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 if (!offset)
773 die_errno("unable to write pack header");
774
775 if (reuse_packfile) {
776 off_t packfile_size;
777 assert(pack_to_stdout);
778
779 packfile_size = write_reused_pack(f);
780 offset += packfile_size;
781 }
782
783 nr_written = 0;
784 for (; i < to_pack.nr_objects; i++) {
785 struct object_entry *e = write_order[i];
786 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
787 break;
788 display_progress(progress_state, written);
789 }
790
791 /*
792 * Did we write the wrong # entries in the header?
793 * If so, rewrite it like in fast-import
794 */
795 if (pack_to_stdout) {
796 sha1close(f, sha1, CSUM_CLOSE);
797 } else if (nr_written == nr_remaining) {
798 sha1close(f, sha1, CSUM_FSYNC);
799 } else {
800 int fd = sha1close(f, sha1, 0);
801 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
802 nr_written, sha1, offset);
803 close(fd);
804 }
805
806 if (!pack_to_stdout) {
807 struct stat st;
808 char tmpname[PATH_MAX];
809
810 /*
811 * Packs are runtime accessed in their mtime
812 * order since newer packs are more likely to contain
813 * younger objects. So if we are creating multiple
814 * packs then we should modify the mtime of later ones
815 * to preserve this property.
816 */
817 if (stat(pack_tmp_name, &st) < 0) {
818 warning("failed to stat %s: %s",
819 pack_tmp_name, strerror(errno));
820 } else if (!last_mtime) {
821 last_mtime = st.st_mtime;
822 } else {
823 struct utimbuf utb;
824 utb.actime = st.st_atime;
825 utb.modtime = --last_mtime;
826 if (utime(pack_tmp_name, &utb) < 0)
827 warning("failed utime() on %s: %s",
828 tmpname, strerror(errno));
829 }
830
831 /* Enough space for "-<sha-1>.pack"? */
832 if (sizeof(tmpname) <= strlen(base_name) + 50)
833 die("pack base name '%s' too long", base_name);
834 snprintf(tmpname, sizeof(tmpname), "%s-", base_name);
835
836 if (write_bitmap_index) {
837 bitmap_writer_set_checksum(sha1);
838 bitmap_writer_build_type_index(written_list, nr_written);
839 }
840
841 finish_tmp_packfile(tmpname, pack_tmp_name,
842 written_list, nr_written,
843 &pack_idx_opts, sha1);
844
845 if (write_bitmap_index) {
846 char *end_of_name_prefix = strrchr(tmpname, 0);
847 sprintf(end_of_name_prefix, "%s.bitmap", sha1_to_hex(sha1));
848
849 stop_progress(&progress_state);
850
851 bitmap_writer_show_progress(progress);
852 bitmap_writer_reuse_bitmaps(&to_pack);
853 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
854 bitmap_writer_build(&to_pack);
855 bitmap_writer_finish(written_list, nr_written,
856 tmpname, write_bitmap_options);
857 write_bitmap_index = 0;
858 }
859
860 free(pack_tmp_name);
861 puts(sha1_to_hex(sha1));
862 }
863
864 /* mark written objects as written to previous pack */
865 for (j = 0; j < nr_written; j++) {
866 written_list[j]->offset = (off_t)-1;
867 }
868 nr_remaining -= nr_written;
869 } while (nr_remaining && i < to_pack.nr_objects);
870
871 free(written_list);
872 free(write_order);
873 stop_progress(&progress_state);
874 if (written != nr_result)
875 die("wrote %"PRIu32" objects while expecting %"PRIu32,
876 written, nr_result);
877}
878
879static void setup_delta_attr_check(struct git_attr_check *check)
880{
881 static struct git_attr *attr_delta;
882
883 if (!attr_delta)
884 attr_delta = git_attr("delta");
885
886 check[0].attr = attr_delta;
887}
888
889static int no_try_delta(const char *path)
890{
891 struct git_attr_check check[1];
892
893 setup_delta_attr_check(check);
894 if (git_check_attr(path, ARRAY_SIZE(check), check))
895 return 0;
896 if (ATTR_FALSE(check->value))
897 return 1;
898 return 0;
899}
900
901/*
902 * When adding an object, check whether we have already added it
903 * to our packing list. If so, we can skip. However, if we are
904 * being asked to excludei t, but the previous mention was to include
905 * it, make sure to adjust its flags and tweak our numbers accordingly.
906 *
907 * As an optimization, we pass out the index position where we would have
908 * found the item, since that saves us from having to look it up again a
909 * few lines later when we want to add the new entry.
910 */
911static int have_duplicate_entry(const unsigned char *sha1,
912 int exclude,
913 uint32_t *index_pos)
914{
915 struct object_entry *entry;
916
917 entry = packlist_find(&to_pack, sha1, index_pos);
918 if (!entry)
919 return 0;
920
921 if (exclude) {
922 if (!entry->preferred_base)
923 nr_result--;
924 entry->preferred_base = 1;
925 }
926
927 return 1;
928}
929
930/*
931 * Check whether we want the object in the pack (e.g., we do not want
932 * objects found in non-local stores if the "--local" option was used).
933 *
934 * As a side effect of this check, we will find the packed version of this
935 * object, if any. We therefore pass out the pack information to avoid having
936 * to look it up again later.
937 */
938static int want_object_in_pack(const unsigned char *sha1,
939 int exclude,
940 struct packed_git **found_pack,
941 off_t *found_offset)
942{
943 struct packed_git *p;
944
945 if (!exclude && local && has_loose_object_nonlocal(sha1))
946 return 0;
947
948 *found_pack = NULL;
949 *found_offset = 0;
950
951 for (p = packed_git; p; p = p->next) {
952 off_t offset = find_pack_entry_one(sha1, p);
953 if (offset) {
954 if (!*found_pack) {
955 if (!is_pack_valid(p)) {
956 warning("packfile %s cannot be accessed", p->pack_name);
957 continue;
958 }
959 *found_offset = offset;
960 *found_pack = p;
961 }
962 if (exclude)
963 return 1;
964 if (incremental)
965 return 0;
966 if (local && !p->pack_local)
967 return 0;
968 if (ignore_packed_keep && p->pack_local && p->pack_keep)
969 return 0;
970 }
971 }
972
973 return 1;
974}
975
976static void create_object_entry(const unsigned char *sha1,
977 enum object_type type,
978 uint32_t hash,
979 int exclude,
980 int no_try_delta,
981 uint32_t index_pos,
982 struct packed_git *found_pack,
983 off_t found_offset)
984{
985 struct object_entry *entry;
986
987 entry = packlist_alloc(&to_pack, sha1, index_pos);
988 entry->hash = hash;
989 if (type)
990 entry->type = type;
991 if (exclude)
992 entry->preferred_base = 1;
993 else
994 nr_result++;
995 if (found_pack) {
996 entry->in_pack = found_pack;
997 entry->in_pack_offset = found_offset;
998 }
999
1000 entry->no_try_delta = no_try_delta;
1001}
1002
1003static const char no_closure_warning[] = N_(
1004"disabling bitmap writing, as some objects are not being packed"
1005);
1006
1007static int add_object_entry(const unsigned char *sha1, enum object_type type,
1008 const char *name, int exclude)
1009{
1010 struct packed_git *found_pack;
1011 off_t found_offset;
1012 uint32_t index_pos;
1013
1014 if (have_duplicate_entry(sha1, exclude, &index_pos))
1015 return 0;
1016
1017 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1018 /* The pack is missing an object, so it will not have closure */
1019 if (write_bitmap_index) {
1020 warning(_(no_closure_warning));
1021 write_bitmap_index = 0;
1022 }
1023 return 0;
1024 }
1025
1026 create_object_entry(sha1, type, pack_name_hash(name),
1027 exclude, name && no_try_delta(name),
1028 index_pos, found_pack, found_offset);
1029
1030 display_progress(progress_state, to_pack.nr_objects);
1031 return 1;
1032}
1033
1034static int add_object_entry_from_bitmap(const unsigned char *sha1,
1035 enum object_type type,
1036 int flags, uint32_t name_hash,
1037 struct packed_git *pack, off_t offset)
1038{
1039 uint32_t index_pos;
1040
1041 if (have_duplicate_entry(sha1, 0, &index_pos))
1042 return 0;
1043
1044 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1045
1046 display_progress(progress_state, to_pack.nr_objects);
1047 return 1;
1048}
1049
1050struct pbase_tree_cache {
1051 unsigned char sha1[20];
1052 int ref;
1053 int temporary;
1054 void *tree_data;
1055 unsigned long tree_size;
1056};
1057
1058static struct pbase_tree_cache *(pbase_tree_cache[256]);
1059static int pbase_tree_cache_ix(const unsigned char *sha1)
1060{
1061 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1062}
1063static int pbase_tree_cache_ix_incr(int ix)
1064{
1065 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1066}
1067
1068static struct pbase_tree {
1069 struct pbase_tree *next;
1070 /* This is a phony "cache" entry; we are not
1071 * going to evict it nor find it through _get()
1072 * mechanism -- this is for the toplevel node that
1073 * would almost always change with any commit.
