta;
877
878 if (!attr_delta)
879 attr_delta = git_attr("delta");
880
881 check[0].attr = attr_delta;
882}
883
884static int no_try_delta(const char *path)
885{
886 struct git_attr_check check[1];
887
888 setup_delta_attr_check(check);
889 if (git_check_attr(path, ARRAY_SIZE(check), check))
890 return 0;
891 if (ATTR_FALSE(check->value))
892 return 1;
893 return 0;
894}
895
896/*
897 * When adding an object, check whether we have already added it
898 * to our packing list. If so, we can skip. However, if we are
899 * being asked to excludei t, but the previous mention was to include
900 * it, make sure to adjust its flags and tweak our numbers accordingly.
901 *
902 * As an optimization, we pass out the index position where we would have
903 * found the item, since that saves us from having to look it up again a
904 * few lines later when we want to add the new entry.
905 */
906static int have_duplicate_entry(const unsigned char *sha1,
907 int exclude,
908 uint32_t *index_pos)
909{
910 struct object_entry *entry;
911
912 entry = packlist_find(&to_pack, sha1, index_pos);
913 if (!entry)
914 return 0;
915
916 if (exclude) {
917 if (!entry->preferred_base)
918 nr_result--;
919 entry->preferred_base = 1;
920 }
921
922 return 1;
923}
924
925/*
926 * Check whether we want the object in the pack (e.g., we do not want
927 * objects found in non-local stores if the "--local" option was used).
928 *
929 * As a side effect of this check, we will find the packed version of this
930 * object, if any. We therefore pass out the pack information to avoid having
931 * to look it up again later.
932 */
933static int want_object_in_pack(const unsigned char *sha1,
934 int exclude,
935 struct packed_git **found_pack,
936 off_t *found_offset)
937{
938 struct packed_git *p;
939
940 if (!exclude && local && has_loose_object_nonlocal(sha1))
941 return 0;
942
943 *found_pack = NULL;
944 *found_offset = 0;
945
946 for (p = packed_git; p; p = p->next) {
947 off_t offset = find_pack_entry_one(sha1, p);
948 if (offset) {
949 if (!*found_pack) {
950 if (!is_pack_valid(p)) {
951 warning("packfile %s cannot be accessed", p->pack_name);
952 continue;
953 }
954 *found_offset = offset;
955 *found_pack = p;
956 }
957 if (exclude)
958 return 1;
959 if (incremental)
960 return 0;
961 if (local && !p->pack_local)
962 return 0;
963 if (ignore_packed_keep && p->pack_local && p->pack_keep)
964 return 0;
965 }
966 }
967
968 return 1;
969}
970
971static void create_object_entry(const unsigned char *sha1,
972 enum object_type type,
973 uint32_t hash,
974 int exclude,
975 int no_try_delta,
976 uint32_t index_pos,
977 struct packed_git *found_pack,
978 off_t found_offset)
979{
980 struct object_entry *entry;
981
982 entry = packlist_alloc(&to_pack, sha1, index_pos);
983 entry->hash = hash;
984 if (type)
985 entry->type = type;
986 if (exclude)
987 entry->preferred_base = 1;
988 else
989 nr_result++;
990 if (found_pack) {
991 entry->in_pack = found_pack;
992 entry->in_pack_offset = found_offset;
993 }
994
995 entry->no_try_delta = no_try_delta;
996}
997
998static int add_object_entry(const unsigned char *sha1, enum object_type type,
999 const char *name, int exclude)
1000{
1001 struct packed_git *found_pack;
1002 off_t found_offset;
1003 uint32_t index_pos;
1004
1005 if (have_duplicate_entry(sha1, exclude, &index_pos))
1006 return 0;
1007
1008 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset))
1009 return 0;
1010
1011 create_object_entry(sha1, type, pack_name_hash(name),
1012 exclude, name && no_try_delta(name),
1013 index_pos, found_pack, found_offset);
1014
1015 display_progress(progress_state, to_pack.nr_objects);
1016 return 1;
1017}
1018
1019static int add_object_entry_from_bitmap(const unsigned char *sha1,
1020 enum object_type type,
1021 int flags, uint32_t name_hash,
1022 struct packed_git *pack, off_t offset)
1023{
1024 uint32_t index_pos;
1025
1026 if (have_duplicate_entry(sha1, 0, &index_pos))
1027 return 0;
1028
1029 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1030
1031 display_progress(progress_state, to_pack.nr_objects);
1032 return 1;
1033}
1034
1035struct pbase_tree_cache {
1036 unsigned char sha1[20];
1037 int ref;
1038 int temporary;
1039 void *tree_data;
1040 unsigned long tree_size;
1041};
1042
1043static struct pbase_tree_cache *(pbase_tree_cache[256]);
1044static int pbase_tree_cache_ix(const unsigned char *sha1)
1045{
1046 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1047}
1048static int pbase_tree_cache_ix_incr(int ix)
1049{
1050 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1051}
1052
1053static struct pbase_tree {
1054 struct pbase_tree *next;
1055 /* This is a phony "cache" entry; we are not
1056 * going to evict it nor find it through _get()
1057 * mechanism -- this is for the toplevel node that
1058 * would almost always change with any commit.
1059 */
1060 struct pbase_tree_cache pcache;
1061} *pbase_tree;
1062
1063static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1064{
1065 struct pbase_tree_cache *ent, *nent;
1066 void *data;
1067 unsigned long size;
1068 enum object_type type;
1069 int neigh;
1070 int my_ix = pbase_tree_cache_ix(sha1);
1071 int available_ix = -1;
1072
1073 /* pbase-tree-cache acts as a limited hashtable.
1074 * your object will be found at your index or within a few
1075 * slots after that slot if it is cached.
1076 */
1077 for (neigh = 0; neigh < 8; neigh++) {
1078 ent = pbase_tree_cache[my_ix];
1079 if (ent && !hashcmp(ent->sha1, sha1)) {
1080 ent->ref++;
1081 return ent;
1082 }
1083 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1084 ((0 <= available_ix) &&
1085 (!ent && pbase_tree_cache[available_ix])))
1086 available_ix = my_ix;
1087 if (!ent)
1088 break;
1089 my_ix = pbase_tree_cache_ix_incr(my_ix);
1090 }
1091
1092 /* Did not find one. Either we got a bogus request or
1093 * we need to read and perhaps cache.
