pack-objects: move in_pack_pos out of struct object_entry
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "attr.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "tree.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
16 #include "diff.h"
17 #include "revision.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.h"
22 #include "progress.h"
23 #include "refs.h"
24 #include "streaming.h"
25 #include "thread-utils.h"
26 #include "pack-bitmap.h"
27 #include "reachable.h"
28 #include "sha1-array.h"
29 #include "argv-array.h"
30 #include "list.h"
31 #include "packfile.h"
32 #include "object-store.h"
33
34 static const char *pack_usage[] = {
35 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
36 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
37 NULL
38 };
39
40 /*
41 * Objects we are going to pack are collected in the `to_pack` structure.
42 * It contains an array (dynamically expanded) of the object data, and a map
43 * that can resolve SHA1s to their position in the array.
44 */
45 static struct packing_data to_pack;
46
47 static struct pack_idx_entry **written_list;
48 static uint32_t nr_result, nr_written;
49
50 static int non_empty;
51 static int reuse_delta = 1, reuse_object = 1;
52 static int keep_unreachable, unpack_unreachable, include_tag;
53 static timestamp_t unpack_unreachable_expiration;
54 static int pack_loose_unreachable;
55 static int local;
56 static int have_non_local_packs;
57 static int incremental;
58 static int ignore_packed_keep;
59 static int allow_ofs_delta;
60 static struct pack_idx_option pack_idx_opts;
61 static const char *base_name;
62 static int progress = 1;
63 static int window = 10;
64 static unsigned long pack_size_limit;
65 static int depth = 50;
66 static int delta_search_threads;
67 static int pack_to_stdout;
68 static int num_preferred_base;
69 static struct progress *progress_state;
70
71 static struct packed_git *reuse_packfile;
72 static uint32_t reuse_packfile_objects;
73 static off_t reuse_packfile_offset;
74
75 static int use_bitmap_index_default = 1;
76 static int use_bitmap_index = -1;
77 static int write_bitmap_index;
78 static uint16_t write_bitmap_options;
79
80 static int exclude_promisor_objects;
81
82 static unsigned long delta_cache_size = 0;
83 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
84 static unsigned long cache_max_small_delta_size = 1000;
85
86 static unsigned long window_memory_limit = 0;
87
88 static struct list_objects_filter_options filter_options;
89
90 enum missing_action {
91 MA_ERROR = 0, /* fail if any missing objects are encountered */
92 MA_ALLOW_ANY, /* silently allow ALL missing objects */
93 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
94 };
95 static enum missing_action arg_missing_action;
96 static show_object_fn fn_show_object;
97
98 /*
99 * stats
100 */
101 static uint32_t written, written_delta;
102 static uint32_t reused, reused_delta;
103
104 /*
105 * Indexed commits
106 */
107 static struct commit **indexed_commits;
108 static unsigned int indexed_commits_nr;
109 static unsigned int indexed_commits_alloc;
110
111 static void index_commit_for_bitmap(struct commit *commit)
112 {
113 if (indexed_commits_nr >= indexed_commits_alloc) {
114 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
115 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
116 }
117
118 indexed_commits[indexed_commits_nr++] = commit;
119 }
120
121 static void *get_delta(struct object_entry *entry)
122 {
123 unsigned long size, base_size, delta_size;
124 void *buf, *base_buf, *delta_buf;
125 enum object_type type;
126
127 buf = read_object_file(&entry->idx.oid, &type, &size);
128 if (!buf)
129 die("unable to read %s", oid_to_hex(&entry->idx.oid));
130 base_buf = read_object_file(&entry->delta->idx.oid, &type, &base_size);
131 if (!base_buf)
132 die("unable to read %s",
133 oid_to_hex(&entry->delta->idx.oid));
134 delta_buf = diff_delta(base_buf, base_size,
135 buf, size, &delta_size, 0);
136 if (!delta_buf || delta_size != entry->delta_size)
137 die("delta size changed");
138 free(buf);
139 free(base_buf);
140 return delta_buf;
141 }
142
143 static unsigned long do_compress(void **pptr, unsigned long size)
144 {
145 git_zstream stream;
146 void *in, *out;
147 unsigned long maxsize;
148
149 git_deflate_init(&stream, pack_compression_level);
150 maxsize = git_deflate_bound(&stream, size);
151
152 in = *pptr;
153 out = xmalloc(maxsize);
154 *pptr = out;
155
156 stream.next_in = in;
157 stream.avail_in = size;
158 stream.next_out = out;
159 stream.avail_out = maxsize;
160 while (git_deflate(&stream, Z_FINISH) == Z_OK)
161 ; /* nothing */
162 git_deflate_end(&stream);
163
164 free(in);
165 return stream.total_out;
166 }
167
168 static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
169 const struct object_id *oid)
170 {
171 git_zstream stream;
172 unsigned char ibuf[1024 * 16];
173 unsigned char obuf[1024 * 16];
174 unsigned long olen = 0;
175
176 git_deflate_init(&stream, pack_compression_level);
177
178 for (;;) {
179 ssize_t readlen;
180 int zret = Z_OK;
181 readlen = read_istream(st, ibuf, sizeof(ibuf));
182 if (readlen == -1)
183 die(_("unable to read %s"), oid_to_hex(oid));
184
185 stream.next_in = ibuf;
186 stream.avail_in = readlen;
187 while ((stream.avail_in || readlen == 0) &&
188 (zret == Z_OK || zret == Z_BUF_ERROR)) {
189 stream.next_out = obuf;
190 stream.avail_out = sizeof(obuf);
191 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
192 hashwrite(f, obuf, stream.next_out - obuf);
193 olen += stream.next_out - obuf;
194 }
195 if (stream.avail_in)
196 die(_("deflate error (%d)"), zret);
197 if (readlen == 0) {
198 if (zret != Z_STREAM_END)
199 die(_("deflate error (%d)"), zret);
200 break;
201 }
202 }
203 git_deflate_end(&stream);
204 return olen;
205 }
206
207 /*
208 * we are going to reuse the existing object data as is. make
209 * sure it is not corrupt.
210 */
211 static int check_pack_inflate(struct packed_git *p,
212 struct pack_window **w_curs,
213 off_t offset,
214 off_t len,
215 unsigned long expect)
216 {
217 git_zstream stream;
218 unsigned char fakebuf[4096], *in;
219 int st;
220
221 memset(&stream, 0, sizeof(stream));
222 git_inflate_init(&stream);
223 do {
224 in = use_pack(p, w_curs, offset, &stream.avail_in);
225 stream.next_in = in;
226 stream.next_out = fakebuf;
227 stream.avail_out = sizeof(fakebuf);
228 st = git_inflate(&stream, Z_FINISH);
229 offset += stream.next_in - in;
230 } while (st == Z_OK || st == Z_BUF_ERROR);
231 git_inflate_end(&stream);
232 return (st == Z_STREAM_END &&
233 stream.total_out == expect &&
234 stream.total_in == len) ? 0 : -1;
235 }
236
237 static void copy_pack_data(struct hashfile *f,
238 struct packed_git *p,
239 struct pack_window **w_curs,
240 off_t offset,
241 off_t len)
242 {
243 unsigned char *in;
244 unsigned long avail;
245
246 while (len) {
247 in = use_pack(p, w_curs, offset, &avail);
248 if (avail > len)
249 avail = (unsigned long)len;
250 hashwrite(f, in, avail);
251 offset += avail;
252 len -= avail;
253 }
254 }
255
256 /* Return 0 if we will bust the pack-size limit */
257 static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
258 unsigned long limit, int usable_delta)
259 {
260 unsigned long size, datalen;
261 unsigned char header[MAX_PACK_OBJECT_HEADER],
262 dheader[MAX_PACK_OBJECT_HEADER];
263 unsigned hdrlen;
264 enum object_type type;
265 void *buf;
266 struct git_istream *st = NULL;
267
268 if (!usable_delta) {
269 if (oe_type(entry) == OBJ_BLOB &&
270 entry->size > big_file_threshold &&
271 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
272 buf = NULL;
273 else {
274 buf = read_object_file(&entry->idx.oid, &type, &size);
275 if (!buf)
276 die(_("unable to read %s"),
277 oid_to_hex(&entry->idx.oid));
278 }
279 /*
280 * make sure no cached delta data remains from a
281 * previous attempt before a pack split occurred.
282 */
283 FREE_AND_NULL(entry->delta_data);
284 entry->z_delta_size = 0;
285 } else if (entry->delta_data) {
286 size = entry->delta_size;
287 buf = entry->delta_data;
288 entry->delta_data = NULL;
289 type = (allow_ofs_delta && entry->delta->idx.offset) ?
290 OBJ_OFS_DELTA : OBJ_REF_DELTA;
291 } else {
292 buf = get_delta(entry);
293 size = entry->delta_size;
294 type = (allow_ofs_delta && entry->delta->idx.offset) ?
295 OBJ_OFS_DELTA : OBJ_REF_DELTA;
296 }
297
298 if (st) /* large blob case, just assume we don't compress well */
299 datalen = size;
300 else if (entry->z_delta_size)
301 datalen = entry->z_delta_size;
302 else
303 datalen = do_compress(&buf, size);
304
305 /*
306 * The object header is a byte of 'type' followed by zero or
307 * more bytes of length.
308 */
309 hdrlen = encode_in_pack_object_header(header, sizeof(header),
310 type, size);
311
312 if (type == OBJ_OFS_DELTA) {
313 /*
314 * Deltas with relative base contain an additional
315 * encoding of the relative offset for the delta
316 * base from this object's position in the pack.
317 */
318 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
319 unsigned pos = sizeof(dheader) - 1;
320 dheader[pos] = ofs & 127;
321 while (ofs >>= 7)
322 dheader[--pos] = 128 | (--ofs & 127);
323 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
324 if (st)
325 close_istream(st);
326 free(buf);
327 return 0;
328 }
329 hashwrite(f, header, hdrlen);
330 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
331 hdrlen += sizeof(dheader) - pos;
332 } else if (type == OBJ_REF_DELTA) {
333 /*
334 * Deltas with a base reference contain
335 * an additional 20 bytes for the base sha1.
336 */
337 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
338 if (st)
339 close_istream(st);
340 free(buf);
341 return 0;
342 }
343 hashwrite(f, header, hdrlen);
344 hashwrite(f, entry->delta->idx.oid.hash, 20);
345 hdrlen += 20;
346 } else {
347 if (limit && hdrlen + datalen + 20 >= limit) {
348 if (st)
349 close_istream(st);
350 free(buf);
351 return 0;
352 }
353 hashwrite(f, header, hdrlen);
354 }
355 if (st) {
356 datalen = write_large_blob_data(st, f, &entry->idx.oid);
357 close_istream(st);
358 } else {
359 hashwrite(f, buf, datalen);
360 free(buf);
361 }
362
363 return hdrlen + datalen;
364 }
365
366 /* Return 0 if we will bust the pack-size limit */
367 static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
368 unsigned long limit, int usable_delta)
369 {
370 struct packed_git *p = entry->in_pack;
371 struct pack_window *w_curs = NULL;
372 struct revindex_entry *revidx;
373 off_t offset;
374 enum object_type type = oe_type(entry);
375 off_t datalen;
376 unsigned char header[MAX_PACK_OBJECT_HEADER],
377 dheader[MAX_PACK_OBJECT_HEADER];
378 unsigned hdrlen;
379
380 if (entry->delta)
381 type = (allow_ofs_delta && entry->delta->idx.offset) ?