1074 */
1075 struct pbase_tree_cache pcache;
1076} *pbase_tree;
1077
1078static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1079{
1080 struct pbase_tree_cache *ent, *nent;
1081 void *data;
1082 unsigned long size;
1083 enum object_type type;
1084 int neigh;
1085 int my_ix = pbase_tree_cache_ix(sha1);
1086 int available_ix = -1;
1087
1088 /* pbase-tree-cache acts as a limited hashtable.
1089 * your object will be found at your index or within a few
1090 * slots after that slot if it is cached.
1091 */
1092 for (neigh = 0; neigh < 8; neigh++) {
1093 ent = pbase_tree_cache[my_ix];
1094 if (ent && !hashcmp(ent->sha1, sha1)) {
1095 ent->ref++;
1096 return ent;
1097 }
1098 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1099 ((0 <= available_ix) &&
1100 (!ent && pbase_tree_cache[available_ix])))
1101 available_ix = my_ix;
1102 if (!ent)
1103 break;
1104 my_ix = pbase_tree_cache_ix_incr(my_ix);
1105 }
1106
1107 /* Did not find one. Either we got a bogus request or
1108 * we need to read and perhaps cache.
1109 */
1110 data = read_sha1_file(sha1, &type, &size);
1111 if (!data)
1112 return NULL;
1113 if (type != OBJ_TREE) {
1114 free(data);
1115 return NULL;
1116 }
1117
1118 /* We need to either cache or return a throwaway copy */
1119
1120 if (available_ix < 0)
1121 ent = NULL;
1122 else {
1123 ent = pbase_tree_cache[available_ix];
1124 my_ix = available_ix;
1125 }
1126
1127 if (!ent) {
1128 nent = xmalloc(sizeof(*nent));
1129 nent->temporary = (available_ix < 0);
1130 }
1131 else {
1132 /* evict and reuse */
1133 free(ent->tree_data);
1134 nent = ent;
1135 }
1136 hashcpy(nent->sha1, sha1);
1137 nent->tree_data = data;
1138 nent->tree_size = size;
1139 nent->ref = 1;
1140 if (!nent->temporary)
1141 pbase_tree_cache[my_ix] = nent;
1142 return nent;
1143}
1144
1145static void pbase_tree_put(struct pbase_tree_cache *cache)
1146{
1147 if (!cache->temporary) {
1148 cache->ref--;
1149 return;
1150 }
1151 free(cache->tree_data);
1152 free(cache);
1153}
1154
1155static int name_cmp_len(const char *name)
1156{
1157 int i;
1158 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1159 ;
1160 return i;
1161}
1162
1163static void add_pbase_object(struct tree_desc *tree,
1164 const char *name,
1165 int cmplen,
1166 const char *fullname)
1167{
1168 struct name_entry entry;
1169 int cmp;
1170
1171 while (tree_entry(tree,&entry)) {
1172 if (S_ISGITLINK(entry.mode))
1173 continue;
1174 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1175 memcmp(name, entry.path, cmplen);
1176 if (cmp > 0)
1177 continue;
1178 if (cmp < 0)
1179 return;
1180 if (name[cmplen] != '/') {
1181 add_object_entry(entry.sha1,
1182 object_type(entry.mode),
1183 fullname, 1);
1184 return;
1185 }
1186 if (S_ISDIR(entry.mode)) {
1187 struct tree_desc sub;
1188 struct pbase_tree_cache *tree;
1189 const char *down = name+cmplen+1;
1190 int downlen = name_cmp_len(down);
1191
1192 tree = pbase_tree_get(entry.sha1);
1193 if (!tree)
1194 return;
1195 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1196
1197 add_pbase_object(&sub, down, downlen, fullname);
1198 pbase_tree_put(tree);
1199 }
1200 }
1201}
1202
1203static unsigned *done_pbase_paths;
1204static int done_pbase_paths_num;
1205static int done_pbase_paths_alloc;
1206static int done_pbase_path_pos(unsigned hash)
1207{
1208 int lo = 0;
1209 int hi = done_pbase_paths_num;
1210 while (lo < hi) {
1211 int mi = (hi + lo) / 2;
1212 if (done_pbase_paths[mi] == hash)
1213 return mi;
1214 if (done_pbase_paths[mi] < hash)
1215 hi = mi;
1216 else
1217 lo = mi + 1;
1218 }
1219 return -lo-1;
1220}
1221
1222static int check_pbase_path(unsigned hash)
1223{
1224 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1225 if (0 <= pos)
1226 return 1;
1227 pos = -pos - 1;
1228 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1229 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1230 done_pbase_paths = xrealloc(done_pbase_paths,
1231 done_pbase_paths_alloc *
1232 sizeof(unsigned));
1233 }
1234 done_pbase_paths_num++;
1235 if (pos < done_pbase_paths_num)
1236 memmove(done_pbase_paths + pos + 1,
1237 done_pbase_paths + pos,
1238 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1239 done_pbase_paths[pos] = hash;
1240 return 0;
1241}
1242
1243static void add_preferred_base_object(const char *name)
1244{
1245 struct pbase_tree *it;
1246 int cmplen;
1247 unsigned hash = pack_name_hash(name);
1248
1249 if (!num_preferred_base || check_pbase_path(hash))
1250 return;
1251
1252 cmplen = name_cmp_len(name);
1253 for (it = pbase_tree; it; it = it->next) {
1254 if (cmplen == 0) {
1255 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1256 }
1257 else {
1258 struct tree_desc tree;
1259 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1260 add_pbase_object(&tree, name, cmplen, name);
1261 }
1262 }
1263}
1264
1265static void add_preferred_base(unsigned char *sha1)
1266{
1267 struct pbase_tree *it;
1268 void *data;
1269 unsigned long size;
1270 unsigned char tree_sha1[20];
1271
1272 if (window <= num_preferred_base++)
1273 return;
1274
1275 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1276 if (!data)
1277 return;
1278
1279 for (it = pbase_tree; it; it = it->next) {
1280 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1281 free(data);
1282 return;
1283 }
1284 }
1285
1286 it = xcalloc(1, sizeof(*it));
1287 it->next = pbase_tree;
1288 pbase_tree = it;
1289
1290 hashcpy(it->pcache.sha1, tree_sha1);
1291 it->pcache.tree_data = data;
1292 it->pcache.tree_size = size;
1293}
1294
1295static void cleanup_preferred_base(void)
1296{
1297 struct pbase_tree *it;
1298 unsigned i;
1299
1300 it = pbase_tree;
1301 pbase_tree = NULL;
1302 while (it) {
1303 struct pbase_tree *this = it;
1304 it = this->next;
1305 free(this->pcache.tree_data);
1306 free(this);
1307 }
1308
1309 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1310 if (!pbase_tree_cache[i])
1311 continue;
1312 free(pbase_tree_cache[i]->tree_data);
1313 free(pbase_tree_cache[i]);
1314 pbase_tree_cache[i] = NULL;
1315 }
1316
1317 free(done_pbase_paths);
1318 done_pbase_paths = NULL;
1319 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1320}
1321
1322static void check_object(struct object_entry *entry)
1323{
1324 if (entry->in_pack) {
1325 struct packed_git *p = entry->in_pack;
1326 struct pack_window *w_curs = NULL;
1327 const unsigned char *base_ref = NULL;
1328 struct object_entry *base_entry;
1329 unsigned long used, used_0;
1330 unsigned long avail;
1331 off_t ofs;
1332 unsigned char *buf, c;
1333
1334 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1335
1336 /*
1337 * We want in_pack_type even if we do not reuse delta
1338 * since non-delta representations could still be reused.