1094 */
1095 data = read_sha1_file(sha1, &type, &size);
1096 if (!data)
1097 return NULL;
1098 if (type != OBJ_TREE) {
1099 free(data);
1100 return NULL;
1101 }
1102
1103 /* We need to either cache or return a throwaway copy */
1104
1105 if (available_ix < 0)
1106 ent = NULL;
1107 else {
1108 ent = pbase_tree_cache[available_ix];
1109 my_ix = available_ix;
1110 }
1111
1112 if (!ent) {
1113 nent = xmalloc(sizeof(*nent));
1114 nent->temporary = (available_ix < 0);
1115 }
1116 else {
1117 /* evict and reuse */
1118 free(ent->tree_data);
1119 nent = ent;
1120 }
1121 hashcpy(nent->sha1, sha1);
1122 nent->tree_data = data;
1123 nent->tree_size = size;
1124 nent->ref = 1;
1125 if (!nent->temporary)
1126 pbase_tree_cache[my_ix] = nent;
1127 return nent;
1128}
1129
1130static void pbase_tree_put(struct pbase_tree_cache *cache)
1131{
1132 if (!cache->temporary) {
1133 cache->ref--;
1134 return;
1135 }
1136 free(cache->tree_data);
1137 free(cache);
1138}
1139
1140static int name_cmp_len(const char *name)
1141{
1142 int i;
1143 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1144 ;
1145 return i;
1146}
1147
1148static void add_pbase_object(struct tree_desc *tree,
1149 const char *name,
1150 int cmplen,
1151 const char *fullname)
1152{
1153 struct name_entry entry;
1154 int cmp;
1155
1156 while (tree_entry(tree,&entry)) {
1157 if (S_ISGITLINK(entry.mode))
1158 continue;
1159 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1160 memcmp(name, entry.path, cmplen);
1161 if (cmp > 0)
1162 continue;
1163 if (cmp < 0)
1164 return;
1165 if (name[cmplen] != '/') {
1166 add_object_entry(entry.sha1,
1167 object_type(entry.mode),
1168 fullname, 1);
1169 return;
1170 }
1171 if (S_ISDIR(entry.mode)) {
1172 struct tree_desc sub;
1173 struct pbase_tree_cache *tree;
1174 const char *down = name+cmplen+1;
1175 int downlen = name_cmp_len(down);
1176
1177 tree = pbase_tree_get(entry.sha1);
1178 if (!tree)
1179 return;
1180 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1181
1182 add_pbase_object(&sub, down, downlen, fullname);
1183 pbase_tree_put(tree);
1184 }
1185 }
1186}
1187
1188static unsigned *done_pbase_paths;
1189static int done_pbase_paths_num;
1190static int done_pbase_paths_alloc;
1191static int done_pbase_path_pos(unsigned hash)
1192{
1193 int lo = 0;
1194 int hi = done_pbase_paths_num;
1195 while (lo < hi) {
1196 int mi = (hi + lo) / 2;
1197 if (done_pbase_paths[mi] == hash)
1198 return mi;
1199 if (done_pbase_paths[mi] < hash)
1200 hi = mi;
1201 else
1202 lo = mi + 1;
1203 }
1204 return -lo-1;
1205}
1206
1207static int check_pbase_path(unsigned hash)
1208{
1209 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1210 if (0 <= pos)
1211 return 1;
1212 pos = -pos - 1;
1213 ALLOC_GROW(done_pbase_paths,
1214 done_pbase_paths_num + 1,
1215 done_pbase_paths_alloc);
1216 done_pbase_paths_num++;
1217 if (pos < done_pbase_paths_num)
1218 memmove(done_pbase_paths + pos + 1,
1219 done_pbase_paths + pos,
1220 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1221 done_pbase_paths[pos] = hash;
1222 return 0;
1223}
1224
1225static void add_preferred_base_object(const char *name)
1226{
1227 struct pbase_tree *it;
1228 int cmplen;
1229 unsigned hash = pack_name_hash(name);
1230
1231 if (!num_preferred_base || check_pbase_path(hash))
1232 return;
1233
1234 cmplen = name_cmp_len(name);
1235 for (it = pbase_tree; it; it = it->next) {
1236 if (cmplen == 0) {
1237 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1238 }
1239 else {
1240 struct tree_desc tree;
1241 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1242 add_pbase_object(&tree, name, cmplen, name);
1243 }
1244 }
1245}
1246
1247static void add_preferred_base(unsigned char *sha1)
1248{
1249 struct pbase_tree *it;
1250 void *data;
1251 unsigned long size;
1252 unsigned char tree_sha1[20];
1253
1254 if (window <= num_preferred_base++)
1255 return;
1256
1257 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1258 if (!data)
1259 return;
1260
1261 for (it = pbase_tree; it; it = it->next) {
1262 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1263 free(data);
1264 return;
1265 }
1266 }
1267
1268 it = xcalloc(1, sizeof(*it));
1269 it->next = pbase_tree;
1270 pbase_tree = it;
1271
1272 hashcpy(it->pcache.sha1, tree_sha1);
1273 it->pcache.tree_data = data;
1274 it->pcache.tree_size = size;
1275}
1276
1277static void cleanup_preferred_base(void)
1278{
1279 struct pbase_tree *it;
1280 unsigned i;
1281
1282 it = pbase_tree;
1283 pbase_tree = NULL;
1284 while (it) {
1285 struct pbase_tree *this = it;
1286 it = this->next;
1287 free(this->pcache.tree_data);
1288 free(this);
1289 }
1290
1291 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1292 if (!pbase_tree_cache[i])
1293 continue;
1294 free(pbase_tree_cache[i]->tree_data);
1295 free(pbase_tree_cache[i]);
1296 pbase_tree_cache[i] = NULL;
1297 }
1298
1299 free(done_pbase_paths);
1300 done_pbase_paths = NULL;
1301 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1302}
1303
1304static void check_object(struct object_entry *entry)
1305{
1306 if (entry->in_pack) {
1307 struct packed_git *p = entry->in_pack;
1308 struct pack_window *w_curs = NULL;
1309 const unsigned char *base_ref = NULL;
1310 struct object_entry *base_entry;
1311 unsigned long used, used_0;
1312 unsigned long avail;
1313 off_t ofs;
1314 unsigned char *buf, c;
1315
1316 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1317
1318 /*
1319 * We want in_pack_type even if we do not reuse delta
1320 * since non-delta representations could still be reused.
1321 */
1322 used = unpack_object_header_buffer(buf, avail,
1323 &entry->in_pack_type,
1324 &entry->size);
1325 if (used == 0)
1326 goto give_up;
1327
1328 /*
1329 * Determine if this is a delta and if so whether we can
1330 * reuse it or not. Otherwise let's find out as cheaply as
1331 * possible what the actual type and size for this object is.
1332 */
1333 switch (entry->in_pack_type) {
1334 default:
1335 /* Not a delta hence we've already got all we need. */
1336 entry->type = entry->in_pack_type;
1337 entry->in_pack_header_size = used;
1338 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1339 goto give_up;
1340 unuse_pack(&w_curs);
1341 return;
1342 case OBJ_REF_DELTA:
1343 if (reuse_delta && !entry->preferred_base)
1344 base_ref = use_pack(p, &w_curs,
1345 entry->in_pack_offset + used, NULL);
1346 entry->in_pack_header_size = used + 20;
1347 break;
1348 case OBJ_OFS_DELTA:
1349 buf = use_pack(p, &w_curs,
1350 entry->in_pack_offset + used, NULL);
1351 used_0 = 0;
1352 c = buf[used_0++];
1353 ofs = c & 127;
1354 while (c & 128) {
1355 ofs += 1;
1356 if (!ofs || MSB(ofs, 7)) {
1357 error("delta base offset overflow in pack for %s",
1358 sha1_to_hex(entry->idx.sha1));
1359 goto give_up;
1360 }
1361 c = buf[used_0++];
1362 ofs = (ofs << 7) + (c & 127);
1363 }
1364 ofs = entry->in_pack_offset - ofs;
1365 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1366 error("delta base offset out of bound for %s",
1367 sha1_to_hex(entry->idx.sha1));
1368 goto give_up;
1369 }
1370 if (reuse_delta && !entry->preferred_base) {
1371 struct revindex_entry *revidx;
1372 revidx = find_pack_revindex(p, ofs);
1373 if (!revidx)
1374 goto give_up;
1375 base_ref = nth_packed_object_sha1(p, revidx->nr);
1376 }
1377 entry->in_pack_header_size = used + used_0;
1378 break;
1379 }
1380
1381 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1382 /*
1383 * If base_ref was set above that means we wish to
1384 * reuse delta data, and we even found that base
1385 * in the list of objects we want to pack. Goodie!