382 OBJ_OFS_DELTA : OBJ_REF_DELTA;
383 hdrlen = encode_in_pack_object_header(header, sizeof(header),
384 type, entry->size);
385
386 offset = entry->in_pack_offset;
387 revidx = find_pack_revindex(p, offset);
388 datalen = revidx[1].offset - offset;
389 if (!pack_to_stdout && p->index_version > 1 &&
390 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
391 error("bad packed object CRC for %s",
392 oid_to_hex(&entry->idx.oid));
393 unuse_pack(&w_curs);
394 return write_no_reuse_object(f, entry, limit, usable_delta);
395 }
396
397 offset += entry->in_pack_header_size;
398 datalen -= entry->in_pack_header_size;
399
400 if (!pack_to_stdout && p->index_version == 1 &&
401 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
402 error("corrupt packed object for %s",
403 oid_to_hex(&entry->idx.oid));
404 unuse_pack(&w_curs);
405 return write_no_reuse_object(f, entry, limit, usable_delta);
406 }
407
408 if (type == OBJ_OFS_DELTA) {
409 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
410 unsigned pos = sizeof(dheader) - 1;
411 dheader[pos] = ofs & 127;
412 while (ofs >>= 7)
413 dheader[--pos] = 128 | (--ofs & 127);
414 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
415 unuse_pack(&w_curs);
416 return 0;
417 }
418 hashwrite(f, header, hdrlen);
419 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
420 hdrlen += sizeof(dheader) - pos;
421 reused_delta++;
422 } else if (type == OBJ_REF_DELTA) {
423 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
424 unuse_pack(&w_curs);
425 return 0;
426 }
427 hashwrite(f, header, hdrlen);
428 hashwrite(f, entry->delta->idx.oid.hash, 20);
429 hdrlen += 20;
430 reused_delta++;
431 } else {
432 if (limit && hdrlen + datalen + 20 >= limit) {
433 unuse_pack(&w_curs);
434 return 0;
435 }
436 hashwrite(f, header, hdrlen);
437 }
438 copy_pack_data(f, p, &w_curs, offset, datalen);
439 unuse_pack(&w_curs);
440 reused++;
441 return hdrlen + datalen;
442 }
443
444 /* Return 0 if we will bust the pack-size limit */
445 static off_t write_object(struct hashfile *f,
446 struct object_entry *entry,
447 off_t write_offset)
448 {
449 unsigned long limit;
450 off_t len;
451 int usable_delta, to_reuse;
452
453 if (!pack_to_stdout)
454 crc32_begin(f);
455
456 /* apply size limit if limited packsize and not first object */
457 if (!pack_size_limit || !nr_written)
458 limit = 0;
459 else if (pack_size_limit <= write_offset)
460 /*
461 * the earlier object did not fit the limit; avoid
462 * mistaking this with unlimited (i.e. limit = 0).
463 */
464 limit = 1;
465 else
466 limit = pack_size_limit - write_offset;
467
468 if (!entry->delta)
469 usable_delta = 0; /* no delta */
470 else if (!pack_size_limit)
471 usable_delta = 1; /* unlimited packfile */
472 else if (entry->delta->idx.offset == (off_t)-1)
473 usable_delta = 0; /* base was written to another pack */
474 else if (entry->delta->idx.offset)
475 usable_delta = 1; /* base already exists in this pack */
476 else
477 usable_delta = 0; /* base could end up in another pack */
478
479 if (!reuse_object)
480 to_reuse = 0; /* explicit */
481 else if (!entry->in_pack)
482 to_reuse = 0; /* can't reuse what we don't have */
483 else if (oe_type(entry) == OBJ_REF_DELTA ||
484 oe_type(entry) == OBJ_OFS_DELTA)
485 /* check_object() decided it for us ... */
486 to_reuse = usable_delta;
487 /* ... but pack split may override that */
488 else if (oe_type(entry) != entry->in_pack_type)
489 to_reuse = 0; /* pack has delta which is unusable */
490 else if (entry->delta)
491 to_reuse = 0; /* we want to pack afresh */
492 else
493 to_reuse = 1; /* we have it in-pack undeltified,
494 * and we do not need to deltify it.
495 */
496
497 if (!to_reuse)
498 len = write_no_reuse_object(f, entry, limit, usable_delta);
499 else
500 len = write_reuse_object(f, entry, limit, usable_delta);
501 if (!len)
502 return 0;
503
504 if (usable_delta)
505 written_delta++;
506 written++;
507 if (!pack_to_stdout)
508 entry->idx.crc32 = crc32_end(f);
509 return len;
510 }
511
512 enum write_one_status {
513 WRITE_ONE_SKIP = -1, /* already written */
514 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
515 WRITE_ONE_WRITTEN = 1, /* normal */
516 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
517 };
518
519 static enum write_one_status write_one(struct hashfile *f,
520 struct object_entry *e,
521 off_t *offset)
522 {
523 off_t size;
524 int recursing;
525
526 /*
527 * we set offset to 1 (which is an impossible value) to mark
528 * the fact that this object is involved in "write its base
529 * first before writing a deltified object" recursion.
530 */
531 recursing = (e->idx.offset == 1);
532 if (recursing) {
533 warning("recursive delta detected for object %s",
534 oid_to_hex(&e->idx.oid));
535 return WRITE_ONE_RECURSIVE;
536 } else if (e->idx.offset || e->preferred_base) {
537 /* offset is non zero if object is written already. */
538 return WRITE_ONE_SKIP;
539 }
540
541 /* if we are deltified, write out base object first. */
542 if (e->delta) {
543 e->idx.offset = 1; /* now recurse */
544 switch (write_one(f, e->delta, offset)) {
545 case WRITE_ONE_RECURSIVE:
546 /* we cannot depend on this one */
547 e->delta = NULL;
548 break;
549 default:
550 break;
551 case WRITE_ONE_BREAK:
552 e->idx.offset = recursing;
553 return WRITE_ONE_BREAK;
554 }
555 }
556
557 e->idx.offset = *offset;
558 size = write_object(f, e, *offset);
559 if (!size) {
560 e->idx.offset = recursing;
561 return WRITE_ONE_BREAK;
562 }
563 written_list[nr_written++] = &e->idx;
564
565 /* make sure off_t is sufficiently large not to wrap */
566 if (signed_add_overflows(*offset, size))
567 die("pack too large for current definition of off_t");
568 *offset += size;
569 return WRITE_ONE_WRITTEN;
570 }
571
572 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
573 void *cb_data)
574 {
575 struct object_id peeled;
576 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
577
578 if (entry)
579 entry->tagged = 1;
580 if (!peel_ref(path, &peeled)) {
581 entry = packlist_find(&to_pack, peeled.hash, NULL);
582 if (entry)
583 entry->tagged = 1;
584 }
585 return 0;
586 }
587
588 static inline void add_to_write_order(struct object_entry **wo,
589 unsigned int *endp,
590 struct object_entry *e)
591 {
592 if (e->filled)
593 return;
594 wo[(*endp)++] = e;
595 e->filled = 1;
596 }
597
598 static void add_descendants_to_write_order(struct object_entry **wo,
599 unsigned int *endp,
600 struct object_entry *e)
601 {
602 int add_to_order = 1;
603 while (e) {
604 if (add_to_order) {
605 struct object_entry *s;
606 /* add this node... */
607 add_to_write_order(wo, endp, e);
608 /* all its siblings... */
609 for (s = e->delta_sibling; s; s = s->delta_sibling) {
610 add_to_write_order(wo, endp, s);
611 }
612 }
613 /* drop down a level to add left subtree nodes if possible */
614 if (e->delta_child) {
615 add_to_order = 1;
616 e = e->delta_child;
617 } else {
618 add_to_order = 0;
619 /* our sibling might have some children, it is next */
620 if (e->delta_sibling) {
621 e = e->delta_sibling;
622 continue;
623 }
624 /* go back to our parent node */
625 e = e->delta;
626 while (e && !e->delta_sibling) {
627 /* we're on the right side of a subtree, keep
628 * going up until we can go right again */
629 e = e->delta;
630 }
631 if (!e) {
632 /* done- we hit our original root node */
633 return;
634 }
635 /* pass it off to sibling at this level */
636 e = e->delta_sibling;
637 }
638 };
639 }
640
641 static void add_family_to_write_order(struct object_entry **wo,
642 unsigned int *endp,
643 struct object_entry *e)
644 {
645 struct object_entry *root;
646
647 for (root = e; root->delta; root = root->delta)
648 ; /* nothing */
649 add_descendants_to_write_order(wo, endp, root);
650 }
651
652 static struct object_entry **compute_write_order(void)
653 {
654 unsigned int i, wo_end, last_untagged;
655
656 struct object_entry **wo;
657 struct object_entry *objects = to_pack.objects;
658
659 for (i = 0; i < to_pack.nr_objects; i++) {
660 objects[i].tagged = 0;
661 objects[i].filled = 0;
662 objects[i].delta_child = NULL;
663 objects[i].delta_sibling = NULL;
664 }
665
666 /*
667 * Fully connect delta_child/delta_sibling network.
668 * Make sure delta_sibling is sorted in the original
669 * recency order.
670 */
671 for (i = to_pack.nr_objects; i > 0;) {
672 struct object_entry *e = &objects[--i];
673 if (!e->delta)
674 continue;
675 /* Mark me as the first child */
676 e->delta_sibling = e->delta->delta_child;
677 e->delta->delta_child = e;
678 }
679
680 /*
681 * Mark objects that are at the tip of tags.
682 */
683 for_each_tag_ref(mark_tagged, NULL);
684
685 /*
686 * Give the objects in the original recency order until
687 * we see a tagged tip.
688 */
689 ALLOC_ARRAY(wo, to_pack.nr_objects);
690 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
691 if (objects[i].tagged)
692 break;
693 add_to_write_order(wo, &wo_end, &objects[i]);
694 }
695 last_untagged = i;
696
697 /*
698 * Then fill all the tagged tips.
699 */
700 for (; i < to_pack.nr_objects; i++) {
701 if (objects[i].tagged)
702 add_to_write_order(wo, &wo_end, &objects[i]);
703 }
704
705 /*
706 * And then all remaining commits and tags.
707 */
708 for (i = last_untagged; i < to_pack.nr_objects; i++) {
709 if (oe_type(&objects[i]) != OBJ_COMMIT &&
710 oe_type(&objects[i]) != OBJ_TAG)
711 continue;
712 add_to_write_order(wo, &wo_end, &objects[i]);
713 }
714
715 /*
716 * And then all the trees.
717 */
718 for (i = last_untagged; i < to_pack.nr_objects; i++) {
719 if (oe_type(&objects[i]) != OBJ_TREE)
720 continue;
721 add_to_write_order(wo, &wo_end, &objects[i]);
722 }
723
724 /*
725 * Finally all the rest in really tight order
726 */
727 for (i = last_untagged; i < to_pack.nr_objects; i++) {
728 if (!objects[i].filled)
729 add_family_to_write_order(wo, &wo_end, &objects[i]);
730 }
731
732 if (wo_end != to_pack.nr_objects)
733 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
734
735 return wo;
736 }
737
738 static off_t write_reused_pack(struct hashfile *f)
739 {
740 unsigned char buffer[8192];
741 off_t to_write, total;
742 int fd;
743
744 if (!is_pack_valid(reuse_packfile))
745 die("packfile is invalid: %s", reuse_packfile->pack_name);
746
747 fd = git_open(reuse_packfile->pack_name);
748 if (fd < 0)
749 die_errno("unable to open packfile for reuse: %s",
750 reuse_packfile->pack_name);
751
752 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
753 die_errno("unable to seek in reused packfile");
754
755 if (reuse_packfile_offset < 0)
756 reuse_packfile_offset = reuse_packfile->pack_size - 20;
757
758 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
759
760 while (to_write) {
761 int read_pack = xread(fd, buffer, sizeof(buffer));
762
763 if (read_pack <= 0)
764 die_errno("unable to read from reused packfile");
765
766 if (read_pack > to_write)
767 read_pack = to_write;
768
769 hashwrite(f, buffer, read_pack);
770 to_write -= read_pack;
771
772 /*
773 * We don't know the actual number of objects written,
774 * only how many bytes written, how many bytes total, and
775 * how many objects total. So we can fake it by pretending all
776 * objects we are writing are the same size. This gives us a
777 * smooth progress meter, and at the end it matches the true
778 * answer.
779 */
780 written = reuse_packfile_objects *
781 (((double)(total - to_write)) / total);
782 display_progress(progress_state, written);
783 }
784
785 close(fd);
786 written = reuse_packfile_objects;
787 display_progress(progress_state, written);
788 return reuse_packfile_offset - sizeof(struct pack_header);
789 }
790
791 static const char no_split_warning[] = N_(
792 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
793 );
794
795 static void write_pack_file(void)
796 {
797 uint32_t i = 0, j;
798 struct hashfile *f;
799 off_t offset;
800 uint32_t nr_remaining = nr_result;
801 time_t last_mtime = 0;
802 struct object_entry **write_order;
803
804 if (progress > pack_to_stdout)
805 progress_state = start_progress(_("Writing objects"), nr_result);
806 ALLOC_ARRAY(written_list, to_pack.nr_objects);
807 write_order = compute_write_order();
808
809 do {
810 struct object_id oid;
811 char *pack_tmp_name = NULL;
812
813 if (pack_to_stdout)
814 f = hashfd_throughput(1, "<stdout>", progress_state);
815 else
816 f = create_tmp_packfile(&pack_tmp_name);
817
818 offset = write_pack_header(f, nr_remaining);
819
820 if (reuse_packfile) {
821 off_t packfile_size;
822 assert(pack_to_stdout);
823
824 packfile_size = write_reused_pack(f);
825 offset += packfile_size;
826 }
827
828 nr_written = 0;
829 for (; i < to_pack.nr_objects; i++) {
830 struct object_entry *e = write_order[i];
831 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
832 break;
833 display_progress(progress_state, written);
834 }
835
836 /*
837 * Did we write the wrong # entries in the header?