1339 */
1340 used = unpack_object_header_buffer(buf, avail,
1341 &entry->in_pack_type,
1342 &entry->size);
1343 if (used == 0)
1344 goto give_up;
1345
1346 /*
1347 * Determine if this is a delta and if so whether we can
1348 * reuse it or not. Otherwise let's find out as cheaply as
1349 * possible what the actual type and size for this object is.
1350 */
1351 switch (entry->in_pack_type) {
1352 default:
1353 /* Not a delta hence we've already got all we need. */
1354 entry->type = entry->in_pack_type;
1355 entry->in_pack_header_size = used;
1356 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1357 goto give_up;
1358 unuse_pack(&w_curs);
1359 return;
1360 case OBJ_REF_DELTA:
1361 if (reuse_delta && !entry->preferred_base)
1362 base_ref = use_pack(p, &w_curs,
1363 entry->in_pack_offset + used, NULL);
1364 entry->in_pack_header_size = used + 20;
1365 break;
1366 case OBJ_OFS_DELTA:
1367 buf = use_pack(p, &w_curs,
1368 entry->in_pack_offset + used, NULL);
1369 used_0 = 0;
1370 c = buf[used_0++];
1371 ofs = c & 127;
1372 while (c & 128) {
1373 ofs += 1;
1374 if (!ofs || MSB(ofs, 7)) {
1375 error("delta base offset overflow in pack for %s",
1376 sha1_to_hex(entry->idx.sha1));
1377 goto give_up;
1378 }
1379 c = buf[used_0++];
1380 ofs = (ofs << 7) + (c & 127);
1381 }
1382 ofs = entry->in_pack_offset - ofs;
1383 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1384 error("delta base offset out of bound for %s",
1385 sha1_to_hex(entry->idx.sha1));
1386 goto give_up;
1387 }
1388 if (reuse_delta && !entry->preferred_base) {
1389 struct revindex_entry *revidx;
1390 revidx = find_pack_revindex(p, ofs);
1391 if (!revidx)
1392 goto give_up;
1393 base_ref = nth_packed_object_sha1(p, revidx->nr);
1394 }
1395 entry->in_pack_header_size = used + used_0;
1396 break;
1397 }
1398
1399 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1400 /*
1401 * If base_ref was set above that means we wish to
1402 * reuse delta data, and we even found that base
1403 * in the list of objects we want to pack. Goodie!
1404 *
1405 * Depth value does not matter - find_deltas() will
1406 * never consider reused delta as the base object to
1407 * deltify other objects against, in order to avoid
1408 * circular deltas.
1409 */
1410 entry->type = entry->in_pack_type;
1411 entry->delta = base_entry;
1412 entry->delta_size = entry->size;
1413 entry->delta_sibling = base_entry->delta_child;
1414 base_entry->delta_child = entry;
1415 unuse_pack(&w_curs);
1416 return;
1417 }
1418
1419 if (entry->type) {
1420 /*
1421 * This must be a delta and we already know what the
1422 * final object type is. Let's extract the actual
1423 * object size from the delta header.
1424 */
1425 entry->size = get_size_from_delta(p, &w_curs,
1426 entry->in_pack_offset + entry->in_pack_header_size);
1427 if (entry->size == 0)
1428 goto give_up;
1429 unuse_pack(&w_curs);
1430 return;
1431 }
1432
1433 /*
1434 * No choice but to fall back to the recursive delta walk
1435 * with sha1_object_info() to find about the object type
1436 * at this point...
1437 */
1438 give_up:
1439 unuse_pack(&w_curs);
1440 }
1441
1442 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1443 /*
1444 * The error condition is checked in prepare_pack(). This is
1445 * to permit a missing preferred base object to be ignored
1446 * as a preferred base. Doing so can result in a larger
1447 * pack file, but the transfer will still take place.
1448 */
1449}
1450
1451static int pack_offset_sort(const void *_a, const void *_b)
1452{
1453 const struct object_entry *a = *(struct object_entry **)_a;
1454 const struct object_entry *b = *(struct object_entry **)_b;
1455
1456 /* avoid filesystem trashing with loose objects */
1457 if (!a->in_pack && !b->in_pack)
1458 return hashcmp(a->idx.sha1, b->idx.sha1);
1459
1460 if (a->in_pack < b->in_pack)
1461 return -1;
1462 if (a->in_pack > b->in_pack)
1463 return 1;
1464 return a->in_pack_offset < b->in_pack_offset ? -1 :
1465 (a->in_pack_offset > b->in_pack_offset);
1466}
1467
1468static void get_object_details(void)
1469{
1470 uint32_t i;
1471 struct object_entry **sorted_by_offset;
1472
1473 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1474 for (i = 0; i < to_pack.nr_objects; i++)
1475 sorted_by_offset[i] = to_pack.objects + i;
1476 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1477
1478 for (i = 0; i < to_pack.nr_objects; i++) {
1479 struct object_entry *entry = sorted_by_offset[i];
1480 check_object(entry);
1481 if (big_file_threshold < entry->size)
1482 entry->no_try_delta = 1;
1483 }
1484
1485 free(sorted_by_offset);
1486}
1487
1488/*
1489 * We search for deltas in a list sorted by type, by filename hash, and then
1490 * by size, so that we see progressively smaller and smaller files.
1491 * That's because we prefer deltas to be from the bigger file
1492 * to the smaller -- deletes are potentially cheaper, but perhaps
1493 * more importantly, the bigger file is likely the more recent
1494 * one. The deepest deltas are therefore the oldest objects which are
1495 * less susceptible to be accessed often.
