convert trivial cases to ALLOC_ARRAY
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "pack-objects.h"
18 #include "progress.h"
19 #include "refs.h"
20 #include "streaming.h"
21 #include "thread-utils.h"
22 #include "pack-bitmap.h"
23 #include "reachable.h"
24 #include "sha1-array.h"
25 #include "argv-array.h"
26
27 static const char *pack_usage[] = {
28 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
29 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
30 NULL
31 };
32
33 /*
34 * Objects we are going to pack are collected in the `to_pack` structure.
35 * It contains an array (dynamically expanded) of the object data, and a map
36 * that can resolve SHA1s to their position in the array.
37 */
38 static struct packing_data to_pack;
39
40 static struct pack_idx_entry **written_list;
41 static uint32_t nr_result, nr_written;
42
43 static int non_empty;
44 static int reuse_delta = 1, reuse_object = 1;
45 static int keep_unreachable, unpack_unreachable, include_tag;
46 static unsigned long unpack_unreachable_expiration;
47 static int local;
48 static int incremental;
49 static int ignore_packed_keep;
50 static int allow_ofs_delta;
51 static struct pack_idx_option pack_idx_opts;
52 static const char *base_name;
53 static int progress = 1;
54 static int window = 10;
55 static unsigned long pack_size_limit;
56 static int depth = 50;
57 static int delta_search_threads;
58 static int pack_to_stdout;
59 static int num_preferred_base;
60 static struct progress *progress_state;
61 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
62 static int pack_compression_seen;
63
64 static struct packed_git *reuse_packfile;
65 static uint32_t reuse_packfile_objects;
66 static off_t reuse_packfile_offset;
67
68 static int use_bitmap_index = 1;
69 static int write_bitmap_index;
70 static uint16_t write_bitmap_options;
71
72 static unsigned long delta_cache_size = 0;
73 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
74 static unsigned long cache_max_small_delta_size = 1000;
75
76 static unsigned long window_memory_limit = 0;
77
78 /*
79 * stats
80 */
81 static uint32_t written, written_delta;
82 static uint32_t reused, reused_delta;
83
84 /*
85 * Indexed commits
86 */
87 static struct commit **indexed_commits;
88 static unsigned int indexed_commits_nr;
89 static unsigned int indexed_commits_alloc;
90
91 static void index_commit_for_bitmap(struct commit *commit)
92 {
93 if (indexed_commits_nr >= indexed_commits_alloc) {
94 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
95 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
96 }
97
98 indexed_commits[indexed_commits_nr++] = commit;
99 }
100
101 static void *get_delta(struct object_entry *entry)
102 {
103 unsigned long size, base_size, delta_size;
104 void *buf, *base_buf, *delta_buf;
105 enum object_type type;
106
107 buf = read_sha1_file(entry->idx.sha1, &type, &size);
108 if (!buf)
109 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
110 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
111 if (!base_buf)
112 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
113 delta_buf = diff_delta(base_buf, base_size,
114 buf, size, &delta_size, 0);
115 if (!delta_buf || delta_size != entry->delta_size)
116 die("delta size changed");
117 free(buf);
118 free(base_buf);
119 return delta_buf;
120 }
121
122 static unsigned long do_compress(void **pptr, unsigned long size)
123 {
124 git_zstream stream;
125 void *in, *out;
126 unsigned long maxsize;
127
128 git_deflate_init(&stream, pack_compression_level);
129 maxsize = git_deflate_bound(&stream, size);
130
131 in = *pptr;
132 out = xmalloc(maxsize);
133 *pptr = out;
134
135 stream.next_in = in;
136 stream.avail_in = size;
137 stream.next_out = out;
138 stream.avail_out = maxsize;
139 while (git_deflate(&stream, Z_FINISH) == Z_OK)
140 ; /* nothing */
141 git_deflate_end(&stream);
142
143 free(in);
144 return stream.total_out;
145 }
146
147 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
148 const unsigned char *sha1)
149 {
150 git_zstream stream;
151 unsigned char ibuf[1024 * 16];
152 unsigned char obuf[1024 * 16];
153 unsigned long olen = 0;
154
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 */
190 static 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
216 static 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 */
236 static 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 */
344 static 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 */
418 static 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
483 enum 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
490 static 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
543 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
544 void *cb_data)
545 {
546 unsigned char peeled[20];
547 struct object_entry *entry = packlist_find(&to_pack, oid->hash, 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
559 static 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
569 static 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
612 static 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
623 static struct object_entry **compute_write_order(void)
624 {
625 unsigned int i, wo_end, last_untagged;
626
627 struct object_entry **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 ALLOC_ARRAY(wo, to_pack.nr_objects);
661 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
662 if (objects[i].tagged)
663 break;
664 add_to_write_order(wo, &wo_end, &objects[i]);
665 }
666 last_untagged = i;
667
668 /*
669 * Then fill all the tagged tips.
670 */
671 for (; i < to_pack.nr_objects; i++) {
672 if (objects[i].tagged)
673 add_to_write_order(wo, &wo_end, &objects[i]);
674 }
675
676 /*
677 * And then all remaining commits and tags.
678 */
679 for (i = last_untagged; i < to_pack.nr_objects; i++) {
680 if (objects[i].type != OBJ_COMMIT &&
681 objects[i].type != OBJ_TAG)
682 continue;
683 add_to_write_order(wo, &wo_end, &objects[i]);
684 }
685
686 /*
687 * And then all the trees.
688 */
689 for (i = last_untagged; i < to_pack.nr_objects; i++) {
690 if (objects[i].type != OBJ_TREE)
691 continue;
692 add_to_write_order(wo, &wo_end, &objects[i]);
693 }
694
695 /*
696 * Finally all the rest in really tight order
697 */
698 for (i = last_untagged; i < to_pack.nr_objects; i++) {
699 if (!objects[i].filled)
700 add_family_to_write_order(wo, &wo_end, &objects[i]);
701 }
702
703 if (wo_end != to_pack.nr_objects)
704 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
705
706 return wo;
707 }
708
709 static off_t write_reused_pack(struct sha1file *f)
710 {
711 unsigned char buffer[8192];
712 off_t to_write, total;
713 int fd;
714
715 if (!is_pack_valid(reuse_packfile))
716 die("packfile is invalid: %s", reuse_packfile->pack_name);
717
718 fd = git_open_noatime(reuse_packfile->pack_name);
719 if (fd < 0)
720 die_errno("unable to open packfile for reuse: %s",
721 reuse_packfile->pack_name);
722
723 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
724 die_errno("unable to seek in reused packfile");
725
726 if (reuse_packfile_offset < 0)
727 reuse_packfile_offset = reuse_packfile->pack_size - 20;
728
729 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
730
731 while (to_write) {
732 int read_pack = xread(fd, buffer, sizeof(buffer));
733
734 if (read_pack <= 0)
735 die_errno("unable to read from reused packfile");
736
737 if (read_pack > to_write)
738 read_pack = to_write;
739
740 sha1write(f, buffer, read_pack);
741 to_write -= read_pack;
742
743 /*
744 * We don't know the actual number of objects written,
745 * only how many bytes written, how many bytes total, and
746 * how many objects total. So we can fake it by pretending all
747 * objects we are writing are the same size. This gives us a
748 * smooth progress meter, and at the end it matches the true
749 * answer.
750 */
751 written = reuse_packfile_objects *
752 (((double)(total - to_write)) / total);
753 display_progress(progress_state, written);
754 }
755
756 close(fd);
757 written = reuse_packfile_objects;
758 display_progress(progress_state, written);
759 return reuse_packfile_offset - sizeof(struct pack_header);
760 }
761
762 static void write_pack_file(void)
763 {
764 uint32_t i = 0, j;
765 struct sha1file *f;
766 off_t offset;
767 uint32_t nr_remaining = nr_result;
768 time_t last_mtime = 0;
769 struct object_entry **write_order;
770
771 if (progress > pack_to_stdout)
772 progress_state = start_progress(_("Writing objects"), nr_result);
773 ALLOC_ARRAY(written_list, to_pack.nr_objects);
774 write_order = compute_write_order();
775
776 do {
777 unsigned char sha1[20];
778 char *pack_tmp_name = NULL;
779
780 if (pack_to_stdout)
781 f = sha1fd_throughput(1, "<stdout>", progress_state);
782 else
783 f = create_tmp_packfile(&pack_tmp_name);
784
785 offset = write_pack_header(f, nr_remaining);
786
787 if (reuse_packfile) {
788 off_t packfile_size;
789 assert(pack_to_stdout);
790
791 packfile_size = write_reused_pack(f);
792 offset += packfile_size;
793 }
794
795 nr_written = 0;
796 for (; i < to_pack.nr_objects; i++) {
797 struct object_entry *e = write_order[i];
798 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
799 break;
800 display_progress(progress_state, written);
801 }
802
803 /*
804 * Did we write the wrong # entries in the header?