838 * If so, rewrite it like in fast-import
839 */
840 if (pack_to_stdout) {
841 hashclose(f, oid.hash, CSUM_CLOSE);
842 } else if (nr_written == nr_remaining) {
843 hashclose(f, oid.hash, CSUM_FSYNC);
844 } else {
845 int fd = hashclose(f, oid.hash, 0);
846 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
847 nr_written, oid.hash, offset);
848 close(fd);
849 if (write_bitmap_index) {
850 warning(_(no_split_warning));
851 write_bitmap_index = 0;
852 }
853 }
854
855 if (!pack_to_stdout) {
856 struct stat st;
857 struct strbuf tmpname = STRBUF_INIT;
858
859 /*
860 * Packs are runtime accessed in their mtime
861 * order since newer packs are more likely to contain
862 * younger objects. So if we are creating multiple
863 * packs then we should modify the mtime of later ones
864 * to preserve this property.
865 */
866 if (stat(pack_tmp_name, &st) < 0) {
867 warning_errno("failed to stat %s", pack_tmp_name);
868 } else if (!last_mtime) {
869 last_mtime = st.st_mtime;
870 } else {
871 struct utimbuf utb;
872 utb.actime = st.st_atime;
873 utb.modtime = --last_mtime;
874 if (utime(pack_tmp_name, &utb) < 0)
875 warning_errno("failed utime() on %s", pack_tmp_name);
876 }
877
878 strbuf_addf(&tmpname, "%s-", base_name);
879
880 if (write_bitmap_index) {
881 bitmap_writer_set_checksum(oid.hash);
882 bitmap_writer_build_type_index(
883 &to_pack, written_list, nr_written);
884 }
885
886 finish_tmp_packfile(&tmpname, pack_tmp_name,
887 written_list, nr_written,
888 &pack_idx_opts, oid.hash);
889
890 if (write_bitmap_index) {
891 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
892
893 stop_progress(&progress_state);
894
895 bitmap_writer_show_progress(progress);
896 bitmap_writer_reuse_bitmaps(&to_pack);
897 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
898 bitmap_writer_build(&to_pack);
899 bitmap_writer_finish(written_list, nr_written,
900 tmpname.buf, write_bitmap_options);
901 write_bitmap_index = 0;
902 }
903
904 strbuf_release(&tmpname);
905 free(pack_tmp_name);
906 puts(oid_to_hex(&oid));
907 }
908
909 /* mark written objects as written to previous pack */
910 for (j = 0; j < nr_written; j++) {
911 written_list[j]->offset = (off_t)-1;
912 }
913 nr_remaining -= nr_written;
914 } while (nr_remaining && i < to_pack.nr_objects);
915
916 free(written_list);
917 free(write_order);
918 stop_progress(&progress_state);
919 if (written != nr_result)
920 die("wrote %"PRIu32" objects while expecting %"PRIu32,
921 written, nr_result);
922 }
923
924 static int no_try_delta(const char *path)
925 {
926 static struct attr_check *check;
927
928 if (!check)
929 check = attr_check_initl("delta", NULL);
930 if (git_check_attr(path, check))
931 return 0;
932 if (ATTR_FALSE(check->items[0].value))
933 return 1;
934 return 0;
935 }
936
937 /*
938 * When adding an object, check whether we have already added it
939 * to our packing list. If so, we can skip. However, if we are
940 * being asked to excludei t, but the previous mention was to include
941 * it, make sure to adjust its flags and tweak our numbers accordingly.
942 *
943 * As an optimization, we pass out the index position where we would have
944 * found the item, since that saves us from having to look it up again a
945 * few lines later when we want to add the new entry.
946 */
947 static int have_duplicate_entry(const struct object_id *oid,
948 int exclude,
949 uint32_t *index_pos)
950 {
951 struct object_entry *entry;
952
953 entry = packlist_find(&to_pack, oid->hash, index_pos);
954 if (!entry)
955 return 0;
956
957 if (exclude) {
958 if (!entry->preferred_base)
959 nr_result--;
960 entry->preferred_base = 1;
961 }
962
963 return 1;
964 }
965
966 static int want_found_object(int exclude, struct packed_git *p)
967 {
968 if (exclude)
969 return 1;
970 if (incremental)
971 return 0;
972
973 /*
974 * When asked to do --local (do not include an object that appears in a
975 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
976 * an object that appears in a pack marked with .keep), finding a pack
977 * that matches the criteria is sufficient for us to decide to omit it.
978 * However, even if this pack does not satisfy the criteria, we need to
979 * make sure no copy of this object appears in _any_ pack that makes us
980 * to omit the object, so we need to check all the packs.
981 *
982 * We can however first check whether these options can possible matter;
983 * if they do not matter we know we want the object in generated pack.
984 * Otherwise, we signal "-1" at the end to tell the caller that we do
985 * not know either way, and it needs to check more packs.
986 */
987 if (!ignore_packed_keep &&
988 (!local || !have_non_local_packs))
989 return 1;
990
991 if (local && !p->pack_local)
992 return 0;
993 if (ignore_packed_keep && p->pack_local && p->pack_keep)
994 return 0;
995
996 /* we don't know yet; keep looking for more packs */
997 return -1;
998 }
999
1000 /*
1001 * Check whether we want the object in the pack (e.g., we do not want
1002 * objects found in non-local stores if the "--local" option was used).
1003 *
1004 * If the caller already knows an existing pack it wants to take the object
1005 * from, that is passed in *found_pack and *found_offset; otherwise this
1006 * function finds if there is any pack that has the object and returns the pack
1007 * and its offset in these variables.
1008 */
1009 static int want_object_in_pack(const struct object_id *oid,
1010 int exclude,
1011 struct packed_git **found_pack,
1012 off_t *found_offset)
1013 {
1014 int want;
1015 struct list_head *pos;
1016
1017 if (!exclude && local && has_loose_object_nonlocal(oid->hash))
1018 return 0;
1019
1020 /*
1021 * If we already know the pack object lives in, start checks from that
1022 * pack - in the usual case when neither --local was given nor .keep files
1023 * are present we will determine the answer right now.
1024 */
1025 if (*found_pack) {
1026 want = want_found_object(exclude, *found_pack);
1027 if (want != -1)
1028 return want;
1029 }
1030 list_for_each(pos, get_packed_git_mru(the_repository)) {
1031 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1032 off_t offset;
1033
1034 if (p == *found_pack)
1035 offset = *found_offset;
1036 else
1037 offset = find_pack_entry_one(oid->hash, p);
1038
1039 if (offset) {
1040 if (!*found_pack) {
1041 if (!is_pack_valid(p))
1042 continue;
1043 *found_offset = offset;
1044 *found_pack = p;
1045 }
1046 want = want_found_object(exclude, p);
1047 if (!exclude && want > 0)
1048 list_move(&p->mru,
1049 get_packed_git_mru(the_repository));
1050 if (want != -1)
1051 return want;
1052 }
1053 }
1054
1055 return 1;
1056 }
1057
1058 static void create_object_entry(const struct object_id *oid,
1059 enum object_type type,
1060 uint32_t hash,
1061 int exclude,
1062 int no_try_delta,
1063 uint32_t index_pos,
1064 struct packed_git *found_pack,
1065 off_t found_offset)
1066 {
1067 struct object_entry *entry;
1068
1069 entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1070 entry->hash = hash;
1071 oe_set_type(entry, type);
1072 if (exclude)
1073 entry->preferred_base = 1;
1074 else
1075 nr_result++;
1076 if (found_pack) {
1077 entry->in_pack = found_pack;
1078 entry->in_pack_offset = found_offset;
1079 }
1080
1081 entry->no_try_delta = no_try_delta;
1082 }
1083
1084 static const char no_closure_warning[] = N_(
1085 "disabling bitmap writing, as some objects are not being packed"
1086 );
1087
1088 static int add_object_entry(const struct object_id *oid, enum object_type type,
1089 const char *name, int exclude)
1090 {
1091 struct packed_git *found_pack = NULL;
1092 off_t found_offset = 0;
1093 uint32_t index_pos;
1094
1095 if (have_duplicate_entry(oid, exclude, &index_pos))
1096 return 0;
1097
1098 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1099 /* The pack is missing an object, so it will not have closure */
1100 if (write_bitmap_index) {
1101 warning(_(no_closure_warning));
1102 write_bitmap_index = 0;
1103 }
1104 return 0;
1105 }
1106
1107 create_object_entry(oid, type, pack_name_hash(name),
1108 exclude, name && no_try_delta(name),
1109 index_pos, found_pack, found_offset);
1110
1111 display_progress(progress_state, nr_result);
1112 return 1;
1113 }
1114
1115 static int add_object_entry_from_bitmap(const struct object_id *oid,
1116 enum object_type type,
1117 int flags, uint32_t name_hash,
1118 struct packed_git *pack, off_t offset)
1119 {
1120 uint32_t index_pos;
1121
1122 if (have_duplicate_entry(oid, 0, &index_pos))
1123 return 0;
1124
1125 if (!want_object_in_pack(oid, 0, &pack, &offset))
1126 return 0;
1127
1128 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1129
1130 display_progress(progress_state, nr_result);
1131 return 1;
1132 }
1133
1134 struct pbase_tree_cache {
1135 struct object_id oid;
1136 int ref;
1137 int temporary;
1138 void *tree_data;
1139 unsigned long tree_size;
1140 };
1141
1142 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1143 static int pbase_tree_cache_ix(const struct object_id *oid)
1144 {
1145 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1146 }
1147 static int pbase_tree_cache_ix_incr(int ix)
1148 {
1149 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1150 }
1151
1152 static struct pbase_tree {
1153 struct pbase_tree *next;
1154 /* This is a phony "cache" entry; we are not
1155 * going to evict it or find it through _get()
1156 * mechanism -- this is for the toplevel node that
1157 * would almost always change with any commit.
1158 */
1159 struct pbase_tree_cache pcache;
1160 } *pbase_tree;
1161
1162 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1163 {
1164 struct pbase_tree_cache *ent, *nent;
1165 void *data;
1166 unsigned long size;
1167 enum object_type type;
1168 int neigh;
1169 int my_ix = pbase_tree_cache_ix(oid);
1170 int available_ix = -1;
1171
1172 /* pbase-tree-cache acts as a limited hashtable.
1173 * your object will be found at your index or within a few
1174 * slots after that slot if it is cached.
1175 */
1176 for (neigh = 0; neigh < 8; neigh++) {
1177 ent = pbase_tree_cache[my_ix];
1178 if (ent && !oidcmp(&ent->oid, oid)) {
1179 ent->ref++;
1180 return ent;
1181 }
1182 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1183 ((0 <= available_ix) &&
1184 (!ent && pbase_tree_cache[available_ix])))
1185 available_ix = my_ix;
1186 if (!ent)
1187 break;
1188 my_ix = pbase_tree_cache_ix_incr(my_ix);
1189 }
1190
1191 /* Did not find one. Either we got a bogus request or
1192 * we need to read and perhaps cache.