1496 */
1497static int type_size_sort(const void *_a, const void *_b)
1498{
1499 const struct object_entry *a = *(struct object_entry **)_a;
1500 const struct object_entry *b = *(struct object_entry **)_b;
1501
1502 if (a->type > b->type)
1503 return -1;
1504 if (a->type < b->type)
1505 return 1;
1506 if (a->hash > b->hash)
1507 return -1;
1508 if (a->hash < b->hash)
1509 return 1;
1510 if (a->preferred_base > b->preferred_base)
1511 return -1;
1512 if (a->preferred_base < b->preferred_base)
1513 return 1;
1514 if (a->size > b->size)
1515 return -1;
1516 if (a->size < b->size)
1517 return 1;
1518 return a < b ? -1 : (a > b); /* newest first */
1519}
1520
1521struct unpacked {
1522 struct object_entry *entry;
1523 void *data;
1524 struct delta_index *index;
1525 unsigned depth;
1526};
1527
1528static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1529 unsigned long delta_size)
1530{
1531 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1532 return 0;
1533
1534 if (delta_size < cache_max_small_delta_size)
1535 return 1;
1536
1537 /* cache delta, if objects are large enough compared to delta size */
1538 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1539 return 1;
1540
1541 return 0;
1542}
1543
1544#ifndef NO_PTHREADS
1545
1546static pthread_mutex_t read_mutex;
1547#define read_lock() pthread_mutex_lock(&read_mutex)
1548#define read_unlock() pthread_mutex_unlock(&read_mutex)
1549
1550static pthread_mutex_t cache_mutex;
1551#define cache_lock() pthread_mutex_lock(&cache_mutex)
1552#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1553
1554static pthread_mutex_t progress_mutex;
1555#define progress_lock() pthread_mutex_lock(&progress_mutex)
1556#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1557
1558#else
1559
1560#define read_lock() (void)0
1561#define read_unlock() (void)0
1562#define cache_lock() (void)0
1563#define cache_unlock() (void)0
1564#define progress_lock() (void)0
1565#define progress_unlock() (void)0
1566
1567#endif
1568
1569static int try_delta(struct unpacked *trg, struct unpacked *src,
1570 unsigned max_depth, unsigned long *mem_usage)
1571{
1572 struct object_entry *trg_entry = trg->entry;
1573 struct object_entry *src_entry = src->entry;
1574 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1575 unsigned ref_depth;
1576 enum object_type type;
1577 void *delta_buf;
1578
1579 /* Don't bother doing diffs between different types */
1580 if (trg_entry->type != src_entry->type)
1581 return -1;
1582
1583 /*
1584 * We do not bother to try a delta that we discarded on an
1585 * earlier try, but only when reusing delta data. Note that
1586 * src_entry that is marked as the preferred_base should always
1587 * be considered, as even if we produce a suboptimal delta against
1588 * it, we will still save the transfer cost, as we already know
1589 * the other side has it and we won't send src_entry at all.
1590 */
1591 if (reuse_delta && trg_entry->in_pack &&
1592 trg_entry->in_pack == src_entry->in_pack &&
1593 !src_entry->preferred_base &&
1594 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1595 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1596 return 0;
1597
1598 /* Let's not bust the allowed depth. */
1599 if (src->depth >= max_depth)
1600 return 0;
1601
1602 /* Now some size filtering heuristics. */
1603 trg_size = trg_entry->size;
1604 if (!trg_entry->delta) {
1605 max_size = trg_size/2 - 20;
1606 ref_depth = 1;
1607 } else {
1608 max_size = trg_entry->delta_size;
1609 ref_depth = trg->depth;
1610 }
1611 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1612 (max_depth - ref_depth + 1);
1613 if (max_size == 0)
1614 return 0;
1615 src_size = src_entry->size;
1616 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1617 if (sizediff >= max_size)
1618 return 0;
1619 if (trg_size < src_size / 32)
1620 return 0;
1621
1622 /* Load data if not already done */
1623 if (!trg->data) {
1624 read_lock();
1625 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1626 read_unlock();
1627 if (!trg->data)
1628 die("object %s cannot be read",
1629 sha1_to_hex(trg_entry->idx.sha1));
1630 if (sz != trg_size)
1631 die("object %s inconsistent object length (%lu vs %lu)",
1632 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1633 *mem_usage += sz;
1634 }
1635 if (!src->data) {
1636 read_lock();
1637 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1638 read_unlock();
1639 if (!src->data) {
1640 if (src_entry->preferred_base) {
1641 static int warned = 0;
1642 if (!warned++)
1643 warning("object %s cannot be read",
1644 sha1_to_hex(src_entry->idx.sha1));
1645 /*
1646 * Those objects are not included in the
1647 * resulting pack. Be resilient and ignore
1648 * them if they can't be read, in case the
1649 * pack could be created nevertheless.
1650 */
1651 return 0;
1652 }
1653 die("object %s cannot be read",
1654 sha1_to_hex(src_entry->idx.sha1));
1655 }
1656 if (sz != src_size)
1657 die("object %s inconsistent object length (%lu vs %lu)",
1658 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1659 *mem_usage += sz;
1660 }
1661 if (!src->index) {
1662 src->index = create_delta_index(src->data, src_size);
1663 if (!src->index) {
1664 static int warned = 0;
1665 if (!warned++)
1666 warning("suboptimal pack - out of memory");
1667 return 0;
1668 }
1669 *mem_usage += sizeof_delta_index(src->index);
1670 }
1671
1672 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1673 if (!delta_buf)
1674 return 0;
1675
1676 if (trg_entry->delta) {
1677 /* Prefer only shallower same-sized deltas. */
1678 if (delta_size == trg_entry->delta_size &&
1679 src->depth + 1 >= trg->depth) {
1680 free(delta_buf);
1681 return 0;
1682 }
1683 }
1684
1685 /*
1686 * Handle memory allocation outside of the cache
1687 * accounting lock. Compiler will optimize the strangeness
1688 * away when NO_PTHREADS is defined.
1689 */
1690 free(trg_entry->delta_data);
1691 cache_lock();
1692 if (trg_entry->delta_data) {
1693 delta_cache_size -= trg_entry->delta_size;
1694 trg_entry->delta_data = NULL;
1695 }
1696 if (delta_cacheable(src_size, trg_size, delta_size)) {
1697 delta_cache_size += delta_size;
1698 cache_unlock();
1699 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1700 } else {
1701 cache_unlock();
1702 free(delta_buf);
1703 }
1704
1705 trg_entry->delta = src_entry;
1706 trg_entry->delta_size = delta_size;
1707 trg->depth = src->depth + 1;
1708
1709 return 1;
1710}
1711
1712static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1713{
1714 struct object_entry *child = me->delta_child;
1715 unsigned int m = n;
1716 while (child) {
1717 unsigned int c = check_delta_limit(child, n + 1);
1718 if (m < c)
1719 m = c;
1720 child = child->delta_sibling;
1721 }
1722 return m;
1723}
1724
1725static unsigned long free_unpacked(struct unpacked *n)
1726{
1727 unsigned long freed_mem = sizeof_delta_index(n->index);
1728 free_delta_index(n->index);
1729 n->index = NULL;
1730 if (n->data) {
1731 freed_mem += n->entry->size;
1732 free(n->data);
1733 n->data = NULL;
1734 }
1735 n->entry = NULL;
1736 n->depth = 0;
1737 return freed_mem;
1738}
1739
1740static void find_deltas(struct object_entry **list, unsigned *list_size,
1741 int window, int depth, unsigned *processed)
1742{
1743 uint32_t i, idx = 0, count = 0;
1744 struct unpacked *array;
1745 unsigned long mem_usage = 0;
1746
1747 array = xcalloc(window, sizeof(struct unpacked));
1748
1749 for (;;) {
1750 struct object_entry *entry;
1751 struct unpacked *n = array + idx;
1752 int j, max_depth, best_base = -1;
1753
1754 progress_lock();
1755 if (!*list_size) {
1756 progress_unlock();
1757 break;
1758 }
1759 entry = *list++;
1760 (*list_size)--;
1761 if (!entry->preferred_base) {
1762 (*processed)++;
1763 display_progress(progress_state, *processed);
1764 }
1765 progress_unlock();
1766
1767 mem_usage -= free_unpacked(n);
1768 n->entry = entry;
1769
1770 while (window_memory_limit &&
1771 mem_usage > window_memory_limit &&
1772 count > 1) {
1773 uint32_t tail = (idx + window - count) % window;
1774 mem_usage -= free_unpacked(array + tail);
1775 count--;
1776 }
1777
1778 /* We do not compute delta to *create* objects we are not
1779 * going to pack.
1780 */
1781 if (entry->preferred_base)
1782 goto next;
1783
1784 /*
1785 * If the current object is at pack edge, take the depth the
1786 * objects that depend on the current object into account
1787 * otherwise they would become too deep.