1386 *
1387 * Depth value does not matter - find_deltas() will
1388 * never consider reused delta as the base object to
1389 * deltify other objects against, in order to avoid
1390 * circular deltas.
1391 */
1392 entry->type = entry->in_pack_type;
1393 entry->delta = base_entry;
1394 entry->delta_size = entry->size;
1395 entry->delta_sibling = base_entry->delta_child;
1396 base_entry->delta_child = entry;
1397 unuse_pack(&w_curs);
1398 return;
1399 }
1400
1401 if (entry->type) {
1402 /*
1403 * This must be a delta and we already know what the
1404 * final object type is. Let's extract the actual
1405 * object size from the delta header.
1406 */
1407 entry->size = get_size_from_delta(p, &w_curs,
1408 entry->in_pack_offset + entry->in_pack_header_size);
1409 if (entry->size == 0)
1410 goto give_up;
1411 unuse_pack(&w_curs);
1412 return;
1413 }
1414
1415 /*
1416 * No choice but to fall back to the recursive delta walk
1417 * with sha1_object_info() to find about the object type
1418 * at this point...
1419 */
1420 give_up:
1421 unuse_pack(&w_curs);
1422 }
1423
1424 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1425 /*
1426 * The error condition is checked in prepare_pack(). This is
1427 * to permit a missing preferred base object to be ignored
1428 * as a preferred base. Doing so can result in a larger
1429 * pack file, but the transfer will still take place.
1430 */
1431}
1432
1433static int pack_offset_sort(const void *_a, const void *_b)
1434{
1435 const struct object_entry *a = *(struct object_entry **)_a;
1436 const struct object_entry *b = *(struct object_entry **)_b;
1437
1438 /* avoid filesystem trashing with loose objects */
1439 if (!a->in_pack && !b->in_pack)
1440 return hashcmp(a->idx.sha1, b->idx.sha1);
1441
1442 if (a->in_pack < b->in_pack)
1443 return -1;
1444 if (a->in_pack > b->in_pack)
1445 return 1;
1446 return a->in_pack_offset < b->in_pack_offset ? -1 :
1447 (a->in_pack_offset > b->in_pack_offset);
1448}
1449
1450static void get_object_details(void)
1451{
1452 uint32_t i;
1453 struct object_entry **sorted_by_offset;
1454
1455 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1456 for (i = 0; i < to_pack.nr_objects; i++)
1457 sorted_by_offset[i] = to_pack.objects + i;
1458 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1459
1460 for (i = 0; i < to_pack.nr_objects; i++) {
1461 struct object_entry *entry = sorted_by_offset[i];
1462 check_object(entry);
1463 if (big_file_threshold < entry->size)
1464 entry->no_try_delta = 1;
1465 }
1466
1467 free(sorted_by_offset);
1468}
1469
1470/*
1471 * We search for deltas in a list sorted by type, by filename hash, and then
1472 * by size, so that we see progressively smaller and smaller files.
1473 * That's because we prefer deltas to be from the bigger file
1474 * to the smaller -- deletes are potentially cheaper, but perhaps
1475 * more importantly, the bigger file is likely the more recent
1476 * one. The deepest deltas are therefore the oldest objects which are
1477 * less susceptible to be accessed often.
1478 */
1479static int type_size_sort(const void *_a, const void *_b)
1480{
1481 const struct object_entry *a = *(struct object_entry **)_a;
1482 const struct object_entry *b = *(struct object_entry **)_b;
1483
1484 if (a->type > b->type)
1485 return -1;
1486 if (a->type < b->type)
1487 return 1;
1488 if (a->hash > b->hash)
1489 return -1;
1490 if (a->hash < b->hash)
1491 return 1;
1492 if (a->preferred_base > b->preferred_base)
1493 return -1;
1494 if (a->preferred_base < b->preferred_base)
1495 return 1;
1496 if (a->size > b->size)
1497 return -1;
1498 if (a->size < b->size)
1499 return 1;
1500 return a < b ? -1 : (a > b); /* newest first */
1501}
1502
1503struct unpacked {
1504 struct object_entry *entry;
1505 void *data;
1506 struct delta_index *index;
1507 unsigned depth;
1508};
1509
1510static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1511 unsigned long delta_size)
1512{
1513 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1514 return 0;
1515
1516 if (delta_size < cache_max_small_delta_size)
1517 return 1;
1518
1519 /* cache delta, if objects are large enough compared to delta size */
1520 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1521 return 1;
1522
1523 return 0;
1524}
1525
1526#ifndef NO_PTHREADS
1527
1528static pthread_mutex_t read_mutex;
1529#define read_lock() pthread_mutex_lock(&read_mutex)
1530#define read_unlock() pthread_mutex_unlock(&read_mutex)
1531
1532static pthread_mutex_t cache_mutex;
1533#define cache_lock() pthread_mutex_lock(&cache_mutex)
1534#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1535
1536static pthread_mutex_t progress_mutex;
1537#define progress_lock() pthread_mutex_lock(&progress_mutex)
1538#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1539
1540#else
1541
1542#define read_lock() (void)0
1543#define read_unlock() (void)0
1544#define cache_lock() (void)0
1545#define cache_unlock() (void)0
1546#define progress_lock() (void)0
1547#define progress_unlock() (void)0
1548
1549#endif
1550
1551static int try_delta(struct unpacked *trg, struct unpacked *src,
1552 unsigned max_depth, unsigned long *mem_usage)
1553{
1554 struct object_entry *trg_entry = trg->entry;
1555 struct object_entry *src_entry = src->entry;
1556 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1557 unsigned ref_depth;
1558 enum object_type type;
1559 void *delta_buf;
1560
1561 /* Don't bother doing diffs between different types */
1562 if (trg_entry->type != src_entry->type)
1563 return -1;
1564
1565 /*
1566 * We do not bother to try a delta that we discarded on an
1567 * earlier try, but only when reusing delta data. Note that
1568 * src_entry that is marked as the preferred_base should always
1569 * be considered, as even if we produce a suboptimal delta against
1570 * it, we will still save the transfer cost, as we already know
1571 * the other side has it and we won't send src_entry at all.