1193 */
1194 data = read_object_file(oid, &type, &size);
1195 if (!data)
1196 return NULL;
1197 if (type != OBJ_TREE) {
1198 free(data);
1199 return NULL;
1200 }
1201
1202 /* We need to either cache or return a throwaway copy */
1203
1204 if (available_ix < 0)
1205 ent = NULL;
1206 else {
1207 ent = pbase_tree_cache[available_ix];
1208 my_ix = available_ix;
1209 }
1210
1211 if (!ent) {
1212 nent = xmalloc(sizeof(*nent));
1213 nent->temporary = (available_ix < 0);
1214 }
1215 else {
1216 /* evict and reuse */
1217 free(ent->tree_data);
1218 nent = ent;
1219 }
1220 oidcpy(&nent->oid, oid);
1221 nent->tree_data = data;
1222 nent->tree_size = size;
1223 nent->ref = 1;
1224 if (!nent->temporary)
1225 pbase_tree_cache[my_ix] = nent;
1226 return nent;
1227 }
1228
1229 static void pbase_tree_put(struct pbase_tree_cache *cache)
1230 {
1231 if (!cache->temporary) {
1232 cache->ref--;
1233 return;
1234 }
1235 free(cache->tree_data);
1236 free(cache);
1237 }
1238
1239 static int name_cmp_len(const char *name)
1240 {
1241 int i;
1242 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1243 ;
1244 return i;
1245 }
1246
1247 static void add_pbase_object(struct tree_desc *tree,
1248 const char *name,
1249 int cmplen,
1250 const char *fullname)
1251 {
1252 struct name_entry entry;
1253 int cmp;
1254
1255 while (tree_entry(tree,&entry)) {
1256 if (S_ISGITLINK(entry.mode))
1257 continue;
1258 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1259 memcmp(name, entry.path, cmplen);
1260 if (cmp > 0)
1261 continue;
1262 if (cmp < 0)
1263 return;
1264 if (name[cmplen] != '/') {
1265 add_object_entry(entry.oid,
1266 object_type(entry.mode),
1267 fullname, 1);
1268 return;
1269 }
1270 if (S_ISDIR(entry.mode)) {
1271 struct tree_desc sub;
1272 struct pbase_tree_cache *tree;
1273 const char *down = name+cmplen+1;
1274 int downlen = name_cmp_len(down);
1275
1276 tree = pbase_tree_get(entry.oid);
1277 if (!tree)
1278 return;
1279 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1280
1281 add_pbase_object(&sub, down, downlen, fullname);
1282 pbase_tree_put(tree);
1283 }
1284 }
1285 }
1286
1287 static unsigned *done_pbase_paths;
1288 static int done_pbase_paths_num;
1289 static int done_pbase_paths_alloc;
1290 static int done_pbase_path_pos(unsigned hash)
1291 {
1292 int lo = 0;
1293 int hi = done_pbase_paths_num;
1294 while (lo < hi) {
1295 int mi = lo + (hi - lo) / 2;
1296 if (done_pbase_paths[mi] == hash)
1297 return mi;
1298 if (done_pbase_paths[mi] < hash)
1299 hi = mi;
1300 else
1301 lo = mi + 1;
1302 }
1303 return -lo-1;
1304 }
1305
1306 static int check_pbase_path(unsigned hash)
1307 {
1308 int pos = done_pbase_path_pos(hash);
1309 if (0 <= pos)
1310 return 1;
1311 pos = -pos - 1;
1312 ALLOC_GROW(done_pbase_paths,
1313 done_pbase_paths_num + 1,
1314 done_pbase_paths_alloc);
1315 done_pbase_paths_num++;
1316 if (pos < done_pbase_paths_num)
1317 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1318 done_pbase_paths_num - pos - 1);
1319 done_pbase_paths[pos] = hash;
1320 return 0;
1321 }
1322
1323 static void add_preferred_base_object(const char *name)
1324 {
1325 struct pbase_tree *it;
1326 int cmplen;
1327 unsigned hash = pack_name_hash(name);
1328
1329 if (!num_preferred_base || check_pbase_path(hash))
1330 return;
1331
1332 cmplen = name_cmp_len(name);
1333 for (it = pbase_tree; it; it = it->next) {
1334 if (cmplen == 0) {
1335 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1336 }
1337 else {
1338 struct tree_desc tree;
1339 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1340 add_pbase_object(&tree, name, cmplen, name);
1341 }
1342 }
1343 }
1344
1345 static void add_preferred_base(struct object_id *oid)
1346 {
1347 struct pbase_tree *it;
1348 void *data;
1349 unsigned long size;
1350 struct object_id tree_oid;
1351
1352 if (window <= num_preferred_base++)
1353 return;
1354
1355 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);
1356 if (!data)
1357 return;
1358
1359 for (it = pbase_tree; it; it = it->next) {
1360 if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1361 free(data);
1362 return;
1363 }
1364 }
1365
1366 it = xcalloc(1, sizeof(*it));
1367 it->next = pbase_tree;
1368 pbase_tree = it;
1369
1370 oidcpy(&it->pcache.oid, &tree_oid);
1371 it->pcache.tree_data = data;
1372 it->pcache.tree_size = size;
1373 }
1374
1375 static void cleanup_preferred_base(void)
1376 {
1377 struct pbase_tree *it;
1378 unsigned i;
1379
1380 it = pbase_tree;
1381 pbase_tree = NULL;
1382 while (it) {
1383 struct pbase_tree *tmp = it;
1384 it = tmp->next;
1385 free(tmp->pcache.tree_data);
1386 free(tmp);
1387 }
1388
1389 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1390 if (!pbase_tree_cache[i])
1391 continue;
1392 free(pbase_tree_cache[i]->tree_data);
1393 FREE_AND_NULL(pbase_tree_cache[i]);
1394 }
1395
1396 FREE_AND_NULL(done_pbase_paths);
1397 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1398 }
1399
1400 static void check_object(struct object_entry *entry)
1401 {
1402 if (entry->in_pack) {
1403 struct packed_git *p = entry->in_pack;
1404 struct pack_window *w_curs = NULL;
1405 const unsigned char *base_ref = NULL;
1406 struct object_entry *base_entry;
1407 unsigned long used, used_0;
1408 unsigned long avail;
1409 off_t ofs;
1410 unsigned char *buf, c;
1411 enum object_type type;
1412
1413 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1414
1415 /*
1416 * We want in_pack_type even if we do not reuse delta
1417 * since non-delta representations could still be reused.
1418 */
1419 used = unpack_object_header_buffer(buf, avail,
1420 &type,
1421 &entry->size);
1422 if (used == 0)
1423 goto give_up;
1424
1425 if (type < 0)
1426 BUG("invalid type %d", type);
1427 entry->in_pack_type = type;
1428
1429 /*
1430 * Determine if this is a delta and if so whether we can
1431 * reuse it or not. Otherwise let's find out as cheaply as
1432 * possible what the actual type and size for this object is.
1433 */
1434 switch (entry->in_pack_type) {
1435 default:
1436 /* Not a delta hence we've already got all we need. */
1437 oe_set_type(entry, entry->in_pack_type);
1438 entry->in_pack_header_size = used;
1439 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1440 goto give_up;
1441 unuse_pack(&w_curs);
1442 return;
1443 case OBJ_REF_DELTA:
1444 if (reuse_delta && !entry->preferred_base)
1445 base_ref = use_pack(p, &w_curs,
1446 entry->in_pack_offset + used, NULL);
1447 entry->in_pack_header_size = used + 20;
1448 break;
1449 case OBJ_OFS_DELTA:
1450 buf = use_pack(p, &w_curs,
1451 entry->in_pack_offset + used, NULL);
1452 used_0 = 0;
1453 c = buf[used_0++];
1454 ofs = c & 127;
1455 while (c & 128) {
1456 ofs += 1;
1457 if (!ofs || MSB(ofs, 7)) {
1458 error("delta base offset overflow in pack for %s",
1459 oid_to_hex(&entry->idx.oid));
1460 goto give_up;
1461 }
1462 c = buf[used_0++];
1463 ofs = (ofs << 7) + (c & 127);
1464 }
1465 ofs = entry->in_pack_offset - ofs;
1466 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1467 error("delta base offset out of bound for %s",
1468 oid_to_hex(&entry->idx.oid));
1469 goto give_up;
1470 }
1471 if (reuse_delta && !entry->preferred_base) {
1472 struct revindex_entry *revidx;
1473 revidx = find_pack_revindex(p, ofs);
1474 if (!revidx)
1475 goto give_up;
1476 base_ref = nth_packed_object_sha1(p, revidx->nr);
1477 }
1478 entry->in_pack_header_size = used + used_0;
1479 break;
1480 }
1481
1482 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1483 /*
1484 * If base_ref was set above that means we wish to
1485 * reuse delta data, and we even found that base
1486 * in the list of objects we want to pack. Goodie!
1487 *
1488 * Depth value does not matter - find_deltas() will
1489 * never consider reused delta as the base object to
1490 * deltify other objects against, in order to avoid
1491 * circular deltas.
1492 */
1493 oe_set_type(entry, entry->in_pack_type);
1494 entry->delta = base_entry;
1495 entry->delta_size = entry->size;
1496 entry->delta_sibling = base_entry->delta_child;
1497 base_entry->delta_child = entry;
1498 unuse_pack(&w_curs);
1499 return;
1500 }
1501
1502 if (oe_type(entry)) {
1503 /*
1504 * This must be a delta and we already know what the
1505 * final object type is. Let's extract the actual
1506 * object size from the delta header.
1507 */
1508 entry->size = get_size_from_delta(p, &w_curs,
1509 entry->in_pack_offset + entry->in_pack_header_size);
1510 if (entry->size == 0)
1511 goto give_up;
1512 unuse_pack(&w_curs);
1513 return;
1514 }
1515
1516 /*
1517 * No choice but to fall back to the recursive delta walk
1518 * with sha1_object_info() to find about the object type
1519 * at this point...
1520 */
1521 give_up:
1522 unuse_pack(&w_curs);
1523 }
1524
1525 oe_set_type(entry, oid_object_info(&entry->idx.oid, &entry->size));
1526 /*
1527 * The error condition is checked in prepare_pack(). This is
1528 * to permit a missing preferred base object to be ignored
1529 * as a preferred base. Doing so can result in a larger
1530 * pack file, but the transfer will still take place.
1531 */
1532 }
1533
1534 static int pack_offset_sort(const void *_a, const void *_b)
1535 {
1536 const struct object_entry *a = *(struct object_entry **)_a;
1537 const struct object_entry *b = *(struct object_entry **)_b;
1538
1539 /* avoid filesystem trashing with loose objects */
1540 if (!a->in_pack && !b->in_pack)
1541 return oidcmp(&a->idx.oid, &b->idx.oid);
1542
1543 if (a->in_pack < b->in_pack)
1544 return -1;
1545 if (a->in_pack > b->in_pack)
1546 return 1;
1547 return a->in_pack_offset < b->in_pack_offset ? -1 :
1548 (a->in_pack_offset > b->in_pack_offset);
1549 }
1550
1551 /*
1552 * Drop an on-disk delta we were planning to reuse. Naively, this would
1553 * just involve blanking out the "delta" field, but we have to deal
1554 * with some extra book-keeping:
1555 *
1556 * 1. Removing ourselves from the delta_sibling linked list.
1557 *
1558 * 2. Updating our size/type to the non-delta representation. These were
1559 * either not recorded initially (size) or overwritten with the delta type
1560 * (type) when check_object() decided to reuse the delta.
1561 *
1562 * 3. Resetting our delta depth, as we are now a base object.
1563 */
1564 static void drop_reused_delta(struct object_entry *entry)
1565 {
1566 struct object_entry **p = &entry->delta->delta_child;
1567 struct object_info oi = OBJECT_INFO_INIT;
1568 enum object_type type;
1569
1570 while (*p) {
1571 if (*p == entry)
1572 *p = (*p)->delta_sibling;
1573 else
1574 p = &(*p)->delta_sibling;
1575 }
1576 entry->delta = NULL;
1577 entry->depth = 0;
1578
1579 oi.sizep = &entry->size;
1580 oi.typep = &type;
1581 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1582 /*
1583 * We failed to get the info from this pack for some reason;
1584 * fall back to sha1_object_info, which may find another copy.
1585 * And if that fails, the error will be recorded in oe_type(entry)
1586 * and dealt with in prepare_pack().
1587 */
1588 oe_set_type(entry, oid_object_info(&entry->idx.oid,
1589 &entry->size));
1590 } else {
1591 oe_set_type(entry, type);
1592 }
1593 }
1594
1595 /*
1596 * Follow the chain of deltas from this entry onward, throwing away any links
1597 * that cause us to hit a cycle (as determined by the DFS state flags in
1598 * the entries).
1599 *
1600 * We also detect too-long reused chains that would violate our --depth
1601 * limit.
1602 */
1603 static void break_delta_chains(struct object_entry *entry)
1604 {
1605 /*
1606 * The actual depth of each object we will write is stored as an int,
1607 * as it cannot exceed our int "depth" limit. But before we break
1608 * changes based no that limit, we may potentially go as deep as the
1609 * number of objects, which is elsewhere bounded to a uint32_t.
1610 */
1611 uint32_t total_depth;
1612 struct object_entry *cur, *next;
1613
1614 for (cur = entry, total_depth = 0;
1615 cur;
1616 cur = cur->delta, total_depth++) {
1617 if (cur->dfs_state == DFS_DONE) {
1618 /*
1619 * We've already seen this object and know it isn't
1620 * part of a cycle. We do need to append its depth
1621 * to our count.
1622 */
1623 total_depth += cur->depth;
1624 break;
1625 }
1626
1627 /*
1628 * We break cycles before looping, so an ACTIVE state (or any
1629 * other cruft which made its way into the state variable)
1630 * is a bug.
1631 */
1632 if (cur->dfs_state != DFS_NONE)
1633 die("BUG: confusing delta dfs state in first pass: %d",
1634 cur->dfs_state);
1635
1636 /*
1637 * Now we know this is the first time we've seen the object. If
1638 * it's not a delta, we're done traversing, but we'll mark it
1639 * done to save time on future traversals.
1640 */
1641 if (!cur->delta) {
1642 cur->dfs_state = DFS_DONE;
1643 break;
1644 }
1645
1646 /*
1647 * Mark ourselves as active and see if the next step causes
1648 * us to cycle to another active object. It's important to do
1649 * this _before_ we loop, because it impacts where we make the
1650 * cut, and thus how our total_depth counter works.