1788 */
1789 max_depth = depth;
1790 if (entry->delta_child) {
1791 max_depth -= check_delta_limit(entry, 0);
1792 if (max_depth <= 0)
1793 goto next;
1794 }
1795
1796 j = window;
1797 while (--j > 0) {
1798 int ret;
1799 uint32_t other_idx = idx + j;
1800 struct unpacked *m;
1801 if (other_idx >= window)
1802 other_idx -= window;
1803 m = array + other_idx;
1804 if (!m->entry)
1805 break;
1806 ret = try_delta(n, m, max_depth, &mem_usage);
1807 if (ret < 0)
1808 break;
1809 else if (ret > 0)
1810 best_base = other_idx;
1811 }
1812
1813 /*
1814 * If we decided to cache the delta data, then it is best
1815 * to compress it right away. First because we have to do
1816 * it anyway, and doing it here while we're threaded will
1817 * save a lot of time in the non threaded write phase,
1818 * as well as allow for caching more deltas within
1819 * the same cache size limit.
1820 * ...
1821 * But only if not writing to stdout, since in that case
1822 * the network is most likely throttling writes anyway,
1823 * and therefore it is best to go to the write phase ASAP
1824 * instead, as we can afford spending more time compressing
1825 * between writes at that moment.
1826 */
1827 if (entry->delta_data && !pack_to_stdout) {
1828 entry->z_delta_size = do_compress(&entry->delta_data,
1829 entry->delta_size);
1830 cache_lock();
1831 delta_cache_size -= entry->delta_size;
1832 delta_cache_size += entry->z_delta_size;
1833 cache_unlock();
1834 }
1835
1836 /* if we made n a delta, and if n is already at max
1837 * depth, leaving it in the window is pointless. we
1838 * should evict it first.
1839 */
1840 if (entry->delta && max_depth <= n->depth)
1841 continue;
1842
1843 /*
1844 * Move the best delta base up in the window, after the
1845 * currently deltified object, to keep it longer. It will
1846 * be the first base object to be attempted next.
1847 */
1848 if (entry->delta) {
1849 struct unpacked swap = array[best_base];
1850 int dist = (window + idx - best_base) % window;
1851 int dst = best_base;
1852 while (dist--) {
1853 int src = (dst + 1) % window;
1854 array[dst] = array[src];
1855 dst = src;
1856 }
1857 array[dst] = swap;
1858 }
1859
1860 next:
1861 idx++;
1862 if (count + 1 < window)
1863 count++;
1864 if (idx >= window)
1865 idx = 0;
1866 }
1867
1868 for (i = 0; i < window; ++i) {
1869 free_delta_index(array[i].index);
1870 free(array[i].data);
1871 }
1872 free(array);
1873}
1874
1875#ifndef NO_PTHREADS
1876
1877static void try_to_free_from_threads(size_t size)
1878{
1879 read_lock();
1880 release_pack_memory(size);
1881 read_unlock();
1882}
1883
1884static try_to_free_t old_try_to_free_routine;
1885
1886/*
1887 * The main thread waits on the condition that (at least) one of the workers
1888 * has stopped working (which is indicated in the .working member of
1889 * struct thread_params).
1890 * When a work thread has completed its work, it sets .working to 0 and
1891 * signals the main thread and waits on the condition that .data_ready
1892 * becomes 1.
1893 */
1894
1895struct thread_params {
1896 pthread_t thread;
1897 struct object_entry **list;
1898 unsigned list_size;
1899 unsigned remaining;
1900 int window;
1901 int depth;
1902 int working;
1903 int data_ready;
1904 pthread_mutex_t mutex;
1905 pthread_cond_t cond;
1906 unsigned *processed;
1907};
1908
1909static pthread_cond_t progress_cond;
1910
1911/*
1912 * Mutex and conditional variable can't be statically-initialized on Windows.
1913 */
1914static void init_threaded_search(void)
1915{
1916 init_recursive_mutex(&read_mutex);
1917 pthread_mutex_init(&cache_mutex, NULL);
1918 pthread_mutex_init(&progress_mutex, NULL);
1919 pthread_cond_init(&progress_cond, NULL);
1920 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1921}
1922
1923static void cleanup_threaded_search(void)
1924{
1925 set_try_to_free_routine(old_try_to_free_routine);
1926 pthread_cond_destroy(&progress_cond);
1927 pthread_mutex_destroy(&read_mutex);
1928 pthread_mutex_destroy(&cache_mutex);
1929 pthread_mutex_destroy(&progress_mutex);
1930}
1931
1932static void *threaded_find_deltas(void *arg)
1933{
1934 struct thread_params *me = arg;
1935
1936 while (me->remaining) {
1937 find_deltas(me->list, &me->remaining,
1938 me->window, me->depth, me->processed);
1939
1940 progress_lock();
1941 me->working = 0;
1942 pthread_cond_signal(&progress_cond);
1943 progress_unlock();
1944
1945 /*
1946 * We must not set ->data_ready before we wait on the
1947 * condition because the main thread may have set it to 1
1948 * before we get here. In order to be sure that new
1949 * work is available if we see 1 in ->data_ready, it
1950 * was initialized to 0 before this thread was spawned
1951 * and we reset it to 0 right away.
1952 */
1953 pthread_mutex_lock(&me->mutex);
1954 while (!me->data_ready)
1955 pthread_cond_wait(&me->cond, &me->mutex);
1956 me->data_ready = 0;
1957 pthread_mutex_unlock(&me->mutex);
1958 }
1959 /* leave ->working 1 so that this doesn't get more work assigned */
1960 return NULL;
1961}
1962
1963static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1964 int window, int depth, unsigned *processed)
1965{
1966 struct thread_params *p;
1967 int i, ret, active_threads = 0;
1968
1969 init_threaded_search();
1970
1971 if (!delta_search_threads) /* --threads=0 means autodetect */
1972 delta_search_threads = online_cpus();
1973 if (delta_search_threads <= 1) {
1974 find_deltas(list, &list_size, window, depth, processed);
1975 cleanup_threaded_search();
1976 return;
1977 }
1978 if (progress > pack_to_stdout)
1979 fprintf(stderr, "Delta compression using up to %d threads.\n",
1980 delta_search_threads);
1981 p = xcalloc(delta_search_threads, sizeof(*p));
1982
1983 /* Partition the work amongst work threads. */
1984 for (i = 0; i < delta_search_threads; i++) {
1985 unsigned sub_size = list_size / (delta_search_threads - i);
1986
1987 /* don't use too small segments or no deltas will be found */
1988 if (sub_size < 2*window && i+1 < delta_search_threads)
1989 sub_size = 0;
1990
1991 p[i].window = window;
1992 p[i].depth = depth;
1993 p[i].processed = processed;
1994 p[i].working = 1;
1995 p[i].data_ready = 0;
1996
1997 /* try to split chunks on "path" boundaries */
1998 while (sub_size && sub_size < list_size &&
1999 list[sub_size]->hash &&
2000 list[sub_size]->hash == list[sub_size-1]->hash)
2001 sub_size++;
2002
2003 p[i].list = list;
2004 p[i].list_size = sub_size;
2005 p[i].remaining = sub_size;
2006
2007 list += sub_size;
2008 list_size -= sub_size;
2009 }
2010
2011 /* Start work threads. */
2012 for (i = 0; i < delta_search_threads; i++) {
2013 if (!p[i].list_size)
2014 continue;
2015 pthread_mutex_init(&p[i].mutex, NULL);
2016 pthread_cond_init(&p[i].cond, NULL);
2017 ret = pthread_create(&p[i].thread, NULL,
2018 threaded_find_deltas, &p[i]);
2019 if (ret)
2020 die("unable to create thread: %s", strerror(ret));
2021 active_threads++;
2022 }
2023
2024 /*
2025 * Now let's wait for work completion. Each time a thread is done
2026 * with its work, we steal half of the remaining work from the
2027 * thread with the largest number of unprocessed objects and give
2028 * it to that newly idle thread. This ensure good load balancing
2029 * until the remaining object list segments are simply too short
2030 * to be worth splitting anymore.