805 * If so, rewrite it like in fast-import
806 */
807 if (pack_to_stdout) {
808 sha1close(f, sha1, CSUM_CLOSE);
809 } else if (nr_written == nr_remaining) {
810 sha1close(f, sha1, CSUM_FSYNC);
811 } else {
812 int fd = sha1close(f, sha1, 0);
813 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
814 nr_written, sha1, offset);
815 close(fd);
816 write_bitmap_index = 0;
817 }
818
819 if (!pack_to_stdout) {
820 struct stat st;
821 struct strbuf tmpname = STRBUF_INIT;
822
823 /*
824 * Packs are runtime accessed in their mtime
825 * order since newer packs are more likely to contain
826 * younger objects. So if we are creating multiple
827 * packs then we should modify the mtime of later ones
828 * to preserve this property.
829 */
830 if (stat(pack_tmp_name, &st) < 0) {
831 warning("failed to stat %s: %s",
832 pack_tmp_name, strerror(errno));
833 } else if (!last_mtime) {
834 last_mtime = st.st_mtime;
835 } else {
836 struct utimbuf utb;
837 utb.actime = st.st_atime;
838 utb.modtime = --last_mtime;
839 if (utime(pack_tmp_name, &utb) < 0)
840 warning("failed utime() on %s: %s",
841 pack_tmp_name, strerror(errno));
842 }
843
844 strbuf_addf(&tmpname, "%s-", base_name);
845
846 if (write_bitmap_index) {
847 bitmap_writer_set_checksum(sha1);
848 bitmap_writer_build_type_index(written_list, nr_written);
849 }
850
851 finish_tmp_packfile(&tmpname, pack_tmp_name,
852 written_list, nr_written,
853 &pack_idx_opts, sha1);
854
855 if (write_bitmap_index) {
856 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
857
858 stop_progress(&progress_state);
859
860 bitmap_writer_show_progress(progress);
861 bitmap_writer_reuse_bitmaps(&to_pack);
862 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
863 bitmap_writer_build(&to_pack);
864 bitmap_writer_finish(written_list, nr_written,
865 tmpname.buf, write_bitmap_options);
866 write_bitmap_index = 0;
867 }
868
869 strbuf_release(&tmpname);
870 free(pack_tmp_name);
871 puts(sha1_to_hex(sha1));
872 }
873
874 /* mark written objects as written to previous pack */
875 for (j = 0; j < nr_written; j++) {
876 written_list[j]->offset = (off_t)-1;
877 }
878 nr_remaining -= nr_written;
879 } while (nr_remaining && i < to_pack.nr_objects);
880
881 free(written_list);
882 free(write_order);
883 stop_progress(&progress_state);
884 if (written != nr_result)
885 die("wrote %"PRIu32" objects while expecting %"PRIu32,
886 written, nr_result);
887 }
888
889 static void setup_delta_attr_check(struct git_attr_check *check)
890 {
891 static struct git_attr *attr_delta;
892
893 if (!attr_delta)
894 attr_delta = git_attr("delta");
895
896 check[0].attr = attr_delta;
897 }
898
899 static int no_try_delta(const char *path)
900 {
901 struct git_attr_check check[1];
902
903 setup_delta_attr_check(check);
904 if (git_check_attr(path, ARRAY_SIZE(check), check))
905 return 0;
906 if (ATTR_FALSE(check->value))
907 return 1;
908 return 0;
909 }
910
911 /*
912 * When adding an object, check whether we have already added it
913 * to our packing list. If so, we can skip. However, if we are
914 * being asked to excludei t, but the previous mention was to include
915 * it, make sure to adjust its flags and tweak our numbers accordingly.
916 *
917 * As an optimization, we pass out the index position where we would have
918 * found the item, since that saves us from having to look it up again a
919 * few lines later when we want to add the new entry.
920 */
921 static int have_duplicate_entry(const unsigned char *sha1,
922 int exclude,
923 uint32_t *index_pos)
924 {
925 struct object_entry *entry;
926
927 entry = packlist_find(&to_pack, sha1, index_pos);
928 if (!entry)
929 return 0;
930
931 if (exclude) {
932 if (!entry->preferred_base)
933 nr_result--;
934 entry->preferred_base = 1;
935 }
936
937 return 1;
938 }
939
940 /*
941 * Check whether we want the object in the pack (e.g., we do not want
942 * objects found in non-local stores if the "--local" option was used).
943 *
944 * As a side effect of this check, we will find the packed version of this
945 * object, if any. We therefore pass out the pack information to avoid having
946 * to look it up again later.
947 */
948 static int want_object_in_pack(const unsigned char *sha1,
949 int exclude,
950 struct packed_git **found_pack,
951 off_t *found_offset)
952 {
953 struct packed_git *p;
954
955 if (!exclude && local && has_loose_object_nonlocal(sha1))
956 return 0;
957
958 *found_pack = NULL;
959 *found_offset = 0;
960
961 for (p = packed_git; p; p = p->next) {
962 off_t offset = find_pack_entry_one(sha1, p);
963 if (offset) {
964 if (!*found_pack) {
965 if (!is_pack_valid(p))
966 continue;
967 *found_offset = offset;
968 *found_pack = p;
969 }
970 if (exclude)
971 return 1;
972 if (incremental)
973 return 0;
974 if (local && !p->pack_local)
975 return 0;
976 if (ignore_packed_keep && p->pack_local && p->pack_keep)
977 return 0;
978 }
979 }
980
981 return 1;
982 }
983
984 static void create_object_entry(const unsigned char *sha1,
985 enum object_type type,
986 uint32_t hash,
987 int exclude,
988 int no_try_delta,
989 uint32_t index_pos,
990 struct packed_git *found_pack,
991 off_t found_offset)
992 {
993 struct object_entry *entry;
994
995 entry = packlist_alloc(&to_pack, sha1, index_pos);
996 entry->hash = hash;
997 if (type)
998 entry->type = type;
999 if (exclude)
1000 entry->preferred_base = 1;
1001 else
1002 nr_result++;
1003 if (found_pack) {
1004 entry->in_pack = found_pack;
1005 entry->in_pack_offset = found_offset;
1006 }
1007
1008 entry->no_try_delta = no_try_delta;
1009 }
1010
1011 static const char no_closure_warning[] = N_(
1012 "disabling bitmap writing, as some objects are not being packed"
1013 );
1014
1015 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1016 const char *name, int exclude)
1017 {
1018 struct packed_git *found_pack;
1019 off_t found_offset;
1020 uint32_t index_pos;
1021
1022 if (have_duplicate_entry(sha1, exclude, &index_pos))
1023 return 0;
1024
1025 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1026 /* The pack is missing an object, so it will not have closure */
1027 if (write_bitmap_index) {
1028 warning(_(no_closure_warning));
1029 write_bitmap_index = 0;
1030 }
1031 return 0;
1032 }
1033
1034 create_object_entry(sha1, type, pack_name_hash(name),
1035 exclude, name && no_try_delta(name),
1036 index_pos, found_pack, found_offset);
1037
1038 display_progress(progress_state, nr_result);
1039 return 1;
1040 }
1041
1042 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1043 enum object_type type,
1044 int flags, uint32_t name_hash,
1045 struct packed_git *pack, off_t offset)
1046 {
1047 uint32_t index_pos;
1048
1049 if (have_duplicate_entry(sha1, 0, &index_pos))
1050 return 0;
1051
1052 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1053
1054 display_progress(progress_state, nr_result);
1055 return 1;
1056 }
1057
1058 struct pbase_tree_cache {
1059 unsigned char sha1[20];
1060 int ref;
1061 int temporary;
1062 void *tree_data;
1063 unsigned long tree_size;
1064 };
1065
1066 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1067 static int pbase_tree_cache_ix(const unsigned char *sha1)
1068 {
1069 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1070 }
1071 static int pbase_tree_cache_ix_incr(int ix)
1072 {
1073 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1074 }
1075
1076 static struct pbase_tree {
1077 struct pbase_tree *next;
1078 /* This is a phony "cache" entry; we are not
1079 * going to evict it or find it through _get()
1080 * mechanism -- this is for the toplevel node that
1081 * would almost always change with any commit.
1082 */
1083 struct pbase_tree_cache pcache;
1084 } *pbase_tree;
1085
1086 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1087 {
1088 struct pbase_tree_cache *ent, *nent;
1089 void *data;
1090 unsigned long size;
1091 enum object_type type;
1092 int neigh;
1093 int my_ix = pbase_tree_cache_ix(sha1);
1094 int available_ix = -1;
1095
1096 /* pbase-tree-cache acts as a limited hashtable.
1097 * your object will be found at your index or within a few
1098 * slots after that slot if it is cached.
1099 */
1100 for (neigh = 0; neigh < 8; neigh++) {
1101 ent = pbase_tree_cache[my_ix];
1102 if (ent && !hashcmp(ent->sha1, sha1)) {
1103 ent->ref++;
1104 return ent;
1105 }
1106 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1107 ((0 <= available_ix) &&
1108 (!ent && pbase_tree_cache[available_ix])))
1109 available_ix = my_ix;
1110 if (!ent)
1111 break;
1112 my_ix = pbase_tree_cache_ix_incr(my_ix);
1113 }
1114
1115 /* Did not find one. Either we got a bogus request or
1116 * we need to read and perhaps cache.