1651 * E.g., We may see a partial loop like:
1652 *
1653 * A -> B -> C -> D -> B
1654 *
1655 * Cutting B->C breaks the cycle. But now the depth of A is
1656 * only 1, and our total_depth counter is at 3. The size of the
1657 * error is always one less than the size of the cycle we
1658 * broke. Commits C and D were "lost" from A's chain.
1659 *
1660 * If we instead cut D->B, then the depth of A is correct at 3.
1661 * We keep all commits in the chain that we examined.
1662 */
1663 cur->dfs_state = DFS_ACTIVE;
1664 if (cur->delta->dfs_state == DFS_ACTIVE) {
1665 drop_reused_delta(cur);
1666 cur->dfs_state = DFS_DONE;
1667 break;
1668 }
1669 }
1670
1671 /*
1672 * And now that we've gone all the way to the bottom of the chain, we
1673 * need to clear the active flags and set the depth fields as
1674 * appropriate. Unlike the loop above, which can quit when it drops a
1675 * delta, we need to keep going to look for more depth cuts. So we need
1676 * an extra "next" pointer to keep going after we reset cur->delta.
1677 */
1678 for (cur = entry; cur; cur = next) {
1679 next = cur->delta;
1680
1681 /*
1682 * We should have a chain of zero or more ACTIVE states down to
1683 * a final DONE. We can quit after the DONE, because either it
1684 * has no bases, or we've already handled them in a previous
1685 * call.
1686 */
1687 if (cur->dfs_state == DFS_DONE)
1688 break;
1689 else if (cur->dfs_state != DFS_ACTIVE)
1690 die("BUG: confusing delta dfs state in second pass: %d",
1691 cur->dfs_state);
1692
1693 /*
1694 * If the total_depth is more than depth, then we need to snip
1695 * the chain into two or more smaller chains that don't exceed
1696 * the maximum depth. Most of the resulting chains will contain
1697 * (depth + 1) entries (i.e., depth deltas plus one base), and
1698 * the last chain (i.e., the one containing entry) will contain
1699 * whatever entries are left over, namely
1700 * (total_depth % (depth + 1)) of them.
1701 *
1702 * Since we are iterating towards decreasing depth, we need to
1703 * decrement total_depth as we go, and we need to write to the
1704 * entry what its final depth will be after all of the
1705 * snipping. Since we're snipping into chains of length (depth
1706 * + 1) entries, the final depth of an entry will be its
1707 * original depth modulo (depth + 1). Any time we encounter an
1708 * entry whose final depth is supposed to be zero, we snip it
1709 * from its delta base, thereby making it so.
1710 */
1711 cur->depth = (total_depth--) % (depth + 1);
1712 if (!cur->depth)
1713 drop_reused_delta(cur);
1714
1715 cur->dfs_state = DFS_DONE;
1716 }
1717 }
1718
1719 static void get_object_details(void)
1720 {
1721 uint32_t i;
1722 struct object_entry **sorted_by_offset;
1723
1724 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1725 for (i = 0; i < to_pack.nr_objects; i++)
1726 sorted_by_offset[i] = to_pack.objects + i;
1727 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1728
1729 for (i = 0; i < to_pack.nr_objects; i++) {
1730 struct object_entry *entry = sorted_by_offset[i];
1731 check_object(entry);
1732 if (big_file_threshold < entry->size)
1733 entry->no_try_delta = 1;
1734 }
1735
1736 /*
1737 * This must happen in a second pass, since we rely on the delta
1738 * information for the whole list being completed.
1739 */
1740 for (i = 0; i < to_pack.nr_objects; i++)
1741 break_delta_chains(&to_pack.objects[i]);
1742
1743 free(sorted_by_offset);
1744 }
1745
1746 /*
1747 * We search for deltas in a list sorted by type, by filename hash, and then
1748 * by size, so that we see progressively smaller and smaller files.
1749 * That's because we prefer deltas to be from the bigger file
1750 * to the smaller -- deletes are potentially cheaper, but perhaps
1751 * more importantly, the bigger file is likely the more recent
1752 * one. The deepest deltas are therefore the oldest objects which are
1753 * less susceptible to be accessed often.
1754 */
1755 static int type_size_sort(const void *_a, const void *_b)
1756 {
1757 const struct object_entry *a = *(struct object_entry **)_a;
1758 const struct object_entry *b = *(struct object_entry **)_b;
1759 enum object_type a_type = oe_type(a);
1760 enum object_type b_type = oe_type(b);
1761
1762 if (a_type > b_type)
1763 return -1;
1764 if (a_type < b_type)
1765 return 1;
1766 if (a->hash > b->hash)
1767 return -1;
1768 if (a->hash < b->hash)
1769 return 1;
1770 if (a->preferred_base > b->preferred_base)
1771 return -1;
1772 if (a->preferred_base < b->preferred_base)
1773 return 1;
1774 if (a->size > b->size)
1775 return -1;
1776 if (a->size < b->size)
1777 return 1;
1778 return a < b ? -1 : (a > b); /* newest first */
1779 }
1780
1781 struct unpacked {
1782 struct object_entry *entry;
1783 void *data;
1784 struct delta_index *index;
1785 unsigned depth;
1786 };
1787
1788 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1789 unsigned long delta_size)
1790 {
1791 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1792 return 0;
1793
1794 if (delta_size < cache_max_small_delta_size)
1795 return 1;
1796
1797 /* cache delta, if objects are large enough compared to delta size */
1798 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1799 return 1;
1800
1801 return 0;
1802 }
1803
1804 #ifndef NO_PTHREADS
1805
1806 static pthread_mutex_t read_mutex;
1807 #define read_lock() pthread_mutex_lock(&read_mutex)
1808 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1809
1810 static pthread_mutex_t cache_mutex;
1811 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1812 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1813
1814 static pthread_mutex_t progress_mutex;
1815 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1816 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1817
1818 #else
1819
1820 #define read_lock() (void)0
1821 #define read_unlock() (void)0
1822 #define cache_lock() (void)0
1823 #define cache_unlock() (void)0
1824 #define progress_lock() (void)0
1825 #define progress_unlock() (void)0
1826
1827 #endif
1828
1829 static int try_delta(struct unpacked *trg, struct unpacked *src,
1830 unsigned max_depth, unsigned long *mem_usage)
1831 {
1832 struct object_entry *trg_entry = trg->entry;
1833 struct object_entry *src_entry = src->entry;
1834 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1835 unsigned ref_depth;
1836 enum object_type type;
1837 void *delta_buf;
1838
1839 /* Don't bother doing diffs between different types */
1840 if (oe_type(trg_entry) != oe_type(src_entry))
1841 return -1;
1842
1843 /*
1844 * We do not bother to try a delta that we discarded on an
1845 * earlier try, but only when reusing delta data. Note that
1846 * src_entry that is marked as the preferred_base should always
1847 * be considered, as even if we produce a suboptimal delta against
1848 * it, we will still save the transfer cost, as we already know
1849 * the other side has it and we won't send src_entry at all.
1850 */
1851 if (reuse_delta && trg_entry->in_pack &&
1852 trg_entry->in_pack == src_entry->in_pack &&
1853 !src_entry->preferred_base &&
1854 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1855 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1856 return 0;
1857
1858 /* Let's not bust the allowed depth. */
1859 if (src->depth >= max_depth)
1860 return 0;
1861
1862 /* Now some size filtering heuristics. */
1863 trg_size = trg_entry->size;
1864 if (!trg_entry->delta) {
1865 max_size = trg_size/2 - 20;
1866 ref_depth = 1;
1867 } else {
1868 max_size = trg_entry->delta_size;
1869 ref_depth = trg->depth;
1870 }
1871 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1872 (max_depth - ref_depth + 1);
1873 if (max_size == 0)
1874 return 0;
1875 src_size = src_entry->size;
1876 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1877 if (sizediff >= max_size)
1878 return 0;
1879 if (trg_size < src_size / 32)
1880 return 0;
1881
1882 /* Load data if not already done */
1883 if (!trg->data) {
1884 read_lock();
1885 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
1886 read_unlock();
1887 if (!trg->data)
1888 die("object %s cannot be read",
1889 oid_to_hex(&trg_entry->idx.oid));
1890 if (sz != trg_size)
1891 die("object %s inconsistent object length (%lu vs %lu)",
1892 oid_to_hex(&trg_entry->idx.oid), sz,
1893 trg_size);
1894 *mem_usage += sz;
1895 }
1896 if (!src->data) {
1897 read_lock();
1898 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
1899 read_unlock();
1900 if (!src->data) {
1901 if (src_entry->preferred_base) {
1902 static int warned = 0;
1903 if (!warned++)
1904 warning("object %s cannot be read",
1905 oid_to_hex(&src_entry->idx.oid));
1906 /*
1907 * Those objects are not included in the
1908 * resulting pack. Be resilient and ignore
1909 * them if they can't be read, in case the
1910 * pack could be created nevertheless.
1911 */
1912 return 0;
1913 }
1914 die("object %s cannot be read",
1915 oid_to_hex(&src_entry->idx.oid));
1916 }
1917 if (sz != src_size)
1918 die("object %s inconsistent object length (%lu vs %lu)",
1919 oid_to_hex(&src_entry->idx.oid), sz,
1920 src_size);
1921 *mem_usage += sz;
1922 }
1923 if (!src->index) {
1924 src->index = create_delta_index(src->data, src_size);
1925 if (!src->index) {
1926 static int warned = 0;
1927 if (!warned++)
1928 warning("suboptimal pack - out of memory");
1929 return 0;
1930 }
1931 *mem_usage += sizeof_delta_index(src->index);
1932 }
1933
1934 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1935 if (!delta_buf)
1936 return 0;
1937
1938 if (trg_entry->delta) {
1939 /* Prefer only shallower same-sized deltas. */
1940 if (delta_size == trg_entry->delta_size &&
1941 src->depth + 1 >= trg->depth) {
1942 free(delta_buf);
1943 return 0;
1944 }
1945 }
1946
1947 /*
1948 * Handle memory allocation outside of the cache
1949 * accounting lock. Compiler will optimize the strangeness
1950 * away when NO_PTHREADS is defined.
1951 */
1952 free(trg_entry->delta_data);
1953 cache_lock();
1954 if (trg_entry->delta_data) {
1955 delta_cache_size -= trg_entry->delta_size;
1956 trg_entry->delta_data = NULL;
1957 }
1958 if (delta_cacheable(src_size, trg_size, delta_size)) {
1959 delta_cache_size += delta_size;
1960 cache_unlock();
1961 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1962 } else {
1963 cache_unlock();
1964 free(delta_buf);
1965 }
1966
1967 trg_entry->delta = src_entry;
1968 trg_entry->delta_size = delta_size;
1969 trg->depth = src->depth + 1;
1970
1971 return 1;
1972 }
1973
1974 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1975 {
1976 struct object_entry *child = me->delta_child;
1977 unsigned int m = n;
1978 while (child) {
1979 unsigned int c = check_delta_limit(child, n + 1);
1980 if (m < c)
1981 m = c;
1982 child = child->delta_sibling;
1983 }
1984 return m;
1985 }
1986
1987 static unsigned long free_unpacked(struct unpacked *n)
1988 {
1989 unsigned long freed_mem = sizeof_delta_index(n->index);
1990 free_delta_index(n->index);
1991 n->index = NULL;
1992 if (n->data) {
1993 freed_mem += n->entry->size;
1994 FREE_AND_NULL(n->data);
1995 }
1996 n->entry = NULL;
1997 n->depth = 0;
1998 return freed_mem;
1999 }
2000
2001 static void find_deltas(struct object_entry **list, unsigned *list_size,
2002 int window, int depth, unsigned *processed)
2003 {
2004 uint32_t i, idx = 0, count = 0;
2005 struct unpacked *array;
2006 unsigned long mem_usage = 0;
2007
2008 array = xcalloc(window, sizeof(struct unpacked));
2009
2010 for (;;) {
2011 struct object_entry *entry;
2012 struct unpacked *n = array + idx;
2013 int j, max_depth, best_base = -1;
2014
2015 progress_lock();
2016 if (!*list_size) {
2017 progress_unlock();
2018 break;
2019 }
2020 entry = *list++;
2021 (*list_size)--;
2022 if (!entry->preferred_base) {
2023 (*processed)++;
2024 display_progress(progress_state, *processed);
2025 }
2026 progress_unlock();
2027
2028 mem_usage -= free_unpacked(n);
2029 n->entry = entry;
2030
2031 while (window_memory_limit &&
2032 mem_usage > window_memory_limit &&
2033 count > 1) {
2034 uint32_t tail = (idx + window - count) % window;
2035 mem_usage -= free_unpacked(array + tail);
2036 count--;
2037 }
2038
2039 /* We do not compute delta to *create* objects we are not
2040 * going to pack.