2031 */
2032 while (active_threads) {
2033 struct thread_params *target = NULL;
2034 struct thread_params *victim = NULL;
2035 unsigned sub_size = 0;
2036
2037 progress_lock();
2038 for (;;) {
2039 for (i = 0; !target && i < delta_search_threads; i++)
2040 if (!p[i].working)
2041 target = &p[i];
2042 if (target)
2043 break;
2044 pthread_cond_wait(&progress_cond, &progress_mutex);
2045 }
2046
2047 for (i = 0; i < delta_search_threads; i++)
2048 if (p[i].remaining > 2*window &&
2049 (!victim || victim->remaining < p[i].remaining))
2050 victim = &p[i];
2051 if (victim) {
2052 sub_size = victim->remaining / 2;
2053 list = victim->list + victim->list_size - sub_size;
2054 while (sub_size && list[0]->hash &&
2055 list[0]->hash == list[-1]->hash) {
2056 list++;
2057 sub_size--;
2058 }
2059 if (!sub_size) {
2060 /*
2061 * It is possible for some "paths" to have
2062 * so many objects that no hash boundary
2063 * might be found. Let's just steal the
2064 * exact half in that case.
2065 */
2066 sub_size = victim->remaining / 2;
2067 list -= sub_size;
2068 }
2069 target->list = list;
2070 victim->list_size -= sub_size;
2071 victim->remaining -= sub_size;
2072 }
2073 target->list_size = sub_size;
2074 target->remaining = sub_size;
2075 target->working = 1;
2076 progress_unlock();
2077
2078 pthread_mutex_lock(&target->mutex);
2079 target->data_ready = 1;
2080 pthread_cond_signal(&target->cond);
2081 pthread_mutex_unlock(&target->mutex);
2082
2083 if (!sub_size) {
2084 pthread_join(target->thread, NULL);
2085 pthread_cond_destroy(&target->cond);
2086 pthread_mutex_destroy(&target->mutex);
2087 active_threads--;
2088 }
2089 }
2090 cleanup_threaded_search();
2091 free(p);
2092}
2093
2094#else
2095#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2096#endif
2097
2098static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
2099{
2100 unsigned char peeled[20];
2101
2102 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
2103 !peel_ref(path, peeled) && /* peelable? */
2104 packlist_find(&to_pack, peeled, NULL)) /* object packed? */
2105 add_object_entry(sha1, OBJ_TAG, NULL, 0);
2106 return 0;
2107}
2108
2109static void prepare_pack(int window, int depth)
2110{
2111 struct object_entry **delta_list;
2112 uint32_t i, nr_deltas;
2113 unsigned n;
2114
2115 get_object_details();
2116
2117 /*
2118 * If we're locally repacking then we need to be doubly careful
2119 * from now on in order to make sure no stealth corruption gets
2120 * propagated to the new pack. Clients receiving streamed packs
2121 * should validate everything they get anyway so no need to incur
2122 * the additional cost here in that case.
2123 */
2124 if (!pack_to_stdout)
2125 do_check_packed_object_crc = 1;
2126
2127 if (!to_pack.nr_objects || !window || !depth)
2128 return;
2129
2130 delta_list = xmalloc(to_pack.nr_objects * sizeof(*delta_list));
2131 nr_deltas = n = 0;
2132
2133 for (i = 0; i < to_pack.nr_objects; i++) {
2134 struct object_entry *entry = to_pack.objects + i;
2135
2136 if (entry->delta)
2137 /* This happens if we decided to reuse existing
2138 * delta from a pack. "reuse_delta &&" is implied.
2139 */
2140 continue;
2141
2142 if (entry->size < 50)
2143 continue;
2144
2145 if (entry->no_try_delta)
2146 continue;
2147
2148 if (!entry->preferred_base) {
2149 nr_deltas++;
2150 if (entry->type < 0)
2151 die("unable to get type of object %s",
2152 sha1_to_hex(entry->idx.sha1));
2153 } else {
2154 if (entry->type < 0) {
2155 /*
2156 * This object is not found, but we
2157 * don't have to include it anyway.
2158 */
2159 continue;
2160 }
2161 }
2162
2163 delta_list[n++] = entry;
2164 }
2165
2166 if (nr_deltas && n > 1) {
2167 unsigned nr_done = 0;
2168 if (progress)
2169 progress_state = start_progress("Compressing objects",
2170 nr_deltas);
2171 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2172 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2173 stop_progress(&progress_state);
2174 if (nr_done != nr_deltas)
2175 die("inconsistency with delta count");
2176 }
2177 free(delta_list);
2178}
2179
2180static int git_pack_config(const char *k, const char *v, void *cb)
2181{
2182 if (!strcmp(k, "pack.window")) {
2183 window = git_config_int(k, v);
2184 return 0;
2185 }
2186 if (!strcmp(k, "pack.windowmemory")) {
2187 window_memory_limit = git_config_ulong(k, v);
2188 return 0;
2189 }
2190 if (!strcmp(k, "pack.depth")) {
2191 depth = git_config_int(k, v);
2192 return 0;
2193 }
2194 if (!strcmp(k, "pack.compression")) {
2195 int level = git_config_int(k, v);
2196 if (level == -1)
2197 level = Z_DEFAULT_COMPRESSION;
2198 else if (level < 0 || level > Z_BEST_COMPRESSION)
2199 die("bad pack compression level %d", level);
2200 pack_compression_level = level;
2201 pack_compression_seen = 1;
2202 return 0;
2203 }
2204 if (!strcmp(k, "pack.deltacachesize")) {
2205 max_delta_cache_size = git_config_int(k, v);
2206 return 0;
2207 }
2208 if (!strcmp(k, "pack.deltacachelimit")) {
2209 cache_max_small_delta_size = git_config_int(k, v);
2210 return 0;
2211 }
2212 if (!strcmp(k, "pack.writebitmaps")) {
2213 write_bitmap_index = git_config_bool(k, v);
2214 return 0;
2215 }
2216 if (!strcmp(k, "pack.writebitmaphashcache")) {
2217 if (git_config_bool(k, v))
2218 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2219 else
2220 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2221 }
2222 if (!strcmp(k, "pack.usebitmaps")) {
2223 use_bitmap_index = git_config_bool(k, v);
2224 return 0;
2225 }
2226 if (!strcmp(k, "pack.threads")) {
2227 delta_search_threads = git_config_int(k, v);
2228 if (delta_search_threads < 0)
2229 die("invalid number of threads specified (%d)",
2230 delta_search_threads);
2231#ifdef NO_PTHREADS
2232 if (delta_search_threads != 1)
2233 warning("no threads support, ignoring %s", k);
2234#endif
2235 return 0;
2236 }
2237 if (!strcmp(k, "pack.indexversion")) {
2238 pack_idx_opts.version = git_config_int(k, v);
2239 if (pack_idx_opts.version > 2)
2240 die("bad pack.indexversion=%"PRIu32,
2241 pack_idx_opts.version);
2242 return 0;
2243 }
2244 return git_default_config(k, v, cb);
2245}
2246
2247static void read_object_list_from_stdin(void)
2248{
2249 char line[40 + 1 + PATH_MAX + 2];
2250 unsigned char sha1[20];
2251
2252 for (;;) {
2253 if (!fgets(line, sizeof(line), stdin)) {
2254 if (feof(stdin))
2255 break;
2256 if (!ferror(stdin))
2257 die("fgets returned NULL, not EOF, not error!");
2258 if (errno != EINTR)
2259 die_errno("fgets");
2260 clearerr(stdin);
2261 continue;
2262 }
2263 if (line[0] == '-') {
2264 if (get_sha1_hex(line+1, sha1))
2265 die("expected edge sha1, got garbage:\n %s",
2266 line);
2267 add_preferred_base(sha1);
2268 continue;
2269 }
2270 if (get_sha1_hex(line, sha1))
2271 die("expected sha1, got garbage:\n %s", line);
2272
2273 add_preferred_base_object(line+41);
2274 add_object_entry(sha1, 0, line+41, 0);
2275 }
2276}
2277
2278#define OBJECT_ADDED (1u<<20)
2279
2280static void show_commit(struct commit *commit, void *data)
2281{
2282 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2283 commit->object.flags |= OBJECT_ADDED;
2284
2285 if (write_bitmap_index)
2286 index_commit_for_bitmap(commit);
2287}
2288
2289static void show_object(struct object *obj,
2290 const struct name_path *path, const char *last,
2291 void *data)
2292{
2293 char *name = path_name(path, last);
2294
2295 add_preferred_base_object(name);
2296 add_object_entry(obj->sha1, obj->type, name, 0);
2297 obj->flags |= OBJECT_ADDED;
2298
2299 /*
2300 * We will have generated the hash from the name,
2301 * but not saved a pointer to it - we can free it
2302 */
2303 free((char *)name);
2304}
2305
2306static void show_edge(struct commit *commit)
2307{
2308 add_preferred_base(commit->object.sha1);
2309}
2310
2311struct in_pack_object {
2312 off_t offset;
2313 struct object *object;
2314};
2315
2316struct in_pack {
2317 int alloc;
2318 int nr;
2319 struct in_pack_object *array;
2320};
2321
2322static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2323{
2324 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2325 in_pack->array[in_pack->nr].object = object;
2326 in_pack->nr++;
2327}
2328
2329/*
2330 * Compare the objects in the offset order, in order to emulate the
2331 * "git rev-list --objects" output that produced the pack originally.