1117 */
1118 data = read_sha1_file(sha1, &type, &size);
1119 if (!data)
1120 return NULL;
1121 if (type != OBJ_TREE) {
1122 free(data);
1123 return NULL;
1124 }
1125
1126 /* We need to either cache or return a throwaway copy */
1127
1128 if (available_ix < 0)
1129 ent = NULL;
1130 else {
1131 ent = pbase_tree_cache[available_ix];
1132 my_ix = available_ix;
1133 }
1134
1135 if (!ent) {
1136 nent = xmalloc(sizeof(*nent));
1137 nent->temporary = (available_ix < 0);
1138 }
1139 else {
1140 /* evict and reuse */
1141 free(ent->tree_data);
1142 nent = ent;
1143 }
1144 hashcpy(nent->sha1, sha1);
1145 nent->tree_data = data;
1146 nent->tree_size = size;
1147 nent->ref = 1;
1148 if (!nent->temporary)
1149 pbase_tree_cache[my_ix] = nent;
1150 return nent;
1151 }
1152
1153 static void pbase_tree_put(struct pbase_tree_cache *cache)
1154 {
1155 if (!cache->temporary) {
1156 cache->ref--;
1157 return;
1158 }
1159 free(cache->tree_data);
1160 free(cache);
1161 }
1162
1163 static int name_cmp_len(const char *name)
1164 {
1165 int i;
1166 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1167 ;
1168 return i;
1169 }
1170
1171 static void add_pbase_object(struct tree_desc *tree,
1172 const char *name,
1173 int cmplen,
1174 const char *fullname)
1175 {
1176 struct name_entry entry;
1177 int cmp;
1178
1179 while (tree_entry(tree,&entry)) {
1180 if (S_ISGITLINK(entry.mode))
1181 continue;
1182 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1183 memcmp(name, entry.path, cmplen);
1184 if (cmp > 0)
1185 continue;
1186 if (cmp < 0)
1187 return;
1188 if (name[cmplen] != '/') {
1189 add_object_entry(entry.sha1,
1190 object_type(entry.mode),
1191 fullname, 1);
1192 return;
1193 }
1194 if (S_ISDIR(entry.mode)) {
1195 struct tree_desc sub;
1196 struct pbase_tree_cache *tree;
1197 const char *down = name+cmplen+1;
1198 int downlen = name_cmp_len(down);
1199
1200 tree = pbase_tree_get(entry.sha1);
1201 if (!tree)
1202 return;
1203 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1204
1205 add_pbase_object(&sub, down, downlen, fullname);
1206 pbase_tree_put(tree);
1207 }
1208 }
1209 }
1210
1211 static unsigned *done_pbase_paths;
1212 static int done_pbase_paths_num;
1213 static int done_pbase_paths_alloc;
1214 static int done_pbase_path_pos(unsigned hash)
1215 {
1216 int lo = 0;
1217 int hi = done_pbase_paths_num;
1218 while (lo < hi) {
1219 int mi = (hi + lo) / 2;
1220 if (done_pbase_paths[mi] == hash)
1221 return mi;
1222 if (done_pbase_paths[mi] < hash)
1223 hi = mi;
1224 else
1225 lo = mi + 1;
1226 }
1227 return -lo-1;
1228 }
1229
1230 static int check_pbase_path(unsigned hash)
1231 {
1232 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1233 if (0 <= pos)
1234 return 1;
1235 pos = -pos - 1;
1236 ALLOC_GROW(done_pbase_paths,
1237 done_pbase_paths_num + 1,
1238 done_pbase_paths_alloc);
1239 done_pbase_paths_num++;
1240 if (pos < done_pbase_paths_num)
1241 memmove(done_pbase_paths + pos + 1,
1242 done_pbase_paths + pos,
1243 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1244 done_pbase_paths[pos] = hash;
1245 return 0;
1246 }
1247
1248 static void add_preferred_base_object(const char *name)
1249 {
1250 struct pbase_tree *it;
1251 int cmplen;
1252 unsigned hash = pack_name_hash(name);
1253
1254 if (!num_preferred_base || check_pbase_path(hash))
1255 return;
1256
1257 cmplen = name_cmp_len(name);
1258 for (it = pbase_tree; it; it = it->next) {
1259 if (cmplen == 0) {
1260 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1261 }
1262 else {
1263 struct tree_desc tree;
1264 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1265 add_pbase_object(&tree, name, cmplen, name);
1266 }
1267 }
1268 }
1269
1270 static void add_preferred_base(unsigned char *sha1)
1271 {
1272 struct pbase_tree *it;
1273 void *data;
1274 unsigned long size;
1275 unsigned char tree_sha1[20];
1276
1277 if (window <= num_preferred_base++)
1278 return;
1279
1280 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1281 if (!data)
1282 return;
1283
1284 for (it = pbase_tree; it; it = it->next) {
1285 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1286 free(data);
1287 return;
1288 }
1289 }
1290
1291 it = xcalloc(1, sizeof(*it));
1292 it->next = pbase_tree;
1293 pbase_tree = it;
1294
1295 hashcpy(it->pcache.sha1, tree_sha1);
1296 it->pcache.tree_data = data;
1297 it->pcache.tree_size = size;
1298 }
1299
1300 static void cleanup_preferred_base(void)
1301 {
1302 struct pbase_tree *it;
1303 unsigned i;
1304
1305 it = pbase_tree;
1306 pbase_tree = NULL;
1307 while (it) {
1308 struct pbase_tree *this = it;
1309 it = this->next;
1310 free(this->pcache.tree_data);
1311 free(this);
1312 }
1313
1314 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1315 if (!pbase_tree_cache[i])
1316 continue;
1317 free(pbase_tree_cache[i]->tree_data);
1318 free(pbase_tree_cache[i]);
1319 pbase_tree_cache[i] = NULL;
1320 }
1321
1322 free(done_pbase_paths);
1323 done_pbase_paths = NULL;
1324 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1325 }
1326
1327 static void check_object(struct object_entry *entry)
1328 {
1329 if (entry->in_pack) {
1330 struct packed_git *p = entry->in_pack;
1331 struct pack_window *w_curs = NULL;
1332 const unsigned char *base_ref = NULL;
1333 struct object_entry *base_entry;
1334 unsigned long used, used_0;
1335 unsigned long avail;
1336 off_t ofs;
1337 unsigned char *buf, c;
1338
1339 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1340
1341 /*
1342 * We want in_pack_type even if we do not reuse delta
1343 * since non-delta representations could still be reused.
1344 */
1345 used = unpack_object_header_buffer(buf, avail,
1346 &entry->in_pack_type,
1347 &entry->size);
1348 if (used == 0)
1349 goto give_up;
1350
1351 /*
1352 * Determine if this is a delta and if so whether we can
1353 * reuse it or not. Otherwise let's find out as cheaply as
1354 * possible what the actual type and size for this object is.
1355 */
1356 switch (entry->in_pack_type) {
1357 default:
1358 /* Not a delta hence we've already got all we need. */
1359 entry->type = entry->in_pack_type;
1360 entry->in_pack_header_size = used;
1361 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1362 goto give_up;
1363 unuse_pack(&w_curs);
1364 return;
1365 case OBJ_REF_DELTA:
1366 if (reuse_delta && !entry->preferred_base)
1367 base_ref = use_pack(p, &w_curs,
1368 entry->in_pack_offset + used, NULL);
1369 entry->in_pack_header_size = used + 20;
1370 break;
1371 case OBJ_OFS_DELTA:
1372 buf = use_pack(p, &w_curs,
1373 entry->in_pack_offset + used, NULL);
1374 used_0 = 0;
1375 c = buf[used_0++];
1376 ofs = c & 127;
1377 while (c & 128) {
1378 ofs += 1;
1379 if (!ofs || MSB(ofs, 7)) {
1380 error("delta base offset overflow in pack for %s",
1381 sha1_to_hex(entry->idx.sha1));
1382 goto give_up;
1383 }
1384 c = buf[used_0++];
1385 ofs = (ofs << 7) + (c & 127);
1386 }
1387 ofs = entry->in_pack_offset - ofs;
1388 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1389 error("delta base offset out of bound for %s",
1390 sha1_to_hex(entry->idx.sha1));
1391 goto give_up;
1392 }
1393 if (reuse_delta && !entry->preferred_base) {
1394 struct revindex_entry *revidx;
1395 revidx = find_pack_revindex(p, ofs);
1396 if (!revidx)
1397 goto give_up;
1398 base_ref = nth_packed_object_sha1(p, revidx->nr);
1399 }
1400 entry->in_pack_header_size = used + used_0;
1401 break;
1402 }
1403
1404 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1405 /*
1406 * If base_ref was set above that means we wish to
1407 * reuse delta data, and we even found that base
1408 * in the list of objects we want to pack. Goodie!