2041 */
2042 if (entry->preferred_base)
2043 goto next;
2044
2045 /*
2046 * If the current object is at pack edge, take the depth the
2047 * objects that depend on the current object into account
2048 * otherwise they would become too deep.
2049 */
2050 max_depth = depth;
2051 if (entry->delta_child) {
2052 max_depth -= check_delta_limit(entry, 0);
2053 if (max_depth <= 0)
2054 goto next;
2055 }
2056
2057 j = window;
2058 while (--j > 0) {
2059 int ret;
2060 uint32_t other_idx = idx + j;
2061 struct unpacked *m;
2062 if (other_idx >= window)
2063 other_idx -= window;
2064 m = array + other_idx;
2065 if (!m->entry)
2066 break;
2067 ret = try_delta(n, m, max_depth, &mem_usage);
2068 if (ret < 0)
2069 break;
2070 else if (ret > 0)
2071 best_base = other_idx;
2072 }
2073
2074 /*
2075 * If we decided to cache the delta data, then it is best
2076 * to compress it right away. First because we have to do
2077 * it anyway, and doing it here while we're threaded will
2078 * save a lot of time in the non threaded write phase,
2079 * as well as allow for caching more deltas within
2080 * the same cache size limit.
2081 * ...
2082 * But only if not writing to stdout, since in that case
2083 * the network is most likely throttling writes anyway,
2084 * and therefore it is best to go to the write phase ASAP
2085 * instead, as we can afford spending more time compressing
2086 * between writes at that moment.
2087 */
2088 if (entry->delta_data && !pack_to_stdout) {
2089 entry->z_delta_size = do_compress(&entry->delta_data,
2090 entry->delta_size);
2091 cache_lock();
2092 delta_cache_size -= entry->delta_size;
2093 delta_cache_size += entry->z_delta_size;
2094 cache_unlock();
2095 }
2096
2097 /* if we made n a delta, and if n is already at max
2098 * depth, leaving it in the window is pointless. we
2099 * should evict it first.
2100 */
2101 if (entry->delta && max_depth <= n->depth)
2102 continue;
2103
2104 /*
2105 * Move the best delta base up in the window, after the
2106 * currently deltified object, to keep it longer. It will
2107 * be the first base object to be attempted next.
2108 */
2109 if (entry->delta) {
2110 struct unpacked swap = array[best_base];
2111 int dist = (window + idx - best_base) % window;
2112 int dst = best_base;
2113 while (dist--) {
2114 int src = (dst + 1) % window;
2115 array[dst] = array[src];
2116 dst = src;
2117 }
2118 array[dst] = swap;
2119 }
2120
2121 next:
2122 idx++;
2123 if (count + 1 < window)
2124 count++;
2125 if (idx >= window)
2126 idx = 0;
2127 }
2128
2129 for (i = 0; i < window; ++i) {
2130 free_delta_index(array[i].index);
2131 free(array[i].data);
2132 }
2133 free(array);
2134 }
2135
2136 #ifndef NO_PTHREADS
2137
2138 static void try_to_free_from_threads(size_t size)
2139 {
2140 read_lock();
2141 release_pack_memory(size);
2142 read_unlock();
2143 }
2144
2145 static try_to_free_t old_try_to_free_routine;
2146
2147 /*
2148 * The main thread waits on the condition that (at least) one of the workers
2149 * has stopped working (which is indicated in the .working member of
2150 * struct thread_params).
2151 * When a work thread has completed its work, it sets .working to 0 and
2152 * signals the main thread and waits on the condition that .data_ready
2153 * becomes 1.
2154 */
2155
2156 struct thread_params {
2157 pthread_t thread;
2158 struct object_entry **list;
2159 unsigned list_size;
2160 unsigned remaining;
2161 int window;
2162 int depth;
2163 int working;
2164 int data_ready;
2165 pthread_mutex_t mutex;
2166 pthread_cond_t cond;
2167 unsigned *processed;
2168 };
2169
2170 static pthread_cond_t progress_cond;
2171
2172 /*
2173 * Mutex and conditional variable can't be statically-initialized on Windows.
2174 */
2175 static void init_threaded_search(void)
2176 {
2177 init_recursive_mutex(&read_mutex);
2178 pthread_mutex_init(&cache_mutex, NULL);
2179 pthread_mutex_init(&progress_mutex, NULL);
2180 pthread_cond_init(&progress_cond, NULL);
2181 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2182 }
2183
2184 static void cleanup_threaded_search(void)
2185 {
2186 set_try_to_free_routine(old_try_to_free_routine);
2187 pthread_cond_destroy(&progress_cond);
2188 pthread_mutex_destroy(&read_mutex);
2189 pthread_mutex_destroy(&cache_mutex);
2190 pthread_mutex_destroy(&progress_mutex);
2191 }
2192
2193 static void *threaded_find_deltas(void *arg)
2194 {
2195 struct thread_params *me = arg;
2196
2197 progress_lock();
2198 while (me->remaining) {
2199 progress_unlock();
2200
2201 find_deltas(me->list, &me->remaining,
2202 me->window, me->depth, me->processed);
2203
2204 progress_lock();
2205 me->working = 0;
2206 pthread_cond_signal(&progress_cond);
2207 progress_unlock();
2208
2209 /*
2210 * We must not set ->data_ready before we wait on the
2211 * condition because the main thread may have set it to 1
2212 * before we get here. In order to be sure that new
2213 * work is available if we see 1 in ->data_ready, it
2214 * was initialized to 0 before this thread was spawned
2215 * and we reset it to 0 right away.
2216 */
2217 pthread_mutex_lock(&me->mutex);
2218 while (!me->data_ready)
2219 pthread_cond_wait(&me->cond, &me->mutex);
2220 me->data_ready = 0;
2221 pthread_mutex_unlock(&me->mutex);
2222
2223 progress_lock();
2224 }
2225 progress_unlock();
2226 /* leave ->working 1 so that this doesn't get more work assigned */
2227 return NULL;
2228 }
2229
2230 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2231 int window, int depth, unsigned *processed)
2232 {
2233 struct thread_params *p;
2234 int i, ret, active_threads = 0;
2235
2236 init_threaded_search();
2237
2238 if (delta_search_threads <= 1) {
2239 find_deltas(list, &list_size, window, depth, processed);
2240 cleanup_threaded_search();
2241 return;
2242 }
2243 if (progress > pack_to_stdout)
2244 fprintf(stderr, "Delta compression using up to %d threads.\n",
2245 delta_search_threads);
2246 p = xcalloc(delta_search_threads, sizeof(*p));
2247
2248 /* Partition the work amongst work threads. */
2249 for (i = 0; i < delta_search_threads; i++) {
2250 unsigned sub_size = list_size / (delta_search_threads - i);
2251
2252 /* don't use too small segments or no deltas will be found */
2253 if (sub_size < 2*window && i+1 < delta_search_threads)
2254 sub_size = 0;
2255
2256 p[i].window = window;
2257 p[i].depth = depth;
2258 p[i].processed = processed;
2259 p[i].working = 1;
2260 p[i].data_ready = 0;
2261
2262 /* try to split chunks on "path" boundaries */
2263 while (sub_size && sub_size < list_size &&
2264 list[sub_size]->hash &&
2265 list[sub_size]->hash == list[sub_size-1]->hash)
2266 sub_size++;
2267
2268 p[i].list = list;
2269 p[i].list_size = sub_size;
2270 p[i].remaining = sub_size;
2271
2272 list += sub_size;
2273 list_size -= sub_size;
2274 }
2275
2276 /* Start work threads. */
2277 for (i = 0; i < delta_search_threads; i++) {
2278 if (!p[i].list_size)
2279 continue;
2280 pthread_mutex_init(&p[i].mutex, NULL);
2281 pthread_cond_init(&p[i].cond, NULL);
2282 ret = pthread_create(&p[i].thread, NULL,
2283 threaded_find_deltas, &p[i]);
2284 if (ret)
2285 die("unable to create thread: %s", strerror(ret));
2286 active_threads++;
2287 }
2288
2289 /*
2290 * Now let's wait for work completion. Each time a thread is done
2291 * with its work, we steal half of the remaining work from the
2292 * thread with the largest number of unprocessed objects and give
2293 * it to that newly idle thread. This ensure good load balancing
2294 * until the remaining object list segments are simply too short
2295 * to be worth splitting anymore.
2296 */
2297 while (active_threads) {
2298 struct thread_params *target = NULL;
2299 struct thread_params *victim = NULL;
2300 unsigned sub_size = 0;
2301
2302 progress_lock();
2303 for (;;) {
2304 for (i = 0; !target && i < delta_search_threads; i++)
2305 if (!p[i].working)
2306 target = &p[i];
2307 if (target)
2308 break;
2309 pthread_cond_wait(&progress_cond, &progress_mutex);
2310 }
2311
2312 for (i = 0; i < delta_search_threads; i++)
2313 if (p[i].remaining > 2*window &&
2314 (!victim || victim->remaining < p[i].remaining))
2315 victim = &p[i];
2316 if (victim) {
2317 sub_size = victim->remaining / 2;
2318 list = victim->list + victim->list_size - sub_size;
2319 while (sub_size && list[0]->hash &&
2320 list[0]->hash == list[-1]->hash) {
2321 list++;
2322 sub_size--;
2323 }
2324 if (!sub_size) {
2325 /*
2326 * It is possible for some "paths" to have
2327 * so many objects that no hash boundary
2328 * might be found. Let's just steal the
2329 * exact half in that case.
2330 */
2331 sub_size = victim->remaining / 2;
2332 list -= sub_size;
2333 }
2334 target->list = list;
2335 victim->list_size -= sub_size;
2336 victim->remaining -= sub_size;
2337 }
2338 target->list_size = sub_size;
2339 target->remaining = sub_size;
2340 target->working = 1;
2341 progress_unlock();
2342
2343 pthread_mutex_lock(&target->mutex);
2344 target->data_ready = 1;
2345 pthread_cond_signal(&target->cond);
2346 pthread_mutex_unlock(&target->mutex);
2347
2348 if (!sub_size) {
2349 pthread_join(target->thread, NULL);
2350 pthread_cond_destroy(&target->cond);
2351 pthread_mutex_destroy(&target->mutex);
2352 active_threads--;
2353 }
2354 }
2355 cleanup_threaded_search();
2356 free(p);
2357 }
2358
2359 #else
2360 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2361 #endif
2362
2363 static void add_tag_chain(const struct object_id *oid)
2364 {
2365 struct tag *tag;
2366
2367 /*
2368 * We catch duplicates already in add_object_entry(), but we'd
2369 * prefer to do this extra check to avoid having to parse the
2370 * tag at all if we already know that it's being packed (e.g., if
2371 * it was included via bitmaps, we would not have parsed it
2372 * previously).
2373 */
2374 if (packlist_find(&to_pack, oid->hash, NULL))
2375 return;
2376
2377 tag = lookup_tag(oid);
2378 while (1) {
2379 if (!tag || parse_tag(tag) || !tag->tagged)
2380 die("unable to pack objects reachable from tag %s",
2381 oid_to_hex(oid));
2382
2383 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2384
2385 if (tag->tagged->type != OBJ_TAG)
2386 return;
2387
2388 tag = (struct tag *)tag->tagged;
2389 }
2390 }
2391
2392 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2393 {
2394 struct object_id peeled;
2395
2396 if (starts_with(path, "refs/tags/") && /* is a tag? */
2397 !peel_ref(path, &peeled) && /* peelable? */
2398 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2399 add_tag_chain(oid);
2400 return 0;
2401 }
2402
2403 static void prepare_pack(int window, int depth)
2404 {
2405 struct object_entry **delta_list;
2406 uint32_t i, nr_deltas;
2407 unsigned n;
2408
2409 get_object_details();
2410
2411 /*
2412 * If we're locally repacking then we need to be doubly careful
2413 * from now on in order to make sure no stealth corruption gets
2414 * propagated to the new pack. Clients receiving streamed packs
2415 * should validate everything they get anyway so no need to incur
2416 * the additional cost here in that case.
2417 */
2418 if (!pack_to_stdout)
2419 do_check_packed_object_crc = 1;
2420
2421 if (!to_pack.nr_objects || !window || !depth)
2422 return;
2423
2424 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2425 nr_deltas = n = 0;
2426
2427 for (i = 0; i < to_pack.nr_objects; i++) {
2428 struct object_entry *entry = to_pack.objects + i;
2429
2430 if (entry->delta)
2431 /* This happens if we decided to reuse existing
2432 * delta from a pack. "reuse_delta &&" is implied.