2332 */
2333static int ofscmp(const void *a_, const void *b_)
2334{
2335 struct in_pack_object *a = (struct in_pack_object *)a_;
2336 struct in_pack_object *b = (struct in_pack_object *)b_;
2337
2338 if (a->offset < b->offset)
2339 return -1;
2340 else if (a->offset > b->offset)
2341 return 1;
2342 else
2343 return hashcmp(a->object->sha1, b->object->sha1);
2344}
2345
2346static void add_objects_in_unpacked_packs(struct rev_info *revs)
2347{
2348 struct packed_git *p;
2349 struct in_pack in_pack;
2350 uint32_t i;
2351
2352 memset(&in_pack, 0, sizeof(in_pack));
2353
2354 for (p = packed_git; p; p = p->next) {
2355 const unsigned char *sha1;
2356 struct object *o;
2357
2358 if (!p->pack_local || p->pack_keep)
2359 continue;
2360 if (open_pack_index(p))
2361 die("cannot open pack index");
2362
2363 ALLOC_GROW(in_pack.array,
2364 in_pack.nr + p->num_objects,
2365 in_pack.alloc);
2366
2367 for (i = 0; i < p->num_objects; i++) {
2368 sha1 = nth_packed_object_sha1(p, i);
2369 o = lookup_unknown_object(sha1);
2370 if (!(o->flags & OBJECT_ADDED))
2371 mark_in_pack_object(o, p, &in_pack);
2372 o->flags |= OBJECT_ADDED;
2373 }
2374 }
2375
2376 if (in_pack.nr) {
2377 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2378 ofscmp);
2379 for (i = 0; i < in_pack.nr; i++) {
2380 struct object *o = in_pack.array[i].object;
2381 add_object_entry(o->sha1, o->type, "", 0);
2382 }
2383 }
2384 free(in_pack.array);
2385}
2386
2387static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2388{
2389 static struct packed_git *last_found = (void *)1;
2390 struct packed_git *p;
2391
2392 p = (last_found != (void *)1) ? last_found : packed_git;
2393
2394 while (p) {
2395 if ((!p->pack_local || p->pack_keep) &&
2396 find_pack_entry_one(sha1, p)) {
2397 last_found = p;
2398 return 1;
2399 }
2400 if (p == last_found)
2401 p = packed_git;
2402 else
2403 p = p->next;
2404 if (p == last_found)
2405 p = p->next;
2406 }
2407 return 0;
2408}
2409
2410static void loosen_unused_packed_objects(struct rev_info *revs)
2411{
2412 struct packed_git *p;
2413 uint32_t i;
2414 const unsigned char *sha1;
2415
2416 for (p = packed_git; p; p = p->next) {
2417 if (!p->pack_local || p->pack_keep)
2418 continue;
2419
2420 if (unpack_unreachable_expiration &&
2421 p->mtime < unpack_unreachable_expiration)
2422 continue;
2423
2424 if (open_pack_index(p))
2425 die("cannot open pack index");
2426
2427 for (i = 0; i < p->num_objects; i++) {
2428 sha1 = nth_packed_object_sha1(p, i);
2429 if (!packlist_find(&to_pack, sha1, NULL) &&
2430 !has_sha1_pack_kept_or_nonlocal(sha1))
2431 if (force_object_loose(sha1, p->mtime))
2432 die("unable to force loose object");
2433 }
2434 }
2435}
2436
2437static int get_object_list_from_bitmap(struct rev_info *revs)
2438{
2439 if (prepare_bitmap_walk(revs) < 0)
2440 return -1;
2441
2442 if (!reuse_partial_packfile_from_bitmap(
2443 &reuse_packfile,
2444 &reuse_packfile_objects,
2445 &reuse_packfile_offset)) {
2446 assert(reuse_packfile_objects);
2447 nr_result += reuse_packfile_objects;
2448
2449 if (progress) {
2450 fprintf(stderr, "Reusing existing pack: %d, done.\n",
2451 reuse_packfile_objects);
2452 fflush(stderr);
2453 }
2454 }
2455
2456 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2457 return 0;
2458}
2459
2460static void get_object_list(int ac, const char **av)
2461{
2462 struct rev_info revs;
2463 char line[1000];
2464 int flags = 0;
2465
2466 init_revisions(&revs, NULL);
2467 save_commit_buffer = 0;
2468 setup_revisions(ac, av, &revs, NULL);
2469
2470 while (fgets(line, sizeof(line), stdin) != NULL) {
2471 int len = strlen(line);
2472 if (len && line[len - 1] == '\n')
2473 line[--len] = 0;
2474 if (!len)
2475 break;
2476 if (*line == '-') {
2477 if (!strcmp(line, "--not")) {
2478 flags ^= UNINTERESTING;
2479 write_bitmap_index = 0;
2480 continue;
2481 }
2482 die("not a rev '%s'", line);
2483 }
2484 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2485 die("bad revision '%s'", line);
2486 }
2487
2488 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2489 return;
2490
2491 if (prepare_revision_walk(&revs))
2492 die("revision walk setup failed");
2493 mark_edges_uninteresting(&revs, show_edge);
2494 traverse_commit_list(&revs, show_commit, show_object, NULL);
2495
2496 if (keep_unreachable)
2497 add_objects_in_unpacked_packs(&revs);
2498 if (unpack_unreachable)
2499 loosen_unused_packed_objects(&revs);
2500}
2501
2502static int option_parse_index_version(const struct option *opt,
2503 const char *arg, int unset)
2504{
2505 char *c;
2506 const char *val = arg;
2507 pack_idx_opts.version = strtoul(val, &c, 10);
2508 if (pack_idx_opts.version > 2)
2509 die(_("unsupported index version %s"), val);
2510 if (*c == ',' && c[1])
2511 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2512 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2513 die(_("bad index version '%s'"), val);
2514 return 0;
2515}
2516
2517static int option_parse_unpack_unreachable(const struct option *opt,
2518 const char *arg, int unset)
2519{
2520 if (unset) {
2521 unpack_unreachable = 0;
2522 unpack_unreachable_expiration = 0;
2523 }
2524 else {
2525 unpack_unreachable = 1;
2526 if (arg)
2527 unpack_unreachable_expiration = approxidate(arg);
2528 }
2529 return 0;
2530}
2531
2532static int option_parse_ulong(const struct option *opt,
2533 const char *arg, int unset)
2534{
2535 if (unset)
2536 die(_("option %s does not accept negative form"),
2537 opt->long_name);
2538
2539 if (!git_parse_ulong(arg, opt->value))
2540 die(_("unable to parse value '%s' for option %s"),
2541 arg, opt->long_name);
2542 return 0;
2543}
2544
2545#define OPT_ULONG(s, l, v, h) \
2546 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2547 PARSE_OPT_NONEG, option_parse_ulong }
2548
2549int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2550{
2551 int use_internal_rev_list = 0;
2552 int thin = 0;
2553 int all_progress_implied = 0;
2554 const char *rp_av[6];
2555 int rp_ac = 0;
2556 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2557 struct option pack_objects_options[] = {
2558 OPT_SET_INT('q', "quiet", &progress,
2559 N_("do not show progress meter"), 0),
2560 OPT_SET_INT(0, "progress", &progress,
2561 N_("show progress meter"), 1),