1409 *
1410 * Depth value does not matter - find_deltas() will
1411 * never consider reused delta as the base object to
1412 * deltify other objects against, in order to avoid
1413 * circular deltas.
1414 */
1415 entry->type = entry->in_pack_type;
1416 entry->delta = base_entry;
1417 entry->delta_size = entry->size;
1418 entry->delta_sibling = base_entry->delta_child;
1419 base_entry->delta_child = entry;
1420 unuse_pack(&w_curs);
1421 return;
1422 }
1423
1424 if (entry->type) {
1425 /*
1426 * This must be a delta and we already know what the
1427 * final object type is. Let's extract the actual
1428 * object size from the delta header.
1429 */
1430 entry->size = get_size_from_delta(p, &w_curs,
1431 entry->in_pack_offset + entry->in_pack_header_size);
1432 if (entry->size == 0)
1433 goto give_up;
1434 unuse_pack(&w_curs);
1435 return;
1436 }
1437
1438 /*
1439 * No choice but to fall back to the recursive delta walk
1440 * with sha1_object_info() to find about the object type
1441 * at this point...
1442 */
1443 give_up:
1444 unuse_pack(&w_curs);
1445 }
1446
1447 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1448 /*
1449 * The error condition is checked in prepare_pack(). This is
1450 * to permit a missing preferred base object to be ignored
1451 * as a preferred base. Doing so can result in a larger
1452 * pack file, but the transfer will still take place.
1453 */
1454 }
1455
1456 static int pack_offset_sort(const void *_a, const void *_b)
1457 {
1458 const struct object_entry *a = *(struct object_entry **)_a;
1459 const struct object_entry *b = *(struct object_entry **)_b;
1460
1461 /* avoid filesystem trashing with loose objects */
1462 if (!a->in_pack && !b->in_pack)
1463 return hashcmp(a->idx.sha1, b->idx.sha1);
1464
1465 if (a->in_pack < b->in_pack)
1466 return -1;
1467 if (a->in_pack > b->in_pack)
1468 return 1;
1469 return a->in_pack_offset < b->in_pack_offset ? -1 :
1470 (a->in_pack_offset > b->in_pack_offset);
1471 }
1472
1473 static void get_object_details(void)
1474 {
1475 uint32_t i;
1476 struct object_entry **sorted_by_offset;
1477
1478 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1479 for (i = 0; i < to_pack.nr_objects; i++)
1480 sorted_by_offset[i] = to_pack.objects + i;
1481 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1482
1483 for (i = 0; i < to_pack.nr_objects; i++) {
1484 struct object_entry *entry = sorted_by_offset[i];
1485 check_object(entry);
1486 if (big_file_threshold < entry->size)
1487 entry->no_try_delta = 1;
1488 }
1489
1490 free(sorted_by_offset);
1491 }
1492
1493 /*
1494 * We search for deltas in a list sorted by type, by filename hash, and then
1495 * by size, so that we see progressively smaller and smaller files.
1496 * That's because we prefer deltas to be from the bigger file
1497 * to the smaller -- deletes are potentially cheaper, but perhaps
1498 * more importantly, the bigger file is likely the more recent
1499 * one. The deepest deltas are therefore the oldest objects which are
1500 * less susceptible to be accessed often.
1501 */
1502 static int type_size_sort(const void *_a, const void *_b)
1503 {
1504 const struct object_entry *a = *(struct object_entry **)_a;
1505 const struct object_entry *b = *(struct object_entry **)_b;
1506
1507 if (a->type > b->type)
1508 return -1;
1509 if (a->type < b->type)
1510 return 1;
1511 if (a->hash > b->hash)
1512 return -1;
1513 if (a->hash < b->hash)
1514 return 1;
1515 if (a->preferred_base > b->preferred_base)
1516 return -1;
1517 if (a->preferred_base < b->preferred_base)
1518 return 1;
1519 if (a->size > b->size)
1520 return -1;
1521 if (a->size < b->size)
1522 return 1;
1523 return a < b ? -1 : (a > b); /* newest first */
1524 }
1525
1526 struct unpacked {
1527 struct object_entry *entry;
1528 void *data;
1529 struct delta_index *index;
1530 unsigned depth;
1531 };
1532
1533 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1534 unsigned long delta_size)
1535 {
1536 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1537 return 0;
1538
1539 if (delta_size < cache_max_small_delta_size)
1540 return 1;
1541
1542 /* cache delta, if objects are large enough compared to delta size */
1543 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1544 return 1;
1545
1546 return 0;
1547 }
1548
1549 #ifndef NO_PTHREADS
1550
1551 static pthread_mutex_t read_mutex;
1552 #define read_lock() pthread_mutex_lock(&read_mutex)
1553 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1554
1555 static pthread_mutex_t cache_mutex;
1556 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1557 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1558
1559 static pthread_mutex_t progress_mutex;
1560 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1561 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1562
1563 #else
1564
1565 #define read_lock() (void)0
1566 #define read_unlock() (void)0
1567 #define cache_lock() (void)0
1568 #define cache_unlock() (void)0
1569 #define progress_lock() (void)0
1570 #define progress_unlock() (void)0
1571
1572 #endif
1573
1574 static int try_delta(struct unpacked *trg, struct unpacked *src,
1575 unsigned max_depth, unsigned long *mem_usage)
1576 {
1577 struct object_entry *trg_entry = trg->entry;
1578 struct object_entry *src_entry = src->entry;
1579 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1580 unsigned ref_depth;
1581 enum object_type type;
1582 void *delta_buf;
1583
1584 /* Don't bother doing diffs between different types */
1585 if (trg_entry->type != src_entry->type)
1586 return -1;
1587
1588 /*
1589 * We do not bother to try a delta that we discarded on an
1590 * earlier try, but only when reusing delta data. Note that
1591 * src_entry that is marked as the preferred_base should always
1592 * be considered, as even if we produce a suboptimal delta against
1593 * it, we will still save the transfer cost, as we already know
1594 * the other side has it and we won't send src_entry at all.
1595 */
1596 if (reuse_delta && trg_entry->in_pack &&
1597 trg_entry->in_pack == src_entry->in_pack &&
1598 !src_entry->preferred_base &&
1599 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1600 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1601 return 0;
1602
1603 /* Let's not bust the allowed depth. */
1604 if (src->depth >= max_depth)
1605 return 0;
1606
1607 /* Now some size filtering heuristics. */
1608 trg_size = trg_entry->size;
1609 if (!trg_entry->delta) {
1610 max_size = trg_size/2 - 20;
1611 ref_depth = 1;
1612 } else {
1613 max_size = trg_entry->delta_size;
1614 ref_depth = trg->depth;
1615 }
1616 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1617 (max_depth - ref_depth + 1);
1618 if (max_size == 0)
1619 return 0;
1620 src_size = src_entry->size;
1621 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1622 if (sizediff >= max_size)
1623 return 0;
1624 if (trg_size < src_size / 32)
1625 return 0;
1626
1627 /* Load data if not already done */
1628 if (!trg->data) {
1629 read_lock();
1630 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1631 read_unlock();
1632 if (!trg->data)
1633 die("object %s cannot be read",
1634 sha1_to_hex(trg_entry->idx.sha1));
1635 if (sz != trg_size)
1636 die("object %s inconsistent object length (%lu vs %lu)",
1637 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1638 *mem_usage += sz;
1639 }
1640 if (!src->data) {
1641 read_lock();
1642 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1643 read_unlock();
1644 if (!src->data) {
1645 if (src_entry->preferred_base) {
1646 static int warned = 0;
1647 if (!warned++)
1648 warning("object %s cannot be read",
1649 sha1_to_hex(src_entry->idx.sha1));
1650 /*
1651 * Those objects are not included in the
1652 * resulting pack. Be resilient and ignore
1653 * them if they can't be read, in case the
1654 * pack could be created nevertheless.
1655 */
1656 return 0;
1657 }
1658 die("object %s cannot be read",
1659 sha1_to_hex(src_entry->idx.sha1));
1660 }
1661 if (sz != src_size)
1662 die("object %s inconsistent object length (%lu vs %lu)",
1663 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1664 *mem_usage += sz;
1665 }
1666 if (!src->index) {
1667 src->index = create_delta_index(src->data, src_size);
1668 if (!src->index) {
1669 static int warned = 0;
1670 if (!warned++)
1671 warning("suboptimal pack - out of memory");
1672 return 0;
1673 }
1674 *mem_usage += sizeof_delta_index(src->index);
1675 }
1676
1677 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1678 if (!delta_buf)
1679 return 0;
1680
1681 if (trg_entry->delta) {
1682 /* Prefer only shallower same-sized deltas. */
1683 if (delta_size == trg_entry->delta_size &&
1684 src->depth + 1 >= trg->depth) {
1685 free(delta_buf);
1686 return 0;
1687 }
1688 }
1689
1690 /*
1691 * Handle memory allocation outside of the cache
1692 * accounting lock. Compiler will optimize the strangeness
1693 * away when NO_PTHREADS is defined.