2433 */
2434 continue;
2435
2436 if (entry->size < 50)
2437 continue;
2438
2439 if (entry->no_try_delta)
2440 continue;
2441
2442 if (!entry->preferred_base) {
2443 nr_deltas++;
2444 if (oe_type(entry) < 0)
2445 die("unable to get type of object %s",
2446 oid_to_hex(&entry->idx.oid));
2447 } else {
2448 if (oe_type(entry) < 0) {
2449 /*
2450 * This object is not found, but we
2451 * don't have to include it anyway.
2452 */
2453 continue;
2454 }
2455 }
2456
2457 delta_list[n++] = entry;
2458 }
2459
2460 if (nr_deltas && n > 1) {
2461 unsigned nr_done = 0;
2462 if (progress)
2463 progress_state = start_progress(_("Compressing objects"),
2464 nr_deltas);
2465 QSORT(delta_list, n, type_size_sort);
2466 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2467 stop_progress(&progress_state);
2468 if (nr_done != nr_deltas)
2469 die("inconsistency with delta count");
2470 }
2471 free(delta_list);
2472 }
2473
2474 static int git_pack_config(const char *k, const char *v, void *cb)
2475 {
2476 if (!strcmp(k, "pack.window")) {
2477 window = git_config_int(k, v);
2478 return 0;
2479 }
2480 if (!strcmp(k, "pack.windowmemory")) {
2481 window_memory_limit = git_config_ulong(k, v);
2482 return 0;
2483 }
2484 if (!strcmp(k, "pack.depth")) {
2485 depth = git_config_int(k, v);
2486 return 0;
2487 }
2488 if (!strcmp(k, "pack.deltacachesize")) {
2489 max_delta_cache_size = git_config_int(k, v);
2490 return 0;
2491 }
2492 if (!strcmp(k, "pack.deltacachelimit")) {
2493 cache_max_small_delta_size = git_config_int(k, v);
2494 return 0;
2495 }
2496 if (!strcmp(k, "pack.writebitmaphashcache")) {
2497 if (git_config_bool(k, v))
2498 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2499 else
2500 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2501 }
2502 if (!strcmp(k, "pack.usebitmaps")) {
2503 use_bitmap_index_default = git_config_bool(k, v);
2504 return 0;
2505 }
2506 if (!strcmp(k, "pack.threads")) {
2507 delta_search_threads = git_config_int(k, v);
2508 if (delta_search_threads < 0)
2509 die("invalid number of threads specified (%d)",
2510 delta_search_threads);
2511 #ifdef NO_PTHREADS
2512 if (delta_search_threads != 1) {
2513 warning("no threads support, ignoring %s", k);
2514 delta_search_threads = 0;
2515 }
2516 #endif
2517 return 0;
2518 }
2519 if (!strcmp(k, "pack.indexversion")) {
2520 pack_idx_opts.version = git_config_int(k, v);
2521 if (pack_idx_opts.version > 2)
2522 die("bad pack.indexversion=%"PRIu32,
2523 pack_idx_opts.version);
2524 return 0;
2525 }
2526 return git_default_config(k, v, cb);
2527 }
2528
2529 static void read_object_list_from_stdin(void)
2530 {
2531 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2532 struct object_id oid;
2533 const char *p;
2534
2535 for (;;) {
2536 if (!fgets(line, sizeof(line), stdin)) {
2537 if (feof(stdin))
2538 break;
2539 if (!ferror(stdin))
2540 die("fgets returned NULL, not EOF, not error!");
2541 if (errno != EINTR)
2542 die_errno("fgets");
2543 clearerr(stdin);
2544 continue;
2545 }
2546 if (line[0] == '-') {
2547 if (get_oid_hex(line+1, &oid))
2548 die("expected edge object ID, got garbage:\n %s",
2549 line);
2550 add_preferred_base(&oid);
2551 continue;
2552 }
2553 if (parse_oid_hex(line, &oid, &p))
2554 die("expected object ID, got garbage:\n %s", line);
2555
2556 add_preferred_base_object(p + 1);
2557 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2558 }
2559 }
2560
2561 /* Remember to update object flag allocation in object.h */
2562 #define OBJECT_ADDED (1u<<20)
2563
2564 static void show_commit(struct commit *commit, void *data)
2565 {
2566 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2567 commit->object.flags |= OBJECT_ADDED;
2568
2569 if (write_bitmap_index)
2570 index_commit_for_bitmap(commit);
2571 }
2572
2573 static void show_object(struct object *obj, const char *name, void *data)
2574 {
2575 add_preferred_base_object(name);
2576 add_object_entry(&obj->oid, obj->type, name, 0);
2577 obj->flags |= OBJECT_ADDED;
2578 }
2579
2580 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2581 {
2582 assert(arg_missing_action == MA_ALLOW_ANY);
2583
2584 /*
2585 * Quietly ignore ALL missing objects. This avoids problems with
2586 * staging them now and getting an odd error later.
2587 */
2588 if (!has_object_file(&obj->oid))
2589 return;
2590
2591 show_object(obj, name, data);
2592 }
2593
2594 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2595 {
2596 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2597
2598 /*
2599 * Quietly ignore EXPECTED missing objects. This avoids problems with
2600 * staging them now and getting an odd error later.
2601 */
2602 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2603 return;
2604
2605 show_object(obj, name, data);
2606 }
2607
2608 static int option_parse_missing_action(const struct option *opt,
2609 const char *arg, int unset)
2610 {
2611 assert(arg);
2612 assert(!unset);
2613
2614 if (!strcmp(arg, "error")) {
2615 arg_missing_action = MA_ERROR;
2616 fn_show_object = show_object;
2617 return 0;
2618 }
2619
2620 if (!strcmp(arg, "allow-any")) {
2621 arg_missing_action = MA_ALLOW_ANY;
2622 fetch_if_missing = 0;
2623 fn_show_object = show_object__ma_allow_any;
2624 return 0;
2625 }
2626
2627 if (!strcmp(arg, "allow-promisor")) {
2628 arg_missing_action = MA_ALLOW_PROMISOR;
2629 fetch_if_missing = 0;
2630 fn_show_object = show_object__ma_allow_promisor;
2631 return 0;
2632 }
2633
2634 die(_("invalid value for --missing"));
2635 return 0;
2636 }
2637
2638 static void show_edge(struct commit *commit)
2639 {
2640 add_preferred_base(&commit->object.oid);
2641 }
2642
2643 struct in_pack_object {
2644 off_t offset;
2645 struct object *object;
2646 };
2647
2648 struct in_pack {
2649 unsigned int alloc;
2650 unsigned int nr;
2651 struct in_pack_object *array;
2652 };
2653
2654 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2655 {
2656 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2657 in_pack->array[in_pack->nr].object = object;
2658 in_pack->nr++;
2659 }
2660
2661 /*
2662 * Compare the objects in the offset order, in order to emulate the
2663 * "git rev-list --objects" output that produced the pack originally.
2664 */
2665 static int ofscmp(const void *a_, const void *b_)
2666 {
2667 struct in_pack_object *a = (struct in_pack_object *)a_;
2668 struct in_pack_object *b = (struct in_pack_object *)b_;
2669
2670 if (a->offset < b->offset)
2671 return -1;
2672 else if (a->offset > b->offset)
2673 return 1;
2674 else
2675 return oidcmp(&a->object->oid, &b->object->oid);
2676 }
2677
2678 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2679 {
2680 struct packed_git *p;
2681 struct in_pack in_pack;
2682 uint32_t i;
2683
2684 memset(&in_pack, 0, sizeof(in_pack));
2685
2686 for (p = get_packed_git(the_repository); p; p = p->next) {
2687 struct object_id oid;
2688 struct object *o;
2689
2690 if (!p->pack_local || p->pack_keep)
2691 continue;
2692 if (open_pack_index(p))
2693 die("cannot open pack index");
2694
2695 ALLOC_GROW(in_pack.array,
2696 in_pack.nr + p->num_objects,
2697 in_pack.alloc);
2698
2699 for (i = 0; i < p->num_objects; i++) {
2700 nth_packed_object_oid(&oid, p, i);
2701 o = lookup_unknown_object(oid.hash);
2702 if (!(o->flags & OBJECT_ADDED))
2703 mark_in_pack_object(o, p, &in_pack);
2704 o->flags |= OBJECT_ADDED;
2705 }
2706 }
2707
2708 if (in_pack.nr) {
2709 QSORT(in_pack.array, in_pack.nr, ofscmp);
2710 for (i = 0; i < in_pack.nr; i++) {
2711 struct object *o = in_pack.array[i].object;
2712 add_object_entry(&o->oid, o->type, "", 0);
2713 }
2714 }
2715 free(in_pack.array);
2716 }
2717
2718 static int add_loose_object(const struct object_id *oid, const char *path,
2719 void *data)
2720 {
2721 enum object_type type = oid_object_info(oid, NULL);
2722
2723 if (type < 0) {
2724 warning("loose object at %s could not be examined", path);
2725 return 0;
2726 }
2727
2728 add_object_entry(oid, type, "", 0);
2729 return 0;
2730 }
2731
2732 /*
2733 * We actually don't even have to worry about reachability here.
2734 * add_object_entry will weed out duplicates, so we just add every
2735 * loose object we find.
2736 */
2737 static void add_unreachable_loose_objects(void)
2738 {
2739 for_each_loose_file_in_objdir(get_object_directory(),
2740 add_loose_object,
2741 NULL, NULL, NULL);
2742 }
2743
2744 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2745 {
2746 static struct packed_git *last_found = (void *)1;
2747 struct packed_git *p;
2748
2749 p = (last_found != (void *)1) ? last_found :
2750 get_packed_git(the_repository);
2751
2752 while (p) {
2753 if ((!p->pack_local || p->pack_keep) &&
2754 find_pack_entry_one(oid->hash, p)) {
2755 last_found = p;
2756 return 1;
2757 }
2758 if (p == last_found)
2759 p = get_packed_git(the_repository);
2760 else
2761 p = p->next;
2762 if (p == last_found)
2763 p = p->next;
2764 }
2765 return 0;
2766 }
2767
2768 /*
2769 * Store a list of sha1s that are should not be discarded
2770 * because they are either written too recently, or are
2771 * reachable from another object that was.
2772 *
2773 * This is filled by get_object_list.
2774 */
2775 static struct oid_array recent_objects;
2776
2777 static int loosened_object_can_be_discarded(const struct object_id *oid,
2778 timestamp_t mtime)
2779 {
2780 if (!unpack_unreachable_expiration)
2781 return 0;
2782 if (mtime > unpack_unreachable_expiration)
2783 return 0;
2784 if (oid_array_lookup(&recent_objects, oid) >= 0)
2785 return 0;
2786 return 1;
2787 }
2788
2789 static void loosen_unused_packed_objects(struct rev_info *revs)
2790 {
2791 struct packed_git *p;
2792 uint32_t i;
2793 struct object_id oid;
2794
2795 for (p = get_packed_git(the_repository); p; p = p->next) {
2796 if (!p->pack_local || p->pack_keep)
2797 continue;
2798
2799 if (open_pack_index(p))
2800 die("cannot open pack index");
2801
2802 for (i = 0; i < p->num_objects; i++) {
2803 nth_packed_object_oid(&oid, p, i);
2804 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2805 !has_sha1_pack_kept_or_nonlocal(&oid) &&
2806 !loosened_object_can_be_discarded(&oid, p->mtime))
2807 if (force_object_loose(&oid, p->mtime))
2808 die("unable to force loose object");
2809 }
2810 }
2811 }
2812
2813 /*
2814 * This tracks any options which pack-reuse code expects to be on, or which a
2815 * reader of the pack might not understand, and which would therefore prevent
2816 * blind reuse of what we have on disk.