2562 OPT_SET_INT(0, "all-progress", &progress,
2563 N_("show progress meter during object writing phase"), 2),
2564 OPT_BOOL(0, "all-progress-implied",
2565 &all_progress_implied,
2566 N_("similar to --all-progress when progress meter is shown")),
2567 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2568 N_("write the pack index file in the specified idx format version"),
2569 0, option_parse_index_version },
2570 OPT_ULONG(0, "max-pack-size", &pack_size_limit,
2571 N_("maximum size of each output pack file")),
2572 OPT_BOOL(0, "local", &local,
2573 N_("ignore borrowed objects from alternate object store")),
2574 OPT_BOOL(0, "incremental", &incremental,
2575 N_("ignore packed objects")),
2576 OPT_INTEGER(0, "window", &window,
2577 N_("limit pack window by objects")),
2578 OPT_ULONG(0, "window-memory", &window_memory_limit,
2579 N_("limit pack window by memory in addition to object limit")),
2580 OPT_INTEGER(0, "depth", &depth,
2581 N_("maximum length of delta chain allowed in the resulting pack")),
2582 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2583 N_("reuse existing deltas")),
2584 OPT_BOOL(0, "reuse-object", &reuse_object,
2585 N_("reuse existing objects")),
2586 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2587 N_("use OFS_DELTA objects")),
2588 OPT_INTEGER(0, "threads", &delta_search_threads,
2589 N_("use threads when searching for best delta matches")),
2590 OPT_BOOL(0, "non-empty", &non_empty,
2591 N_("do not create an empty pack output")),
2592 OPT_BOOL(0, "revs", &use_internal_rev_list,
2593 N_("read revision arguments from standard input")),
2594 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2595 N_("limit the objects to those that are not yet packed"),
2596 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2597 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2598 N_("include objects reachable from any reference"),
2599 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2600 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2601 N_("include objects referred by reflog entries"),
2602 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2603 OPT_BOOL(0, "stdout", &pack_to_stdout,
2604 N_("output pack to stdout")),
2605 OPT_BOOL(0, "include-tag", &include_tag,
2606 N_("include tag objects that refer to objects to be packed")),
2607 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2608 N_("keep unreachable objects")),
2609 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2610 N_("unpack unreachable objects newer than <time>"),
2611 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2612 OPT_BOOL(0, "thin", &thin,
2613 N_("create thin packs")),
2614 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2615 N_("ignore packs that have companion .keep file")),
2616 OPT_INTEGER(0, "compression", &pack_compression_level,
2617 N_("pack compression level")),
2618 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2619 N_("do not hide commits by grafts"), 0),
2620 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2621 N_("use a bitmap index if available to speed up counting objects")),
2622 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2623 N_("write a bitmap index together with the pack index")),
2624 OPT_END(),
2625 };
2626
2627 read_replace_refs = 0;
2628
2629 reset_pack_idx_option(&pack_idx_opts);
2630 git_config(git_pack_config, NULL);
2631 if (!pack_compression_seen && core_compression_seen)
2632 pack_compression_level = core_compression_level;
2633
2634 progress = isatty(2);
2635 argc = parse_options(argc, argv, prefix, pack_objects_options,
2636 pack_usage, 0);
2637
2638 if (argc) {
2639 base_name = argv[0];
2640 argc--;
2641 }
2642 if (pack_to_stdout != !base_name || argc)
2643 usage_with_options(pack_usage, pack_objects_options);
2644
2645 rp_av[rp_ac++] = "pack-objects";
2646 if (thin) {
2647 use_internal_rev_list = 1;
2648 rp_av[rp_ac++] = "--objects-edge";
2649 } else
2650 rp_av[rp_ac++] = "--objects";
2651
2652 if (rev_list_all) {
2653 use_internal_rev_list = 1;
2654 rp_av[rp_ac++] = "--all";
2655 }
2656 if (rev_list_reflog) {
2657 use_internal_rev_list = 1;
2658 rp_av[rp_ac++] = "--reflog";
2659 }
2660 if (rev_list_unpacked) {
2661 use_internal_rev_list = 1;
2662 rp_av[rp_ac++] = "--unpacked";
2663 }
2664
2665 if (!reuse_object)
2666 reuse_delta = 0;
2667 if (pack_compression_level == -1)
2668 pack_compression_level = Z_DEFAULT_COMPRESSION;
2669 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2670 die("bad pack compression level %d", pack_compression_level);
2671#ifdef NO_PTHREADS
2672 if (delta_search_threads != 1)
2673 warning("no threads support, ignoring --threads");
2674#endif
2675 if (!pack_to_stdout && !pack_size_limit)
2676 pack_size_limit = pack_size_limit_cfg;
2677 if (pack_to_stdout && pack_size_limit)
2678 die("--max-pack-size cannot be used to build a pack for transfer.");
2679 if (pack_size_limit && pack_size_limit < 1024*1024) {
2680 warning("minimum pack size limit is 1 MiB");
2681 pack_size_limit = 1024*1024;
2682 }
2683
2684 if (!pack_to_stdout && thin)
2685 die("--thin cannot be used to build an indexable pack.");
2686
2687 if (keep_unreachable && unpack_unreachable)
2688 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2689
2690 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2691 use_bitmap_index = 0;
2692
2693 if (pack_to_stdout || !rev_list_all)
2694 write_bitmap_index = 0;
2695
2696 if (progress && all_progress_implied)
2697 progress = 2;
2698
2699 prepare_packed_git();
2700
2701 if (progress)
2702 progress_state = start_progress("Counting objects", 0);
2703 if (!use_internal_rev_list)
2704 read_object_list_from_stdin();
2705 else {
2706 rp_av[rp_ac] = NULL;
2707 get_object_list(rp_ac, rp_av);
2708 }
2709 cleanup_preferred_base();
2710 if (include_tag && nr_result)
2711 for_each_ref(add_ref_tag, NULL);
2712 stop_progress(&progress_state);
2713
2714 if (non_empty && !nr_result)
2715 return 0;
2716 if (nr_result)
2717 prepare_pack(window, depth);
2718 write_pack_file();
2719 if (progress)
2720 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2721 " reused %"PRIu32" (delta %"PRIu32")\n",
2722 written, written_delta, reused, reused_delta);
2723 return 0;
2724}