1694 */
1695 free(trg_entry->delta_data);
1696 cache_lock();
1697 if (trg_entry->delta_data) {
1698 delta_cache_size -= trg_entry->delta_size;
1699 trg_entry->delta_data = NULL;
1700 }
1701 if (delta_cacheable(src_size, trg_size, delta_size)) {
1702 delta_cache_size += delta_size;
1703 cache_unlock();
1704 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1705 } else {
1706 cache_unlock();
1707 free(delta_buf);
1708 }
1709
1710 trg_entry->delta = src_entry;
1711 trg_entry->delta_size = delta_size;
1712 trg->depth = src->depth + 1;
1713
1714 return 1;
1715 }
1716
1717 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1718 {
1719 struct object_entry *child = me->delta_child;
1720 unsigned int m = n;
1721 while (child) {
1722 unsigned int c = check_delta_limit(child, n + 1);
1723 if (m < c)
1724 m = c;
1725 child = child->delta_sibling;
1726 }
1727 return m;
1728 }
1729
1730 static unsigned long free_unpacked(struct unpacked *n)
1731 {
1732 unsigned long freed_mem = sizeof_delta_index(n->index);
1733 free_delta_index(n->index);
1734 n->index = NULL;
1735 if (n->data) {
1736 freed_mem += n->entry->size;
1737 free(n->data);
1738 n->data = NULL;
1739 }
1740 n->entry = NULL;
1741 n->depth = 0;
1742 return freed_mem;
1743 }
1744
1745 static void find_deltas(struct object_entry **list, unsigned *list_size,
1746 int window, int depth, unsigned *processed)
1747 {
1748 uint32_t i, idx = 0, count = 0;
1749 struct unpacked *array;
1750 unsigned long mem_usage = 0;
1751
1752 array = xcalloc(window, sizeof(struct unpacked));
1753
1754 for (;;) {
1755 struct object_entry *entry;
1756 struct unpacked *n = array + idx;
1757 int j, max_depth, best_base = -1;
1758
1759 progress_lock();
1760 if (!*list_size) {
1761 progress_unlock();
1762 break;
1763 }
1764 entry = *list++;
1765 (*list_size)--;
1766 if (!entry->preferred_base) {
1767 (*processed)++;
1768 display_progress(progress_state, *processed);
1769 }
1770 progress_unlock();
1771
1772 mem_usage -= free_unpacked(n);
1773 n->entry = entry;
1774
1775 while (window_memory_limit &&
1776 mem_usage > window_memory_limit &&
1777 count > 1) {
1778 uint32_t tail = (idx + window - count) % window;
1779 mem_usage -= free_unpacked(array + tail);
1780 count--;
1781 }
1782
1783 /* We do not compute delta to *create* objects we are not
1784 * going to pack.
1785 */
1786 if (entry->preferred_base)
1787 goto next;
1788
1789 /*
1790 * If the current object is at pack edge, take the depth the
1791 * objects that depend on the current object into account
1792 * otherwise they would become too deep.
1793 */
1794 max_depth = depth;
1795 if (entry->delta_child) {
1796 max_depth -= check_delta_limit(entry, 0);
1797 if (max_depth <= 0)
1798 goto next;
1799 }
1800
1801 j = window;
1802 while (--j > 0) {
1803 int ret;
1804 uint32_t other_idx = idx + j;
1805 struct unpacked *m;
1806 if (other_idx >= window)
1807 other_idx -= window;
1808 m = array + other_idx;
1809 if (!m->entry)
1810 break;
1811 ret = try_delta(n, m, max_depth, &mem_usage);
1812 if (ret < 0)
1813 break;
1814 else if (ret > 0)
1815 best_base = other_idx;
1816 }
1817
1818 /*
1819 * If we decided to cache the delta data, then it is best
1820 * to compress it right away. First because we have to do
1821 * it anyway, and doing it here while we're threaded will
1822 * save a lot of time in the non threaded write phase,
1823 * as well as allow for caching more deltas within
1824 * the same cache size limit.
1825 * ...
1826 * But only if not writing to stdout, since in that case
1827 * the network is most likely throttling writes anyway,
1828 * and therefore it is best to go to the write phase ASAP
1829 * instead, as we can afford spending more time compressing
1830 * between writes at that moment.
1831 */
1832 if (entry->delta_data && !pack_to_stdout) {
1833 entry->z_delta_size = do_compress(&entry->delta_data,
1834 entry->delta_size);
1835 cache_lock();
1836 delta_cache_size -= entry->delta_size;
1837 delta_cache_size += entry->z_delta_size;
1838 cache_unlock();
1839 }
1840
1841 /* if we made n a delta, and if n is already at max
1842 * depth, leaving it in the window is pointless. we
1843 * should evict it first.
1844 */
1845 if (entry->delta && max_depth <= n->depth)
1846 continue;
1847
1848 /*
1849 * Move the best delta base up in the window, after the
1850 * currently deltified object, to keep it longer. It will
1851 * be the first base object to be attempted next.
1852 */
1853 if (entry->delta) {
1854 struct unpacked swap = array[best_base];
1855 int dist = (window + idx - best_base) % window;
1856 int dst = best_base;
1857 while (dist--) {
1858 int src = (dst + 1) % window;
1859 array[dst] = array[src];
1860 dst = src;
1861 }
1862 array[dst] = swap;
1863 }
1864
1865 next:
1866 idx++;
1867 if (count + 1 < window)
1868 count++;
1869 if (idx >= window)
1870 idx = 0;
1871 }
1872
1873 for (i = 0; i < window; ++i) {
1874 free_delta_index(array[i].index);
1875 free(array[i].data);
1876 }
1877 free(array);
1878 }
1879
1880 #ifndef NO_PTHREADS
1881
1882 static void try_to_free_from_threads(size_t size)
1883 {
1884 read_lock();
1885 release_pack_memory(size);
1886 read_unlock();
1887 }
1888
1889 static try_to_free_t old_try_to_free_routine;
1890
1891 /*
1892 * The main thread waits on the condition that (at least) one of the workers
1893 * has stopped working (which is indicated in the .working member of
1894 * struct thread_params).
1895 * When a work thread has completed its work, it sets .working to 0 and
1896 * signals the main thread and waits on the condition that .data_ready
1897 * becomes 1.
1898 */
1899
1900 struct thread_params {
1901 pthread_t thread;
1902 struct object_entry **list;
1903 unsigned list_size;
1904 unsigned remaining;
1905 int window;
1906 int depth;
1907 int working;
1908 int data_ready;
1909 pthread_mutex_t mutex;
1910 pthread_cond_t cond;
1911 unsigned *processed;
1912 };
1913
1914 static pthread_cond_t progress_cond;
1915
1916 /*
1917 * Mutex and conditional variable can't be statically-initialized on Windows.
1918 */
1919 static void init_threaded_search(void)
1920 {
1921 init_recursive_mutex(&read_mutex);
1922 pthread_mutex_init(&cache_mutex, NULL);
1923 pthread_mutex_init(&progress_mutex, NULL);
1924 pthread_cond_init(&progress_cond, NULL);
1925 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1926 }
1927
1928 static void cleanup_threaded_search(void)
1929 {
1930 set_try_to_free_routine(old_try_to_free_routine);
1931 pthread_cond_destroy(&progress_cond);
1932 pthread_mutex_destroy(&read_mutex);
1933 pthread_mutex_destroy(&cache_mutex);
1934 pthread_mutex_destroy(&progress_mutex);
1935 }
1936
1937 static void *threaded_find_deltas(void *arg)
1938 {
1939 struct thread_params *me = arg;
1940
1941 while (me->remaining) {
1942 find_deltas(me->list, &me->remaining,
1943 me->window, me->depth, me->processed);
1944
1945 progress_lock();
1946 me->working = 0;
1947 pthread_cond_signal(&progress_cond);
1948 progress_unlock();
1949
1950 /*
1951 * We must not set ->data_ready before we wait on the
1952 * condition because the main thread may have set it to 1
1953 * before we get here. In order to be sure that new
1954 * work is available if we see 1 in ->data_ready, it
1955 * was initialized to 0 before this thread was spawned
1956 * and we reset it to 0 right away.