2817 */
2818 static int pack_options_allow_reuse(void)
2819 {
2820 return pack_to_stdout &&
2821 allow_ofs_delta &&
2822 !ignore_packed_keep &&
2823 (!local || !have_non_local_packs) &&
2824 !incremental;
2825 }
2826
2827 static int get_object_list_from_bitmap(struct rev_info *revs)
2828 {
2829 if (prepare_bitmap_walk(revs) < 0)
2830 return -1;
2831
2832 if (pack_options_allow_reuse() &&
2833 !reuse_partial_packfile_from_bitmap(
2834 &reuse_packfile,
2835 &reuse_packfile_objects,
2836 &reuse_packfile_offset)) {
2837 assert(reuse_packfile_objects);
2838 nr_result += reuse_packfile_objects;
2839 display_progress(progress_state, nr_result);
2840 }
2841
2842 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2843 return 0;
2844 }
2845
2846 static void record_recent_object(struct object *obj,
2847 const char *name,
2848 void *data)
2849 {
2850 oid_array_append(&recent_objects, &obj->oid);
2851 }
2852
2853 static void record_recent_commit(struct commit *commit, void *data)
2854 {
2855 oid_array_append(&recent_objects, &commit->object.oid);
2856 }
2857
2858 static void get_object_list(int ac, const char **av)
2859 {
2860 struct rev_info revs;
2861 char line[1000];
2862 int flags = 0;
2863
2864 init_revisions(&revs, NULL);
2865 save_commit_buffer = 0;
2866 setup_revisions(ac, av, &revs, NULL);
2867
2868 /* make sure shallows are read */
2869 is_repository_shallow();
2870
2871 while (fgets(line, sizeof(line), stdin) != NULL) {
2872 int len = strlen(line);
2873 if (len && line[len - 1] == '\n')
2874 line[--len] = 0;
2875 if (!len)
2876 break;
2877 if (*line == '-') {
2878 if (!strcmp(line, "--not")) {
2879 flags ^= UNINTERESTING;
2880 write_bitmap_index = 0;
2881 continue;
2882 }
2883 if (starts_with(line, "--shallow ")) {
2884 struct object_id oid;
2885 if (get_oid_hex(line + 10, &oid))
2886 die("not an SHA-1 '%s'", line + 10);
2887 register_shallow(&oid);
2888 use_bitmap_index = 0;
2889 continue;
2890 }
2891 die("not a rev '%s'", line);
2892 }
2893 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2894 die("bad revision '%s'", line);
2895 }
2896
2897 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2898 return;
2899
2900 if (prepare_revision_walk(&revs))
2901 die("revision walk setup failed");
2902 mark_edges_uninteresting(&revs, show_edge);
2903
2904 if (!fn_show_object)
2905 fn_show_object = show_object;
2906 traverse_commit_list_filtered(&filter_options, &revs,
2907 show_commit, fn_show_object, NULL,
2908 NULL);
2909
2910 if (unpack_unreachable_expiration) {
2911 revs.ignore_missing_links = 1;
2912 if (add_unseen_recent_objects_to_traversal(&revs,
2913 unpack_unreachable_expiration))
2914 die("unable to add recent objects");
2915 if (prepare_revision_walk(&revs))
2916 die("revision walk setup failed");
2917 traverse_commit_list(&revs, record_recent_commit,
2918 record_recent_object, NULL);
2919 }
2920
2921 if (keep_unreachable)
2922 add_objects_in_unpacked_packs(&revs);
2923 if (pack_loose_unreachable)
2924 add_unreachable_loose_objects();
2925 if (unpack_unreachable)
2926 loosen_unused_packed_objects(&revs);
2927
2928 oid_array_clear(&recent_objects);
2929 }
2930
2931 static int option_parse_index_version(const struct option *opt,
2932 const char *arg, int unset)
2933 {
2934 char *c;
2935 const char *val = arg;
2936 pack_idx_opts.version = strtoul(val, &c, 10);
2937 if (pack_idx_opts.version > 2)
2938 die(_("unsupported index version %s"), val);
2939 if (*c == ',' && c[1])
2940 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2941 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2942 die(_("bad index version '%s'"), val);
2943 return 0;
2944 }
2945
2946 static int option_parse_unpack_unreachable(const struct option *opt,
2947 const char *arg, int unset)
2948 {
2949 if (unset) {
2950 unpack_unreachable = 0;
2951 unpack_unreachable_expiration = 0;
2952 }
2953 else {
2954 unpack_unreachable = 1;
2955 if (arg)
2956 unpack_unreachable_expiration = approxidate(arg);
2957 }
2958 return 0;
2959 }
2960
2961 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2962 {
2963 int use_internal_rev_list = 0;
2964 int thin = 0;
2965 int shallow = 0;
2966 int all_progress_implied = 0;
2967 struct argv_array rp = ARGV_ARRAY_INIT;
2968 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2969 int rev_list_index = 0;
2970 struct option pack_objects_options[] = {
2971 OPT_SET_INT('q', "quiet", &progress,
2972 N_("do not show progress meter"), 0),
2973 OPT_SET_INT(0, "progress", &progress,
2974 N_("show progress meter"), 1),
2975 OPT_SET_INT(0, "all-progress", &progress,
2976 N_("show progress meter during object writing phase"), 2),
2977 OPT_BOOL(0, "all-progress-implied",
2978 &all_progress_implied,
2979 N_("similar to --all-progress when progress meter is shown")),
2980 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2981 N_("write the pack index file in the specified idx format version"),
2982 0, option_parse_index_version },
2983 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2984 N_("maximum size of each output pack file")),
2985 OPT_BOOL(0, "local", &local,
2986 N_("ignore borrowed objects from alternate object store")),
2987 OPT_BOOL(0, "incremental", &incremental,
2988 N_("ignore packed objects")),
2989 OPT_INTEGER(0, "window", &window,
2990 N_("limit pack window by objects")),
2991 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2992 N_("limit pack window by memory in addition to object limit")),
2993 OPT_INTEGER(0, "depth", &depth,
2994 N_("maximum length of delta chain allowed in the resulting pack")),
2995 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2996 N_("reuse existing deltas")),
2997 OPT_BOOL(0, "reuse-object", &reuse_object,
2998 N_("reuse existing objects")),
2999 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3000 N_("use OFS_DELTA objects")),
3001 OPT_INTEGER(0, "threads", &delta_search_threads,
3002 N_("use threads when searching for best delta matches")),
3003 OPT_BOOL(0, "non-empty", &non_empty,
3004 N_("do not create an empty pack output")),
3005 OPT_BOOL(0, "revs", &use_internal_rev_list,
3006 N_("read revision arguments from standard input")),
3007 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
3008 N_("limit the objects to those that are not yet packed"),
3009 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3010 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
3011 N_("include objects reachable from any reference"),
3012 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3013 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
3014 N_("include objects referred by reflog entries"),
3015 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3016 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
3017 N_("include objects referred to by the index"),
3018 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3019 OPT_BOOL(0, "stdout", &pack_to_stdout,
3020 N_("output pack to stdout")),
3021 OPT_BOOL(0, "include-tag", &include_tag,
3022 N_("include tag objects that refer to objects to be packed")),
3023 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3024 N_("keep unreachable objects")),
3025 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3026 N_("pack loose unreachable objects")),
3027 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3028 N_("unpack unreachable objects newer than <time>"),
3029 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3030 OPT_BOOL(0, "thin", &thin,
3031 N_("create thin packs")),
3032 OPT_BOOL(0, "shallow", &shallow,
3033 N_("create packs suitable for shallow fetches")),
3034 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
3035 N_("ignore packs that have companion .keep file")),
3036 OPT_INTEGER(0, "compression", &pack_compression_level,
3037 N_("pack compression level")),
3038 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3039 N_("do not hide commits by grafts"), 0),
3040 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3041 N_("use a bitmap index if available to speed up counting objects")),
3042 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3043 N_("write a bitmap index together with the pack index")),
3044 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3045 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3046 N_("handling for missing objects"), PARSE_OPT_NONEG,
3047 option_parse_missing_action },
3048 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3049 N_("do not pack objects in promisor packfiles")),
3050 OPT_END(),
3051 };
3052
3053 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3054 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3055
3056 check_replace_refs = 0;
3057
3058 reset_pack_idx_option(&pack_idx_opts);
3059 git_config(git_pack_config, NULL);
3060
3061 progress = isatty(2);
3062 argc = parse_options(argc, argv, prefix, pack_objects_options,
3063 pack_usage, 0);
3064
3065 if (argc) {
3066 base_name = argv[0];
3067 argc--;
3068 }
3069 if (pack_to_stdout != !base_name || argc)
3070 usage_with_options(pack_usage, pack_objects_options);
3071
3072 if (depth >= (1 << OE_DEPTH_BITS)) {
3073 warning(_("delta chain depth %d is too deep, forcing %d"),
3074 depth, (1 << OE_DEPTH_BITS) - 1);
3075 depth = (1 << OE_DEPTH_BITS) - 1;
3076 }
3077
3078 argv_array_push(&rp, "pack-objects");
3079 if (thin) {
3080 use_internal_rev_list = 1;
3081 argv_array_push(&rp, shallow
3082 ? "--objects-edge-aggressive"
3083 : "--objects-edge");
3084 } else
3085 argv_array_push(&rp, "--objects");
3086
3087 if (rev_list_all) {
3088 use_internal_rev_list = 1;
3089 argv_array_push(&rp, "--all");
3090 }
3091 if (rev_list_reflog) {
3092 use_internal_rev_list = 1;
3093 argv_array_push(&rp, "--reflog");
3094 }
3095 if (rev_list_index) {
3096 use_internal_rev_list = 1;
3097 argv_array_push(&rp, "--indexed-objects");
3098 }
3099 if (rev_list_unpacked) {
3100 use_internal_rev_list = 1;
3101 argv_array_push(&rp, "--unpacked");
3102 }
3103
3104 if (exclude_promisor_objects) {
3105 use_internal_rev_list = 1;
3106 fetch_if_missing = 0;
3107 argv_array_push(&rp, "--exclude-promisor-objects");
3108 }
3109
3110 if (!reuse_object)
3111 reuse_delta = 0;
3112 if (pack_compression_level == -1)
3113 pack_compression_level = Z_DEFAULT_COMPRESSION;
3114 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3115 die("bad pack compression level %d", pack_compression_level);
3116
3117 if (!delta_search_threads) /* --threads=0 means autodetect */
3118 delta_search_threads = online_cpus();
3119
3120 #ifdef NO_PTHREADS
3121 if (delta_search_threads != 1)
3122 warning("no threads support, ignoring --threads");
3123 #endif
3124 if (!pack_to_stdout && !pack_size_limit)
3125 pack_size_limit = pack_size_limit_cfg;
3126 if (pack_to_stdout && pack_size_limit)
3127 die("--max-pack-size cannot be used to build a pack for transfer.");
3128 if (pack_size_limit && pack_size_limit < 1024*1024) {
3129 warning("minimum pack size limit is 1 MiB");
3130 pack_size_limit = 1024*1024;
3131 }
3132
3133 if (!pack_to_stdout && thin)
3134 die("--thin cannot be used to build an indexable pack.");
3135
3136 if (keep_unreachable && unpack_unreachable)
3137 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3138 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3139 unpack_unreachable_expiration = 0;
3140
3141 if (filter_options.choice) {
3142 if (!pack_to_stdout)
3143 die("cannot use --filter without --stdout.");
3144 use_bitmap_index = 0;
3145 }
3146
3147 /*
3148 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3149 *
3150 * - to produce good pack (with bitmap index not-yet-packed objects are
3151 * packed in suboptimal order).
3152 *
3153 * - to use more robust pack-generation codepath (avoiding possible
3154 * bugs in bitmap code and possible bitmap index corruption).
3155 */
3156 if (!pack_to_stdout)
3157 use_bitmap_index_default = 0;
3158
3159 if (use_bitmap_index < 0)
3160 use_bitmap_index = use_bitmap_index_default;
3161
3162 /* "hard" reasons not to use bitmaps; these just won't work at all */
3163 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3164 use_bitmap_index = 0;
3165
3166 if (pack_to_stdout || !rev_list_all)
3167 write_bitmap_index = 0;
3168
3169 if (progress && all_progress_implied)
3170 progress = 2;
3171
3172 if (ignore_packed_keep) {
3173 struct packed_git *p;
3174 for (p = get_packed_git(the_repository); p; p = p->next)
3175 if (p->pack_local && p->pack_keep)
3176 break;
3177 if (!p) /* no keep-able packs found */
3178 ignore_packed_keep = 0;
3179 }
3180 if (local) {
3181 /*
3182 * unlike ignore_packed_keep above, we do not want to
3183 * unset "local" based on looking at packs, as it
3184 * also covers non-local objects
3185 */
3186 struct packed_git *p;
3187 for (p = get_packed_git(the_repository); p; p = p->next) {
3188 if (!p->pack_local) {
3189 have_non_local_packs = 1;
3190 break;
3191 }
3192 }
3193 }
3194
3195 if (progress)
3196 progress_state = start_progress(_("Counting objects"), 0);
3197 if (!use_internal_rev_list)
3198 read_object_list_from_stdin();
3199 else {
3200 get_object_list(rp.argc, rp.argv);
3201 argv_array_clear(&rp);
3202 }
3203 cleanup_preferred_base();
3204 if (include_tag && nr_result)
3205 for_each_ref(add_ref_tag, NULL);
3206 stop_progress(&progress_state);
3207
3208 if (non_empty && !nr_result)
3209 return 0;
3210 if (nr_result)
3211 prepare_pack(window, depth);
3212 write_pack_file();
3213 if (progress)
3214 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3215 " reused %"PRIu32" (delta %"PRIu32")\n",
3216 written, written_delta, reused, reused_delta);
3217 return 0;
3218 }