1572 */
1573 if (reuse_delta && trg_entry->in_pack &&
1574 trg_entry->in_pack == src_entry->in_pack &&
1575 !src_entry->preferred_base &&
1576 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1577 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1578 return 0;
1579
1580 /* Let's not bust the allowed depth. */
1581 if (src->depth >= max_depth)
1582 return 0;
1583
1584 /* Now some size filtering heuristics. */
1585 trg_size = trg_entry->size;
1586 if (!trg_entry->delta) {
1587 max_size = trg_size/2 - 20;
1588 ref_depth = 1;
1589 } else {
1590 max_size = trg_entry->delta_size;
1591 ref_depth = trg->depth;
1592 }
1593 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1594 (max_depth - ref_depth + 1);
1595 if (max_size == 0)
1596 return 0;
1597 src_size = src_entry->size;
1598 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1599 if (sizediff >= max_size)
1600 return 0;
1601 if (trg_size < src_size / 32)
1602 return 0;
1603
1604 /* Load data if not already done */
1605 if (!trg->data) {
1606 read_lock();
1607 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1608 read_unlock();
1609 if (!trg->data)
1610 die("object %s cannot be read",
1611 sha1_to_hex(trg_entry->idx.sha1));
1612 if (sz != trg_size)
1613 die("object %s inconsistent object length (%lu vs %lu)",
1614 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1615 *mem_usage += sz;
1616 }
1617 if (!src->data) {
1618 read_lock();
1619 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1620 read_unlock();
1621 if (!src->data) {
1622 if (src_entry->preferred_base) {
1623 static int warned = 0;
1624 if (!warned++)
1625 warning("object %s cannot be read",
1626 sha1_to_hex(src_entry->idx.sha1));
1627 /*
1628 * Those objects are not included in the
1629 * resulting pack. Be resilient and ignore
1630 * them if they can't be read, in case the
1631 * pack could be created nevertheless.
1632 */
1633 return 0;
1634 }
1635 die("object %s cannot be read",
1636 sha1_to_hex(src_entry->idx.sha1));
1637 }
1638 if (sz != src_size)
1639 die("object %s inconsistent object length (%lu vs %lu)",
1640 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1641 *mem_usage += sz;
1642 }
1643 if (!src->index) {
1644 src->index = create_delta_index(src->data, src_size);
1645 if (!src->index) {
1646 static int warned = 0;
1647 if (!warned++)
1648 warning("suboptimal pack - out of memory");
1649 return 0;
1650 }
1651 *mem_usage += sizeof_delta_index(src->index);
1652 }
1653
1654 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1655 if (!delta_buf)
1656 return 0;
1657
1658 if (trg_entry->delta) {
1659 /* Prefer only shallower same-sized deltas. */
1660 if (delta_size == trg_entry->delta_size &&
1661 src->depth + 1 >= trg->depth) {
1662 free(delta_buf);
1663 return 0;
1664 }
1665 }
1666
1667 /*
1668 * Handle memory allocation outside of the cache
1669 * accounting lock. Compiler will optimize the strangeness
1670 * away when NO_PTHREADS is defined.
1671 */
1672 free(trg_entry->delta_data);
1673 cache_lock();
1674 if (trg_entry->delta_data) {
1675 delta_cache_size -= trg_entry->delta_size;
1676 trg_entry->delta_data = NULL;
1677 }
1678 if (delta_cacheable(src_size, trg_size, delta_size)) {
1679 delta_cache_size += delta_size;
1680 cache_unlock();
1681 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1682 } else {
1683 cache_unlock();
1684 free(delta_buf);
1685 }
1686
1687 trg_entry->delta = src_entry;
1688 trg_entry->delta_size = delta_size;
1689 trg->depth = src->depth + 1;
1690
1691 return 1;
1692}
1693
1694static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1695{
1696 struct object_entry *child = me->delta_child;
1697 unsigned int m = n;
1698 while (child) {
1699 unsigned int c = check_delta_limit(child, n + 1);
1700 if (m < c)
1701 m = c;
1702 child = child->delta_sibling;
1703 }
1704 return m;
1705}
1706
1707static unsigned long free_unpacked(struct unpacked *n)
1708{
1709 unsigned long freed_mem = sizeof_delta_index(n->index);
1710 free_delta_index(n->index);
1711 n->index = NULL;
1712 if (n->data) {
1713 freed_mem += n->entry->size;
1714 free(n->data);
1715 n->data = NULL;
1716 }
1717 n->entry = NULL;
1718 n->depth = 0;
1719 return freed_mem;
1720}
1721
1722static void find_deltas(struct object_entry **list, unsigned *list_size,
1723 int window, int depth, unsigned *processed)
1724{
1725 uint32_t i, idx = 0, count = 0;
1726 struct unpacked *array;
1727 unsigned long mem_usage = 0;
1728
1729 array = xcalloc(window, sizeof(struct unpacked));
1730
1731 for (;;) {
1732 struct object_entry *entry;
1733 struct unpacked *n = array + idx;
1734 int j, max_depth, best_base = -1;
1735
1736 progress_lock();
1737 if (!*list_size) {
1738 progress_unlock();
1739 break;
1740 }
1741 entry = *list++;
1742 (*list_size)--;
1743 if (!entry->preferred_base) {
1744 (*processed)++;
1745 display_progress(progress_state, *processed);
1746 }
1747 progress_unlock();
1748
1749 mem_usage -= free_unpacked(n);
1750 n->entry = entry;
1751
1752 while (window_memory_limit &&
1753 mem_usage > window_memory_limit &&
1754 count > 1) {
1755 uint32_t tail = (idx + window - count) % window;
1756 mem_usage -= free_unpacked(array + tail);
1757 count--;
1758 }
1759
1760 /* We do not compute delta to *create* objects we are not
1761 * going to pack.
1762 */
1763 if (entry->preferred_base)
1764 goto next;
1765
1766 /*
1767 * If the current object is at pack edge, take the depth the
1768 * objects that depend on the current object into account
1769 * otherwise they would become too deep.
1770 */
1771 max_depth = depth;
1772 if (entry->delta_child) {
1773 max_depth -= check_delta_limit(entry, 0);
1774 if (max_depth <= 0)
1775 goto next;
1776 }
1777
1778 j = window;
1779 while (--j > 0) {
1780 int ret;
1781 uint32_t other_idx = idx + j;
1782 struct unpacked *m;
1783 if (other_idx >= window)
1784 other_idx -= window;
1785 m = array + other_idx;
1786 if (!m->entry)
1787 break;
1788 ret = try_delta(n, m, max_depth, &mem_usage);
1789 if (ret < 0)
1790 break;
1791 else if (ret > 0)
1792 best_base = other_idx;
1793 }
1794
1795 /*
1796 * If we decided to cache the delta data, then it is best
1797 * to compress it right away. First because we have to do
1798 * it anyway, and doing it here while we're threaded will
1799 * save a lot of time in the non threaded write phase,
1800 * as well as allow for caching more deltas within
1801 * the same cache size limit.
1802 * ...
1803 * But only if not writing to stdout, since in that case
1804 * the network is most likely throttling writes anyway,
1805 * and therefore it is best to go to the write phase ASAP
1806 * instead, as we can afford spending more time compressing
1807 * between writes at that moment.
1808 */
1809 if (entry->delta_data && !pack_to_stdout) {
1810 entry->z_delta_size = do_compress(&entry->delta_data,
1811 entry->delta_size);
1812 cache_lock();
1813 delta_cache_size -= entry->delta_size;
1814 delta_cache_size += entry->z_delta_size;
1815 cache_unlock();
1816 }
1817
1818 /* if we made n a delta, and if n is already at max
1819 * depth, leaving it in the window is pointless. we
1820 * should evict it first.
1821 */
1822 if (entry->delta && max_depth <= n->depth)
1823 continue;
1824
1825 /*
1826 * Move the best delta base up in the window, after the
1827 * currently deltified object, to keep it longer. It will
1828 * be the first base object to be attempted next.