1957 */
1958 pthread_mutex_lock(&me->mutex);
1959 while (!me->data_ready)
1960 pthread_cond_wait(&me->cond, &me->mutex);
1961 me->data_ready = 0;
1962 pthread_mutex_unlock(&me->mutex);
1963 }
1964 /* leave ->working 1 so that this doesn't get more work assigned */
1965 return NULL;
1966 }
1967
1968 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1969 int window, int depth, unsigned *processed)
1970 {
1971 struct thread_params *p;
1972 int i, ret, active_threads = 0;
1973
1974 init_threaded_search();
1975
1976 if (delta_search_threads <= 1) {
1977 find_deltas(list, &list_size, window, depth, processed);
1978 cleanup_threaded_search();
1979 return;
1980 }
1981 if (progress > pack_to_stdout)
1982 fprintf(stderr, "Delta compression using up to %d threads.\n",
1983 delta_search_threads);
1984 p = xcalloc(delta_search_threads, sizeof(*p));
1985
1986 /* Partition the work amongst work threads. */
1987 for (i = 0; i < delta_search_threads; i++) {
1988 unsigned sub_size = list_size / (delta_search_threads - i);
1989
1990 /* don't use too small segments or no deltas will be found */
1991 if (sub_size < 2*window && i+1 < delta_search_threads)
1992 sub_size = 0;
1993
1994 p[i].window = window;
1995 p[i].depth = depth;
1996 p[i].processed = processed;
1997 p[i].working = 1;
1998 p[i].data_ready = 0;
1999
2000 /* try to split chunks on "path" boundaries */
2001 while (sub_size && sub_size < list_size &&
2002 list[sub_size]->hash &&
2003 list[sub_size]->hash == list[sub_size-1]->hash)
2004 sub_size++;
2005
2006 p[i].list = list;
2007 p[i].list_size = sub_size;
2008 p[i].remaining = sub_size;
2009
2010 list += sub_size;
2011 list_size -= sub_size;
2012 }
2013
2014 /* Start work threads. */
2015 for (i = 0; i < delta_search_threads; i++) {
2016 if (!p[i].list_size)
2017 continue;
2018 pthread_mutex_init(&p[i].mutex, NULL);
2019 pthread_cond_init(&p[i].cond, NULL);
2020 ret = pthread_create(&p[i].thread, NULL,
2021 threaded_find_deltas, &p[i]);
2022 if (ret)
2023 die("unable to create thread: %s", strerror(ret));
2024 active_threads++;
2025 }
2026
2027 /*
2028 * Now let's wait for work completion. Each time a thread is done
2029 * with its work, we steal half of the remaining work from the
2030 * thread with the largest number of unprocessed objects and give
2031 * it to that newly idle thread. This ensure good load balancing
2032 * until the remaining object list segments are simply too short
2033 * to be worth splitting anymore.
2034 */
2035 while (active_threads) {
2036 struct thread_params *target = NULL;
2037 struct thread_params *victim = NULL;
2038 unsigned sub_size = 0;
2039
2040 progress_lock();
2041 for (;;) {
2042 for (i = 0; !target && i < delta_search_threads; i++)
2043 if (!p[i].working)
2044 target = &p[i];
2045 if (target)
2046 break;
2047 pthread_cond_wait(&progress_cond, &progress_mutex);
2048 }
2049
2050 for (i = 0; i < delta_search_threads; i++)
2051 if (p[i].remaining > 2*window &&
2052 (!victim || victim->remaining < p[i].remaining))
2053 victim = &p[i];
2054 if (victim) {
2055 sub_size = victim->remaining / 2;
2056 list = victim->list + victim->list_size - sub_size;
2057 while (sub_size && list[0]->hash &&
2058 list[0]->hash == list[-1]->hash) {
2059 list++;
2060 sub_size--;
2061 }
2062 if (!sub_size) {
2063 /*
2064 * It is possible for some "paths" to have
2065 * so many objects that no hash boundary
2066 * might be found. Let's just steal the
2067 * exact half in that case.
2068 */
2069 sub_size = victim->remaining / 2;
2070 list -= sub_size;
2071 }
2072 target->list = list;
2073 victim->list_size -= sub_size;
2074 victim->remaining -= sub_size;
2075 }
2076 target->list_size = sub_size;
2077 target->remaining = sub_size;
2078 target->working = 1;
2079 progress_unlock();
2080
2081 pthread_mutex_lock(&target->mutex);
2082 target->data_ready = 1;
2083 pthread_cond_signal(&target->cond);
2084 pthread_mutex_unlock(&target->mutex);
2085
2086 if (!sub_size) {
2087 pthread_join(target->thread, NULL);
2088 pthread_cond_destroy(&target->cond);
2089 pthread_mutex_destroy(&target->mutex);
2090 active_threads--;
2091 }
2092 }
2093 cleanup_threaded_search();
2094 free(p);
2095 }
2096
2097 #else
2098 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2099 #endif
2100
2101 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2102 {
2103 struct object_id peeled;
2104
2105 if (starts_with(path, "refs/tags/") && /* is a tag? */
2106 !peel_ref(path, peeled.hash) && /* peelable? */
2107 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2108 add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2109 return 0;
2110 }
2111
2112 static void prepare_pack(int window, int depth)
2113 {
2114 struct object_entry **delta_list;
2115 uint32_t i, nr_deltas;
2116 unsigned n;
2117
2118 get_object_details();
2119
2120 /*
2121 * If we're locally repacking then we need to be doubly careful
2122 * from now on in order to make sure no stealth corruption gets
2123 * propagated to the new pack. Clients receiving streamed packs
2124 * should validate everything they get anyway so no need to incur
2125 * the additional cost here in that case.
2126 */
2127 if (!pack_to_stdout)
2128 do_check_packed_object_crc = 1;
2129
2130 if (!to_pack.nr_objects || !window || !depth)
2131 return;
2132
2133 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2134 nr_deltas = n = 0;
2135
2136 for (i = 0; i < to_pack.nr_objects; i++) {
2137 struct object_entry *entry = to_pack.objects + i;
2138
2139 if (entry->delta)
2140 /* This happens if we decided to reuse existing
2141 * delta from a pack. "reuse_delta &&" is implied.
2142 */
2143 continue;
2144
2145 if (entry->size < 50)
2146 continue;
2147
2148 if (entry->no_try_delta)
2149 continue;
2150
2151 if (!entry->preferred_base) {
2152 nr_deltas++;
2153 if (entry->type < 0)
2154 die("unable to get type of object %s",
2155 sha1_to_hex(entry->idx.sha1));
2156 } else {
2157 if (entry->type < 0) {
2158 /*
2159 * This object is not found, but we
2160 * don't have to include it anyway.
2161 */
2162 continue;
2163 }
2164 }
2165
2166 delta_list[n++] = entry;
2167 }
2168
2169 if (nr_deltas && n > 1) {
2170 unsigned nr_done = 0;
2171 if (progress)
2172 progress_state = start_progress(_("Compressing objects"),
2173 nr_deltas);
2174 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2175 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2176 stop_progress(&progress_state);
2177 if (nr_done != nr_deltas)
2178 die("inconsistency with delta count");
2179 }
2180 free(delta_list);
2181 }
2182
2183 static int git_pack_config(const char *k, const char *v, void *cb)
2184 {
2185 if (!strcmp(k, "pack.window")) {
2186 window = git_config_int(k, v);
2187 return 0;
2188 }
2189 if (!strcmp(k, "pack.windowmemory")) {
2190 window_memory_limit = git_config_ulong(k, v);
2191 return 0;
2192 }
2193 if (!strcmp(k, "pack.depth")) {
2194 depth = git_config_int(k, v);
2195 return 0;
2196 }
2197 if (!strcmp(k, "pack.compression")) {
2198 int level = git_config_int(k, v);
2199 if (level == -1)
2200 level = Z_DEFAULT_COMPRESSION;
2201 else if (level < 0 || level > Z_BEST_COMPRESSION)
2202 die("bad pack compression level %d", level);
2203 pack_compression_level = level;
2204 pack_compression_seen = 1;
2205 return 0;
2206 }
2207 if (!strcmp(k, "pack.deltacachesize")) {
2208 max_delta_cache_size = git_config_int(k, v);
2209 return 0;
2210 }
2211 if (!strcmp(k, "pack.deltacachelimit")) {
2212 cache_max_small_delta_size = git_config_int(k, v);
2213 return 0;
2214 }
2215 if (!strcmp(k, "pack.writebitmaphashcache")) {
2216 if (git_config_bool(k, v))
2217 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2218 else
2219 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2220 }
2221 if (!strcmp(k, "pack.usebitmaps")) {
2222 use_bitmap_index = git_config_bool(k, v);
2223 return 0;
2224 }
2225 if (!strcmp(k, "pack.threads")) {
2226 delta_search_threads = git_config_int(k, v);
2227 if (delta_search_threads < 0)
2228 die("invalid number of threads specified (%d)",
2229 delta_search_threads);
2230 #ifdef NO_PTHREADS
2231 if (delta_search_threads != 1)
2232 warning("no threads support, ignoring %s", k);
2233 #endif
2234 return 0;
2235 }
2236 if (!strcmp(k, "pack.indexversion")) {
2237 pack_idx_opts.version = git_config_int(k, v);
2238 if (pack_idx_opts.version > 2)
2239 die("bad pack.indexversion=%"PRIu32,
2240 pack_idx_opts.version);
2241 return 0;
2242 }
2243 return git_default_config(k, v, cb);
2244 }
2245
2246 static void read_object_list_from_stdin(void)
2247 {
2248 char line[40 + 1 + PATH_MAX + 2];
2249 unsigned char sha1[20];
2250
2251 for (;;) {
2252 if (!fgets(line, sizeof(line), stdin)) {
2253 if (feof(stdin))
2254 break;
2255 if (!ferror(stdin))
2256 die("fgets returned NULL, not EOF, not error!");
2257 if (errno != EINTR)
2258 die_errno("fgets");
2259 clearerr(stdin);
2260 continue;
2261 }
2262 if (line[0] == '-') {
2263 if (get_sha1_hex(line+1, sha1))
2264 die("expected edge sha1, got garbage:\n %s",
2265 line);
2266 add_preferred_base(sha1);
2267 continue;
2268 }
2269 if (get_sha1_hex(line, sha1))
2270 die("expected sha1, got garbage:\n %s", line);
2271
2272 add_preferred_base_object(line+41);
2273 add_object_entry(sha1, 0, line+41, 0);
2274 }
2275 }
2276
2277 #define OBJECT_ADDED (1u<<20)
2278
2279 static void show_commit(struct commit *commit, void *data)
2280 {
2281 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2282 commit->object.flags |= OBJECT_ADDED;
2283
2284 if (write_bitmap_index)
2285 index_commit_for_bitmap(commit);
2286 }
2287
2288 static void show_object(struct object *obj,
2289 const struct name_path *path, const char *last,
2290 void *data)
2291 {
2292 char *name = path_name(path, last);
2293
2294 add_preferred_base_object(name);
2295 add_object_entry(obj->oid.hash, obj->type, name, 0);
2296 obj->flags |= OBJECT_ADDED;
2297
2298 /*
2299 * We will have generated the hash from the name,
2300 * but not saved a pointer to it - we can free it
2301 */
2302 free((char *)name);
2303 }
2304
2305 static void show_edge(struct commit *commit)
2306 {
2307 add_preferred_base(commit->object.oid.hash);
2308 }
2309
2310 struct in_pack_object {
2311 off_t offset;
2312 struct object *object;
2313 };
2314
2315 struct in_pack {
2316 int alloc;
2317 int nr;
2318 struct in_pack_object *array;
2319 };
2320
2321 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2322 {
2323 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2324 in_pack->array[in_pack->nr].object = object;
2325 in_pack->nr++;
2326 }
2327
2328 /*
2329 * Compare the objects in the offset order, in order to emulate the
2330 * "git rev-list --objects" output that produced the pack originally.
2331 */
2332 static int ofscmp(const void *a_, const void *b_)
2333 {
2334 struct in_pack_object *a = (struct in_pack_object *)a_;
2335 struct in_pack_object *b = (struct in_pack_object *)b_;
2336
2337 if (a->offset < b->offset)
2338 return -1;
2339 else if (a->offset > b->offset)
2340 return 1;
2341 else
2342 return oidcmp(&a->object->oid, &b->object->oid);
2343 }
2344
2345 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2346 {
2347 struct packed_git *p;
2348 struct in_pack in_pack;
2349 uint32_t i;
2350
2351 memset(&in_pack, 0, sizeof(in_pack));
2352
2353 for (p = packed_git; p; p = p->next) {
2354 const unsigned char *sha1;
2355 struct object *o;
2356
2357 if (!p->pack_local || p->pack_keep)
2358 continue;
2359 if (open_pack_index(p))
2360 die("cannot open pack index");
2361
2362 ALLOC_GROW(in_pack.array,
2363 in_pack.nr + p->num_objects,
2364 in_pack.alloc);
2365
2366 for (i = 0; i < p->num_objects; i++) {
2367 sha1 = nth_packed_object_sha1(p, i);
2368 o = lookup_unknown_object(sha1);
2369 if (!(o->flags & OBJECT_ADDED))
2370 mark_in_pack_object(o, p, &in_pack);
2371 o->flags |= OBJECT_ADDED;
2372 }
2373 }
2374
2375 if (in_pack.nr) {
2376 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2377 ofscmp);
2378 for (i = 0; i < in_pack.nr; i++) {
2379 struct object *o = in_pack.array[i].object;
2380 add_object_entry(o->oid.hash, o->type, "", 0);
2381 }
2382 }
2383 free(in_pack.array);
2384 }
2385
2386 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2387 {
2388 static struct packed_git *last_found = (void *)1;
2389 struct packed_git *p;
2390
2391 p = (last_found != (void *)1) ? last_found : packed_git;
2392
2393 while (p) {
2394 if ((!p->pack_local || p->pack_keep) &&
2395 find_pack_entry_one(sha1, p)) {
2396 last_found = p;
2397 return 1;
2398 }
2399 if (p == last_found)
2400 p = packed_git;
2401 else
2402 p = p->next;
2403 if (p == last_found)
2404 p = p->next;
2405 }
2406 return 0;
2407 }
2408
2409 /*
2410 * Store a list of sha1s that are should not be discarded
2411 * because they are either written too recently, or are
2412 * reachable from another object that was.
2413 *
2414 * This is filled by get_object_list.
2415 */
2416 static struct sha1_array recent_objects;
2417
2418 static int loosened_object_can_be_discarded(const unsigned char *sha1,
2419 unsigned long mtime)
2420 {
2421 if (!unpack_unreachable_expiration)
2422 return 0;
2423 if (mtime > unpack_unreachable_expiration)
2424 return 0;
2425 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2426 return 0;
2427 return 1;
2428 }
2429
2430 static void loosen_unused_packed_objects(struct rev_info *revs)
2431 {
2432 struct packed_git *p;
2433 uint32_t i;
2434 const unsigned char *sha1;
2435
2436 for (p = packed_git; p; p = p->next) {
2437 if (!p->pack_local || p->pack_keep)
2438 continue;
2439
2440 if (open_pack_index(p))
2441 die("cannot open pack index");
2442
2443 for (i = 0; i < p->num_objects; i++) {
2444 sha1 = nth_packed_object_sha1(p, i);
2445 if (!packlist_find(&to_pack, sha1, NULL) &&
2446 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2447 !loosened_object_can_be_discarded(sha1, p->mtime))
2448 if (force_object_loose(sha1, p->mtime))
2449 die("unable to force loose object");
2450 }
2451 }
2452 }
2453
2454 /*
2455 * This tracks any options which a reader of the pack might
2456 * not understand, and which would therefore prevent blind reuse
2457 * of what we have on disk.