1829 */
1830 if (entry->delta) {
1831 struct unpacked swap = array[best_base];
1832 int dist = (window + idx - best_base) % window;
1833 int dst = best_base;
1834 while (dist--) {
1835 int src = (dst + 1) % window;
1836 array[dst] = array[src];
1837 dst = src;
1838 }
1839 array[dst] = swap;
1840 }
1841
1842 next:
1843 idx++;
1844 if (count + 1 < window)
1845 count++;
1846 if (idx >= window)
1847 idx = 0;
1848 }
1849
1850 for (i = 0; i < window; ++i) {
1851 free_delta_index(array[i].index);
1852 free(array[i].data);
1853 }
1854 free(array);
1855}
1856
1857#ifndef NO_PTHREADS
1858
1859static void try_to_free_from_threads(size_t size)
1860{
1861 read_lock();
1862 release_pack_memory(size);
1863 read_unlock();
1864}
1865
1866static try_to_free_t old_try_to_free_routine;
1867
1868/*
1869 * The main thread waits on the condition that (at least) one of the workers
1870 * has stopped working (which is indicated in the .working member of
1871 * struct thread_params).
1872 * When a work thread has completed its work, it sets .working to 0 and
1873 * signals the main thread and waits on the condition that .data_ready
1874 * becomes 1.
1875 */
1876
1877struct thread_params {
1878 pthread_t thread;
1879 struct object_entry **list;
1880 unsigned list_size;
1881 unsigned remaining;
1882 int window;
1883 int depth;
1884 int working;
1885 int data_ready;
1886 pthread_mutex_t mutex;
1887 pthread_cond_t cond;
1888 unsigned *processed;
1889};
1890
1891static pthread_cond_t progress_cond;
1892
1893/*
1894 * Mutex and conditional variable can't be statically-initialized on Windows.
1895 */
1896static void init_threaded_search(void)
1897{
1898 init_recursive_mutex(&read_mutex);
1899 pthread_mutex_init(&cache_mutex, NULL);
1900 pthread_mutex_init(&progress_mutex, NULL);
1901 pthread_cond_init(&progress_cond, NULL);
1902 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1903}
1904
1905static void cleanup_threaded_search(void)
1906{
1907 set_try_to_free_routine(old_try_to_free_routine);
1908 pthread_cond_destroy(&progress_cond);
1909 pthread_mutex_destroy(&read_mutex);
1910 pthread_mutex_destroy(&cache_mutex);
1911 pthread_mutex_destroy(&progress_mutex);
1912}
1913
1914static void *threaded_find_deltas(void *arg)
1915{
1916 struct thread_params *me = arg;
1917
1918 while (me->remaining) {
1919 find_deltas(me->list, &me->remaining,
1920 me->window, me->depth, me->processed);
1921
1922 progress_lock();
1923 me->working = 0;
1924 pthread_cond_signal(&progress_cond);
1925 progress_unlock();
1926
1927 /*
1928 * We must not set ->data_ready before we wait on the
1929 * condition because the main thread may have set it to 1
1930 * before we get here. In order to be sure that new
1931 * work is available if we see 1 in ->data_ready, it
1932 * was initialized to 0 before this thread was spawned
1933 * and we reset it to 0 right away.
1934 */
1935 pthread_mutex_lock(&me->mutex);
1936 while (!me->data_ready)
1937 pthread_cond_wait(&me->cond, &me->mutex);
1938 me->data_ready = 0;
1939 pthread_mutex_unlock(&me->mutex);
1940 }
1941 /* leave ->working 1 so that this doesn't get more work assigned */
1942 return NULL;
1943}
1944
1945static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1946 int window, int depth, unsigned *processed)
1947{
1948 struct thread_params *p;
1949 int i, ret, active_threads = 0;
1950
1951 init_threaded_search();
1952
1953 if (!delta_search_threads) /* --threads=0 means autodetect */
1954 delta_search_threads = online_cpus();
1955 if (delta_search_threads <= 1) {
1956 find_deltas(list, &list_size, window, depth, processed);
1957 cleanup_threaded_search();
1958 return;
1959 }
1960 if (progress > pack_to_stdout)
1961 fprintf(stderr, "Delta compression using up to %d threads.\n",
1962 delta_search_threads);
1963 p = xcalloc(delta_search_threads, sizeof(*p));
1964
1965 /* Partition the work amongst work threads. */
1966 for (i = 0; i < delta_search_threads; i++) {
1967 unsigned sub_size = list_size / (delta_search_threads - i);
1968
1969 /* don't use too small segments or no deltas will be found */
1970 if (sub_size < 2*window && i+1 < delta_search_threads)
1971 sub_size = 0;
1972
1973 p[i].window = window;
1974 p[i].depth = depth;
1975 p[i].processed = processed;
1976 p[i].working = 1;
1977 p[i].data_ready = 0;
1978
1979 /* try to split chunks on "path" boundaries */
1980 while (sub_size && sub_size < list_size &&
1981 list[sub_size]->hash &&
1982 list[sub_size]->hash == list[sub_size-1]->hash)
1983 sub_size++;
1984
1985 p[i].list = list;
1986 p[i].list_size = sub_size;
1987 p[i].remaining = sub_size;
1988
1989 list += sub_size;
1990 list_size -= sub_size;
1991 }
1992
1993 /* Start work threads. */
1994 for (i = 0; i < delta_search_threads; i++) {
1995 if (!p[i].list_size)
1996 continue;
1997 pthread_mutex_init(&p[i].mutex, NULL);
1998 pthread_cond_init(&p[i].cond, NULL);
1999 ret = pthread_create(&p[i].thread, NULL,
2000 threaded_find_deltas, &p[i]);
2001 if (ret)
2002 die("unable to create thread: %s", strerror(ret));
2003 active_threads++;
2004 }
2005
2006 /*
2007 * Now let's wait for work completion. Each time a thread is done
2008 * with its work, we steal half of the remaining work from the
2009 * thread with the largest number of unprocessed objects and give
2010 * it to that newly idle thread. This ensure good load balancing
2011 * until the remaining object list segments are simply too short
2012 * to be worth splitting anymore.
2013 */
2014 while (active_threads) {
2015 struct thread_params *target = NULL;
2016 struct thread_params *victim = NULL;
2017 unsigned sub_size = 0;
2018
2019 progress_lock();
2020 for (;;) {
2021 for (i = 0; !target && i < delta_search_threads; i++)
2022 if (!p[i].working)
2023 target = &p[i];
2024 if (target)
2025 break;
2026 pthread_cond_wait(&progress_cond, &progress_mutex);
2027 }
2028
2029 for (i = 0; i < delta_search_threads; i++)
2030 if (p[i].remaining > 2*window &&
2031 (!victim || victim->remaining < p[i].remaining))
2032 victim = &p[i];
2033 if (victim) {
2034 sub_size = victim->remaining / 2;
2035 list = victim->list + victim->list_size - sub_size;
2036 while (sub_size && list[0]->hash &&
2037 list[0]->hash == list[-1]->hash) {
2038 list++;
2039 sub_size--;
2040 }
2041 if (!sub_size) {
2042 /*
2043 * It is possible for some "paths" to have
2044 * so many objects that no hash boundary
2045 * might be found. Let's just steal the
2046 * exact half in that case.