2458 */
2459 static int pack_options_allow_reuse(void)
2460 {
2461 return allow_ofs_delta;
2462 }
2463
2464 static int get_object_list_from_bitmap(struct rev_info *revs)
2465 {
2466 if (prepare_bitmap_walk(revs) < 0)
2467 return -1;
2468
2469 if (pack_options_allow_reuse() &&
2470 !reuse_partial_packfile_from_bitmap(
2471 &reuse_packfile,
2472 &reuse_packfile_objects,
2473 &reuse_packfile_offset)) {
2474 assert(reuse_packfile_objects);
2475 nr_result += reuse_packfile_objects;
2476 display_progress(progress_state, nr_result);
2477 }
2478
2479 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2480 return 0;
2481 }
2482
2483 static void record_recent_object(struct object *obj,
2484 const struct name_path *path,
2485 const char *last,
2486 void *data)
2487 {
2488 sha1_array_append(&recent_objects, obj->oid.hash);
2489 }
2490
2491 static void record_recent_commit(struct commit *commit, void *data)
2492 {
2493 sha1_array_append(&recent_objects, commit->object.oid.hash);
2494 }
2495
2496 static void get_object_list(int ac, const char **av)
2497 {
2498 struct rev_info revs;
2499 char line[1000];
2500 int flags = 0;
2501
2502 init_revisions(&revs, NULL);
2503 save_commit_buffer = 0;
2504 setup_revisions(ac, av, &revs, NULL);
2505
2506 /* make sure shallows are read */
2507 is_repository_shallow();
2508
2509 while (fgets(line, sizeof(line), stdin) != NULL) {
2510 int len = strlen(line);
2511 if (len && line[len - 1] == '\n')
2512 line[--len] = 0;
2513 if (!len)
2514 break;
2515 if (*line == '-') {
2516 if (!strcmp(line, "--not")) {
2517 flags ^= UNINTERESTING;
2518 write_bitmap_index = 0;
2519 continue;
2520 }
2521 if (starts_with(line, "--shallow ")) {
2522 unsigned char sha1[20];
2523 if (get_sha1_hex(line + 10, sha1))
2524 die("not an SHA-1 '%s'", line + 10);
2525 register_shallow(sha1);
2526 use_bitmap_index = 0;
2527 continue;
2528 }
2529 die("not a rev '%s'", line);
2530 }
2531 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2532 die("bad revision '%s'", line);
2533 }
2534
2535 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2536 return;
2537
2538 if (prepare_revision_walk(&revs))
2539 die("revision walk setup failed");
2540 mark_edges_uninteresting(&revs, show_edge);
2541 traverse_commit_list(&revs, show_commit, show_object, NULL);
2542
2543 if (unpack_unreachable_expiration) {
2544 revs.ignore_missing_links = 1;
2545 if (add_unseen_recent_objects_to_traversal(&revs,
2546 unpack_unreachable_expiration))
2547 die("unable to add recent objects");
2548 if (prepare_revision_walk(&revs))
2549 die("revision walk setup failed");
2550 traverse_commit_list(&revs, record_recent_commit,
2551 record_recent_object, NULL);
2552 }
2553
2554 if (keep_unreachable)
2555 add_objects_in_unpacked_packs(&revs);
2556 if (unpack_unreachable)
2557 loosen_unused_packed_objects(&revs);
2558
2559 sha1_array_clear(&recent_objects);
2560 }
2561
2562 static int option_parse_index_version(const struct option *opt,
2563 const char *arg, int unset)
2564 {
2565 char *c;
2566 const char *val = arg;
2567 pack_idx_opts.version = strtoul(val, &c, 10);
2568 if (pack_idx_opts.version > 2)
2569 die(_("unsupported index version %s"), val);
2570 if (*c == ',' && c[1])
2571 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2572 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2573 die(_("bad index version '%s'"), val);
2574 return 0;
2575 }
2576
2577 static int option_parse_unpack_unreachable(const struct option *opt,
2578 const char *arg, int unset)
2579 {
2580 if (unset) {
2581 unpack_unreachable = 0;
2582 unpack_unreachable_expiration = 0;
2583 }
2584 else {
2585 unpack_unreachable = 1;
2586 if (arg)
2587 unpack_unreachable_expiration = approxidate(arg);
2588 }
2589 return 0;
2590 }
2591
2592 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2593 {
2594 int use_internal_rev_list = 0;
2595 int thin = 0;
2596 int shallow = 0;
2597 int all_progress_implied = 0;
2598 struct argv_array rp = ARGV_ARRAY_INIT;
2599 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2600 int rev_list_index = 0;
2601 struct option pack_objects_options[] = {
2602 OPT_SET_INT('q', "quiet", &progress,
2603 N_("do not show progress meter"), 0),
2604 OPT_SET_INT(0, "progress", &progress,
2605 N_("show progress meter"), 1),
2606 OPT_SET_INT(0, "all-progress", &progress,
2607 N_("show progress meter during object writing phase"), 2),
2608 OPT_BOOL(0, "all-progress-implied",
2609 &all_progress_implied,
2610 N_("similar to --all-progress when progress meter is shown")),
2611 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2612 N_("write the pack index file in the specified idx format version"),
2613 0, option_parse_index_version },
2614 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2615 N_("maximum size of each output pack file")),
2616 OPT_BOOL(0, "local", &local,
2617 N_("ignore borrowed objects from alternate object store")),
2618 OPT_BOOL(0, "incremental", &incremental,
2619 N_("ignore packed objects")),
2620 OPT_INTEGER(0, "window", &window,
2621 N_("limit pack window by objects")),
2622 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2623 N_("limit pack window by memory in addition to object limit")),
2624 OPT_INTEGER(0, "depth", &depth,
2625 N_("maximum length of delta chain allowed in the resulting pack")),
2626 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2627 N_("reuse existing deltas")),
2628 OPT_BOOL(0, "reuse-object", &reuse_object,
2629 N_("reuse existing objects")),
2630 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2631 N_("use OFS_DELTA objects")),
2632 OPT_INTEGER(0, "threads", &delta_search_threads,
2633 N_("use threads when searching for best delta matches")),
2634 OPT_BOOL(0, "non-empty", &non_empty,
2635 N_("do not create an empty pack output")),
2636 OPT_BOOL(0, "revs", &use_internal_rev_list,
2637 N_("read revision arguments from standard input")),
2638 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2639 N_("limit the objects to those that are not yet packed"),
2640 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2641 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2642 N_("include objects reachable from any reference"),
2643 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2644 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2645 N_("include objects referred by reflog entries"),
2646 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2647 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2648 N_("include objects referred to by the index"),
2649 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2650 OPT_BOOL(0, "stdout", &pack_to_stdout,
2651 N_("output pack to stdout")),
2652 OPT_BOOL(0, "include-tag", &include_tag,
2653 N_("include tag objects that refer to objects to be packed")),
2654 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2655 N_("keep unreachable objects")),
2656 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2657 N_("unpack unreachable objects newer than <time>"),
2658 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2659 OPT_BOOL(0, "thin", &thin,
2660 N_("create thin packs")),
2661 OPT_BOOL(0, "shallow", &shallow,
2662 N_("create packs suitable for shallow fetches")),
2663 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2664 N_("ignore packs that have companion .keep file")),
2665 OPT_INTEGER(0, "compression", &pack_compression_level,
2666 N_("pack compression level")),
2667 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2668 N_("do not hide commits by grafts"), 0),
2669 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2670 N_("use a bitmap index if available to speed up counting objects")),
2671 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2672 N_("write a bitmap index together with the pack index")),
2673 OPT_END(),
2674 };
2675
2676 check_replace_refs = 0;
2677
2678 reset_pack_idx_option(&pack_idx_opts);
2679 git_config(git_pack_config, NULL);
2680 if (!pack_compression_seen && core_compression_seen)
2681 pack_compression_level = core_compression_level;
2682
2683 progress = isatty(2);
2684 argc = parse_options(argc, argv, prefix, pack_objects_options,
2685 pack_usage, 0);
2686
2687 if (argc) {
2688 base_name = argv[0];
2689 argc--;
2690 }
2691 if (pack_to_stdout != !base_name || argc)
2692 usage_with_options(pack_usage, pack_objects_options);
2693
2694 argv_array_push(&rp, "pack-objects");
2695 if (thin) {
2696 use_internal_rev_list = 1;
2697 argv_array_push(&rp, shallow
2698 ? "--objects-edge-aggressive"
2699 : "--objects-edge");
2700 } else
2701 argv_array_push(&rp, "--objects");
2702
2703 if (rev_list_all) {
2704 use_internal_rev_list = 1;
2705 argv_array_push(&rp, "--all");
2706 }
2707 if (rev_list_reflog) {
2708 use_internal_rev_list = 1;
2709 argv_array_push(&rp, "--reflog");
2710 }
2711 if (rev_list_index) {
2712 use_internal_rev_list = 1;
2713 argv_array_push(&rp, "--indexed-objects");
2714 }
2715 if (rev_list_unpacked) {
2716 use_internal_rev_list = 1;
2717 argv_array_push(&rp, "--unpacked");
2718 }
2719
2720 if (!reuse_object)
2721 reuse_delta = 0;
2722 if (pack_compression_level == -1)
2723 pack_compression_level = Z_DEFAULT_COMPRESSION;
2724 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2725 die("bad pack compression level %d", pack_compression_level);
2726
2727 if (!delta_search_threads) /* --threads=0 means autodetect */
2728 delta_search_threads = online_cpus();
2729
2730 #ifdef NO_PTHREADS
2731 if (delta_search_threads != 1)
2732 warning("no threads support, ignoring --threads");
2733 #endif
2734 if (!pack_to_stdout && !pack_size_limit)
2735 pack_size_limit = pack_size_limit_cfg;
2736 if (pack_to_stdout && pack_size_limit)
2737 die("--max-pack-size cannot be used to build a pack for transfer.");
2738 if (pack_size_limit && pack_size_limit < 1024*1024) {
2739 warning("minimum pack size limit is 1 MiB");
2740 pack_size_limit = 1024*1024;
2741 }
2742
2743 if (!pack_to_stdout && thin)
2744 die("--thin cannot be used to build an indexable pack.");
2745
2746 if (keep_unreachable && unpack_unreachable)
2747 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2748 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2749 unpack_unreachable_expiration = 0;
2750
2751 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2752 use_bitmap_index = 0;
2753
2754 if (pack_to_stdout || !rev_list_all)
2755 write_bitmap_index = 0;
2756
2757 if (progress && all_progress_implied)
2758 progress = 2;
2759
2760 prepare_packed_git();
2761
2762 if (progress)
2763 progress_state = start_progress(_("Counting objects"), 0);
2764 if (!use_internal_rev_list)
2765 read_object_list_from_stdin();
2766 else {
2767 get_object_list(rp.argc, rp.argv);
2768 argv_array_clear(&rp);
2769 }
2770 cleanup_preferred_base();
2771 if (include_tag && nr_result)
2772 for_each_ref(add_ref_tag, NULL);
2773 stop_progress(&progress_state);
2774
2775 if (non_empty && !nr_result)
2776 return 0;
2777 if (nr_result)
2778 prepare_pack(window, depth);
2779 write_pack_file();
2780 if (progress)
2781 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2782 " reused %"PRIu32" (delta %"PRIu32")\n",
2783 written, written_delta, reused, reused_delta);
2784 return 0;
2785 }