2047 */
2048 sub_size = victim->remaining / 2;
2049 list -= sub_size;
2050 }
2051 target->list = list;
2052 victim->list_size -= sub_size;
2053 victim->remaining -= sub_size;
2054 }
2055 target->list_size = sub_size;
2056 target->remaining = sub_size;
2057 target->working = 1;
2058 progress_unlock();
2059
2060 pthread_mutex_lock(&target->mutex);
2061 target->data_ready = 1;
2062 pthread_cond_signal(&target->cond);
2063 pthread_mutex_unlock(&target->mutex);
2064
2065 if (!sub_size) {
2066 pthread_join(target->thread, NULL);
2067 pthread_cond_destroy(&target->cond);
2068 pthread_mutex_destroy(&target->mutex);
2069 active_threads--;
2070 }
2071 }
2072 cleanup_threaded_search();
2073 free(p);
2074}
2075
2076#else
2077#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2078#endif
2079
2080static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
2081{
2082 unsigned char peeled[20];
2083
2084 if (starts_with(path, "refs/tags/") && /* is a tag? */
2085 !peel_ref(path, peeled) && /* peelable? */
2086 packlist_find(&to_pack, peeled, NULL)) /* object packed? */
2087 add_object_entry(sha1, OBJ_TAG, NULL, 0);
2088 return 0;
2089}
2090
2091static void prepare_pack(int window, int depth)
2092{
2093 struct object_entry **delta_list;
2094 uint32_t i, nr_deltas;
2095 unsigned n;
2096
2097 get_object_details();
2098
2099 /*
2100 * If we're locally repacking then we need to be doubly careful
2101 * from now on in order to make sure no stealth corruption gets
2102 * propagated to the new pack. Clients receiving streamed packs
2103 * should validate everything they get anyway so no need to incur
2104 * the additional cost here in that case.
2105 */
2106 if (!pack_to_stdout)
2107 do_check_packed_object_crc = 1;
2108
2109 if (!to_pack.nr_objects || !window || !depth)
2110 return;
2111
2112 delta_list = xmalloc(to_pack.nr_objects * sizeof(*delta_list));
2113 nr_deltas = n = 0;
2114
2115 for (i = 0; i < to_pack.nr_objects; i++) {
2116 struct object_entry *entry = to_pack.objects + i;
2117
2118 if (entry->delta)
2119 /* This happens if we decided to reuse existing
2120 * delta from a pack. "reuse_delta &&" is implied.
2121 */
2122 continue;
2123
2124 if (entry->size < 50)
2125 continue;
2126
2127 if (entry->no_try_delta)
2128 continue;
2129
2130 if (!entry->preferred_base) {
2131 nr_deltas++;
2132 if (entry->type < 0)
2133 die("unable to get type of object %s",
2134 sha1_to_hex(entry->idx.sha1));
2135 } else {
2136 if (entry->type < 0) {
2137 /*
2138 * This object is not found, but we
2139 * don't have to include it anyway.
2140 */
2141 continue;
2142 }
2143 }
2144
2145 delta_list[n++] = entry;
2146 }
2147
2148 if (nr_deltas && n > 1) {
2149 unsigned nr_done = 0;
2150 if (progress)
2151 progress_state = start_progress(_("Compressing objects"),
2152 nr_deltas);
2153 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2154 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2155 stop_progress(&progress_state);
2156 if (nr_done != nr_deltas)
2157 die("inconsistency with delta count");
2158 }
2159 free(delta_list);
2160}
2161
2162static int git_pack_config(const char *k, const char *v, void *cb)
2163{
2164 if (!strcmp(k, "pack.window")) {
2165 window = git_config_int(k, v);
2166 return 0;
2167 }
2168 if (!strcmp(k, "pack.windowmemory")) {
2169 window_memory_limit = git_config_ulong(k, v);
2170 return 0;
2171 }
2172 if (!strcmp(k, "pack.depth")) {
2173 depth = git_config_int(k, v);
2174 return 0;
2175 }
2176 if (!strcmp(k, "pack.compression")) {
2177 int level = git_config_int(k, v);
2178 if (level == -1)
2179 level = Z_DEFAULT_COMPRESSION;
2180 else if (level < 0 || level > Z_BEST_COMPRESSION)
2181 die("bad pack compression level %d", level);
2182 pack_compression_level = level;
2183 pack_compression_seen = 1;
2184 return 0;
2185 }
2186 if (!strcmp(k, "pack.deltacachesize")) {
2187 max_delta_cache_size = git_config_int(k, v);
2188 return 0;
2189 }
2190 if (!strcmp(k, "pack.deltacachelimit")) {
2191 cache_max_small_delta_size = git_config_int(k, v);
2192 return 0;
2193 }
2194 if (!strcmp(k, "pack.writebitmaps")) {
2195 write_bitmap_index = git_config_bool(k, v);
2196 return 0;
2197 }
2198 if (!strcmp(k, "pack.writebitmaphashcache")) {
2199 if (git_config_bool(k, v))
2200 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2201 else
2202 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2203 }
2204 if (!strcmp(k, "pack.usebitmaps")) {
2205 use_bitmap_index = git_config_bool(k, v);
2206 return 0;
2207 }
2208 if (!strcmp(k, "pack.threads")) {
2209 delta_search_threads = git_config_int(k, v);
2210 if (delta_search_threads < 0)
2211 die("invalid number of threads specified (%d)",
2212 delta_search_threads);
2213#ifdef NO_PTHREADS
2214 if (delta_search_threads != 1)
2215 warning("no threads support, ignoring %s", k);
2216#endif
2217 return 0;
2218 }
2219 if (!strcmp(k, "pack.indexversion")) {
2220 pack_idx_opts.version = git_config_int(k, v);
2221 if (pack_idx_opts.version > 2)
2222 die("bad pack.indexversion=%"PRIu32,
2223 pack_idx_opts.version);
2224 return 0;
2225 }
2226 return git_default_config(k, v, cb);
2227}
2228
2229static void read_object_list_from_stdin(void)
2230{
2231 char line[40 + 1 + PATH_MAX + 2];
2232 unsigned char sha1[20];
2233
2234 for (;;) {
2235 if (!fgets(line, sizeof(line), stdin)) {
2236 if (feof(stdin))
2237 break;
2238 if (!ferror(stdin))
2239 die("fgets returned NULL, not EOF, not error!");
2240 if (errno != EINTR)
2241 die_errno("fgets");
2242 clearerr(stdin);
2243 continue;
2244 }
2245 if (line[0] == '-') {
2246 if (get_sha1_hex(line+1, sha1))
2247 die("expected edge sha1, got garbage:\n %s",
2248 line);
2249 add_preferred_base(sha1);
2250 continue;
2251 }
2252 if (get_sha1_hex(line, sha1))
2253 die("expected sha1, got garbage:\n %s", line);
2254
2255 add_preferred_base_object(line+41);
2256 add_object_entry(sha1, 0, line+41, 0);
2257 }
2258}
2259
2260#define OBJECT_ADDED (1u<<20)
2261
2262static void show_commit(struct commit *commit, void *data)
2263{
2264 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2265 commit->object.flags |= OBJECT_ADDED;
2266
2267 if (write_bitmap_index)
2268 index_commit_for_bitmap(commit);
2269}
2270
2271static void show_object(struct object *obj,
2272 const struct name_path *path, const char *last,
2273 void *data)
2274{
2275 char *name = path_name(path, last);
2276
2277 add_preferred_base_object(name);
2278 add_object_entry(obj->sha1, obj->type, name, 0);
2279 obj->flags |= OBJECT_ADDED;
2280
2281 /*
2282 * We will have generated the hash from the name,
2283 * but not saved a pointer to it - we can free it
2284 */
2285 free((char *)name);
2286}
2287
2288static void show_edge(struct commit *commit)
2289{
2290 add_preferred_base(commit->object.sha1);
2291}
2292
2293struct in_pack_object {
2294 off_t offset;
2295 struct object *object;
2296};
2297
2298struct in_pack {
2299 int alloc;
2300 int nr;
2301 struct in_pack_object *array;
2302};
2303
2304static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2305{
2306 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2307 in_pack->array[in_pack->nr].object = object;
2308 in_pack->nr++;
2309}
2310
2311/*
2312 * Compare the objects in the offset order, in order to emulate the
2313 * "git rev-list --objects" output that produced the pack originally.
2314 */
2315static int ofscmp(const void *a_, const void *b_)
2316{
2317 struct in_pack_object *a = (struct in_pack_object *)a_;
2318 struct in_pack_object *b = (struct in_pack_object *)b_;
2319
2320 if (a->offset < b->offset)
2321 return -1;
2322 else if (a->offset > b->offset)
2323 return 1;
2324 else
2325 return hashcmp(a->object->sha1, b->object->sha1);
2326}
2327
2328static void add_objects_in_unpacked_packs(struct rev_info *revs)
2329{
2330 struct packed_git *p;
2331 struct in_pack in_pack;
2332 uint32_t i;
2333
2334 memset(&in_pack, 0, sizeof(in_pack));
2335
2336 for (p = packed_git; p; p = p->next) {
2337 const unsigned char *sha1;
2338 struct object *o;
2339
2340 if (!p->pack_local || p->pack_keep)
2341 continue;
2342 if (open_pack_index(p))
2343 die("cannot open pack index");
2344
2345 ALLOC_GROW(in_pack.array,
2346 in_pack.nr + p->num_objects,
2347 in_pack.alloc);
2348
2349 for (i = 0; i < p->num_objects; i++) {
2350 sha1 = nth_packed_object_sha1(p, i);
2351 o = lookup_unknown_object(sha1);
2352 if (!(o->flags & OBJECT_ADDED))
2353 mark_in_pack_object(o, p, &in_pack);
2354 o->flags |= OBJECT_ADDED;
2355 }
2356 }
2357
2358 if (in_pack.nr) {
2359 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2360 ofscmp);
2361 for (i = 0; i < in_pack.nr; i++) {
2362 struct object *o = in_pack.array[i].object;
2363 add_object_entry(o->sha1, o->type, "", 0);
2364 }
2365 }
2366 free(in_pack.array);
2367}
2368
2369static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2370{
2371 static struct packed_git *last_found = (void *)1;
2372 struct packed_git *p;
2373
2374 p = (last_found != (void *)1) ? last_found : packed_git;
2375
2376 while (p) {
2377 if ((!p->pack_local || p->pack_keep) &&
2378 find_pack_entry_one(sha1, p)) {
2379 last_found = p;
2380 return 1;
2381 }
2382 if (p == last_found)
2383 p = packed_git;
2384 else
2385 p = p->next;
2386 if (p == last_found)
2387 p = p->next;
2388 }
2389 return 0;
2390}
2391
2392static void loosen_unused_packed_objects(struct rev_info *revs)
2393{
2394 struct packed_git *p;
2395 uint32_t i;
2396 const unsigned char *sha1;
2397
2398 for (p = packed_git; p; p = p->next) {
2399 if (!p->pack_local || p->pack_keep)
2400 continue;
2401
2402 if (unpack_unreachable_expiration &&
2403 p->mtime < unpack_unreachable_expiration)
2404 continue;
2405
2406 if (open_pack_index(p))
2407 die("cannot open pack index");
2408
2409 for (i = 0; i < p->num_objects; i++) {
2410 sha1 = nth_packed_object_sha1(p, i);
2411 if (!packlist_find(&to_pack, sha1, NULL) &&
2412 !has_sha1_pack_kept_or_nonlocal(sha1))
2413 if (force_object_loose(sha1, p->mtime))
2414 die("unable to force loose object");
2415 }
2416 }
2417}
2418
2419static int get_object_list_from_bitmap(struct rev_info *revs)
2420{
2421 if (prepare_bitmap_walk(revs) < 0)
2422 return -1;
2423
2424 if (!reuse_partial_packfile_from_bitmap(
2425 &reuse_packfile,
2426 &reuse_packfile_objects,
2427 &reuse_packfile_offset)) {
2428 assert(reuse_packfile_objects);
2429 nr_result += reuse_packfile_objects;
2430
2431 if (progress) {
2432 fprintf(stderr, "Reusing existing pack: %d, done.\n",
2433 reuse_packfile_objects);
2434 fflush(stderr);
2435 }
2436 }
2437
2438 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2439 return 0;
2440}
2441
2442static void get_object_list(int ac, const char **av)
2443{
2444 struct rev_info revs;
2445 char line[1000];
2446 int flags = 0;
2447
2448 init_revisions(&revs, NULL);
2449 save_commit_buffer = 0;
2450 setup_revisions(ac, av, &revs, NULL);
2451
2452 /* make sure shallows are read */
2453 is_repository_shallow();
2454
2455 while (fgets(line, sizeof(line), stdin) != NULL) {
2456 int len = strlen(line);
2457 if (len && line[len - 1] == '\n')
2458 line[--len] = 0;
2459 if (!len)
2460 break;
2461 if (*line == '-') {
2462 if (!strcmp(line, "--not")) {
2463 flags ^= UNINTERESTING;
2464 write_bitmap_index = 0;
2465 continue;
2466 }
2467 if (starts_with(line, "--shallow ")) {
2468 unsigned char sha1[20];
2469 if (get_sha1_hex(line + 10, sha1))
2470 die("not an SHA-1 '%s'", line + 10);
2471 register_shallow(sha1);
2472 continue;
2473 }
2474 die("not a rev '%s'", line);
2475 }
2476 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2477 die("bad revision '%s'", line);
2478 }
2479
2480 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2481 return;
2482
2483 if (prepare_revision_walk(&revs))
2484 die("revision walk setup failed");
2485 mark_edges_uninteresting(&revs, show_edge);
2486 traverse_commit_list(&revs, show_commit, show_object, NULL);
2487
2488 if (keep_unreachable)
2489 add_objects_in_unpacked_packs(&revs);
2490 if (unpack_unreachable)
2491 loosen_unused_packed_objects(&revs);
2492}
2493
2494static int option_parse_index_version(const struct option *opt,
2495 const char *arg, int unset)
2496{
2497 char *c;
2498 const char *val = arg;
2499 pack_idx_opts.version = strtoul(val, &c, 10);
2500 if (pack_idx_opts.version > 2)
2501 die(_("unsupported index version %s"), val);
2502 if (*c == ',' && c[1])
2503 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2504 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2505 die(_("bad index version '%s'"), val);
2506 return 0;
2507}
2508
2509static int option_parse_unpack_unreachable(const struct option *opt,
2510 const char *arg, int unset)
2511{
2512 if (unset) {
2513 unpack_unreachable = 0;
2514 unpack_unreachable_expiration = 0;
2515 }
2516 else {
2517 unpack_unreachable = 1;
2518 if (arg)
2519 unpack_unreachable_expiration = approxidate(arg);
2520 }
2521 return 0;
2522}
2523
2524static int option_parse_ulong(const struct option *opt,
2525 const char *arg, int unset)
2526{
2527 if (unset)
2528 die(_("option %s does not accept negative form"),
2529 opt->long_name);
2530
2531 if (!git_parse_ulong(arg, opt->value))
2532 die(_("unable to parse value '%s' for option %s"),
2533 arg, opt->long_name);
2534 return 0;
2535}
2536
2537#define OPT_ULONG(s, l, v, h) \
2538 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2539 PARSE_OPT_NONEG, option_parse_ulong }
2540
2541int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2542{
2543 int use_internal_rev_list = 0;
2544 int thin = 0;
2545 int all_progress_implied = 0;
2546 const char *rp_av[6];
2547 int rp_ac = 0;
2548 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2549 struct option pack_objects_options[] = {
2550 OPT_SET_INT('q', "quiet", &progress,
2551 N_("do not show progress meter"), 0),
2552 OPT_SET_INT(0, "progress", &progress,
2553 N_("show progress meter"), 1),
2554 OPT_SET_INT(0, "all-progress", &progress,
2555 N_("show progress meter during object writing phase"), 2),
2556 OPT_BOOL(0, "all-progress-implied",
2557 &all_progress_implied,
2558 N_("similar to --all-progress when progress meter is shown")),
2559 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2560 N_("write the pack index file in the specified idx format version"),
2561 0, option_parse_index_version },
2562 OPT_ULONG(0, "max-pack-size", &pack_size_limit,
2563 N_("maximum size of each output pack file")),
2564 OPT_BOOL(0, "local", &local,
2565 N_("ignore borrowed objects from alternate object store")),
2566 OPT_BOOL(0, "incremental", &incremental,
2567 N_("ignore packed objects")),
2568 OPT_INTEGER(0, "window", &window,
2569 N_("limit pack window by objects")),
2570 OPT_ULONG(0, "window-memory", &window_memory_limit,
2571 N_("limit pack window by memory in addition to object limit")),
2572 OPT_INTEGER(0, "depth", &depth,
2573 N_("maximum length of delta chain allowed in the resulting pack")),
2574 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2575 N_("reuse existing deltas")),
2576 OPT_BOOL(0, "reuse-object", &reuse_object,
2577 N_("reuse existing objects")),
2578 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2579 N_("use OFS_DELTA objects")),
2580 OPT_INTEGER(0, "threads", &delta_search_threads,
2581 N_("use threads when searching for best delta matches")),
2582 OPT_BOOL(0, "non-empty", &non_empty,
2583 N_("do not create an empty pack output")),
2584 OPT_BOOL(0, "revs", &use_internal_rev_list,
2585 N_("read revision arguments from standard input")),
2586 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2587 N_("limit the objects to those that are not yet packed"),
2588 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2589 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2590 N_("include objects reachable from any reference"),
2591 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2592 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2593 N_("include objects referred by reflog entries"),
2594 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2595 OPT_BOOL(0, "stdout", &pack_to_stdout,
2596 N_("output pack to stdout")),
2597 OPT_BOOL(0, "include-tag", &include_tag,
2598 N_("include tag objects that refer to objects to be packed")),
2599 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2600 N_("keep unreachable objects")),
2601 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2602 N_("unpack unreachable objects newer than <time>"),
2603 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2604 OPT_BOOL(0, "thin", &thin,
2605 N_("create thin packs")),
2606 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2607 N_("ignore packs that have companion .keep file")),
2608 OPT_INTEGER(0, "compression", &pack_compression_level,
2609 N_("pack compression level")),
2610 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2611 N_("do not hide commits by grafts"), 0),
2612 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2613 N_("use a bitmap index if available to speed up counting objects")),
2614 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2615 N_("write a bitmap index together with the pack index")),
2616 OPT_END(),
2617 };
2618
2619 check_replace_refs = 0;
2620
2621 reset_pack_idx_option(&pack_idx_opts);
2622 git_config(git_pack_config, NULL);
2623 if (!pack_compression_seen && core_compression_seen)
2624 pack_compression_level = core_compression_level;
2625
2626 progress = isatty(2);
2627 argc = parse_options(argc, argv, prefix, pack_objects_options,
2628 pack_usage, 0);
2629
2630 if (argc) {
2631 base_name = argv[0];
2632 argc--;
2633 }
2634 if (pack_to_stdout != !base_name || argc)
2635 usage_with_options(pack_usage, pack_objects_options);
2636
2637 rp_av[rp_ac++] = "pack-objects";
2638 if (thin) {
2639 use_internal_rev_list = 1;
2640 rp_av[rp_ac++] = "--objects-edge";
2641 } else
2642 rp_av[rp_ac++] = "--objects";
2643
2644 if (rev_list_all) {
2645 use_internal_rev_list = 1;
2646 rp_av[rp_ac++] = "--all";
2647 }
2648 if (rev_list_reflog) {
2649 use_internal_rev_list = 1;
2650 rp_av[rp_ac++] = "--reflog";
2651 }
2652 if (rev_list_unpacked) {
2653 use_internal_rev_list = 1;
2654 rp_av[rp_ac++] = "--unpacked";
2655 }
2656
2657 if (!reuse_object)
2658 reuse_delta = 0;
2659 if (pack_compression_level == -1)
2660 pack_compression_level = Z_DEFAULT_COMPRESSION;
2661 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2662 die("bad pack compression level %d", pack_compression_level);
2663#ifdef NO_PTHREADS
2664 if (delta_search_threads != 1)
2665 warning("no threads support, ignoring --threads");
2666#endif
2667 if (!pack_to_stdout && !pack_size_limit)
2668 pack_size_limit = pack_size_limit_cfg;
2669 if (pack_to_stdout && pack_size_limit)
2670 die("--max-pack-size cannot be used to build a pack for transfer.");
2671 if (pack_size_limit && pack_size_limit < 1024*1024) {
2672 warning("minimum pack size limit is 1 MiB");
2673 pack_size_limit = 1024*1024;
2674 }
2675
2676 if (!pack_to_stdout && thin)
2677 die("--thin cannot be used to build an indexable pack.");
2678
2679 if (keep_unreachable && unpack_unreachable)
2680 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2681
2682 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2683 use_bitmap_index = 0;
2684
2685 if (pack_to_stdout || !rev_list_all)
2686 write_bitmap_index = 0;
2687
2688 if (progress && all_progress_implied)
2689 progress = 2;
2690
2691 prepare_packed_git();
2692
2693 if (progress)
2694 progress_state = start_progress(_("Counting objects"), 0);
2695 if (!use_internal_rev_list)
2696 read_object_list_from_stdin();
2697 else {
2698 rp_av[rp_ac] = NULL;
2699 get_object_list(rp_ac, rp_av);
2700 }
2701 cleanup_preferred_base();
2702 if (include_tag && nr_result)
2703 for_each_ref(add_ref_tag, NULL);
2704 stop_progress(&progress_state);
2705
2706 if (non_empty && !nr_result)
2707 return 0;
2708 if (nr_result)
2709 prepare_pack(window, depth);
2710 write_pack_file();
2711 if (progress)
2712 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2713 " reused %"PRIu32" (delta %"PRIu32")\n",
2714 written, written_delta, reused, reused_delta);
2715 return 0;
2716}