The sixth batch for 2.26
[git.git] / builtin / pack-objects.c
blob940fbcb7b375b69fe8094d9d03f313a279d2df44
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 "delta-islands.h"
28 #include "reachable.h"
29 #include "sha1-array.h"
30 #include "argv-array.h"
31 #include "list.h"
32 #include "packfile.h"
33 #include "object-store.h"
34 #include "dir.h"
35 #include "midx.h"
36 #include "trace2.h"
38 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
39 #define SIZE(obj) oe_size(&to_pack, obj)
40 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
41 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
42 #define DELTA(obj) oe_delta(&to_pack, obj)
43 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
44 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
45 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
46 #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
47 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
48 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
49 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
51 static const char *pack_usage[] = {
52 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
53 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
54 NULL
58 * Objects we are going to pack are collected in the `to_pack` structure.
59 * It contains an array (dynamically expanded) of the object data, and a map
60 * that can resolve SHA1s to their position in the array.
62 static struct packing_data to_pack;
64 static struct pack_idx_entry **written_list;
65 static uint32_t nr_result, nr_written, nr_seen;
66 static struct bitmap_index *bitmap_git;
67 static uint32_t write_layer;
69 static int non_empty;
70 static int reuse_delta = 1, reuse_object = 1;
71 static int keep_unreachable, unpack_unreachable, include_tag;
72 static timestamp_t unpack_unreachable_expiration;
73 static int pack_loose_unreachable;
74 static int local;
75 static int have_non_local_packs;
76 static int incremental;
77 static int ignore_packed_keep_on_disk;
78 static int ignore_packed_keep_in_core;
79 static int allow_ofs_delta;
80 static struct pack_idx_option pack_idx_opts;
81 static const char *base_name;
82 static int progress = 1;
83 static int window = 10;
84 static unsigned long pack_size_limit;
85 static int depth = 50;
86 static int delta_search_threads;
87 static int pack_to_stdout;
88 static int sparse;
89 static int thin;
90 static int num_preferred_base;
91 static struct progress *progress_state;
93 static struct packed_git *reuse_packfile;
94 static uint32_t reuse_packfile_objects;
95 static struct bitmap *reuse_packfile_bitmap;
97 static int use_bitmap_index_default = 1;
98 static int use_bitmap_index = -1;
99 static int allow_pack_reuse = 1;
100 static enum {
101 WRITE_BITMAP_FALSE = 0,
102 WRITE_BITMAP_QUIET,
103 WRITE_BITMAP_TRUE,
104 } write_bitmap_index;
105 static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
107 static int exclude_promisor_objects;
109 static int use_delta_islands;
111 static unsigned long delta_cache_size = 0;
112 static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
113 static unsigned long cache_max_small_delta_size = 1000;
115 static unsigned long window_memory_limit = 0;
117 static struct list_objects_filter_options filter_options;
119 enum missing_action {
120 MA_ERROR = 0, /* fail if any missing objects are encountered */
121 MA_ALLOW_ANY, /* silently allow ALL missing objects */
122 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
124 static enum missing_action arg_missing_action;
125 static show_object_fn fn_show_object;
128 * stats
130 static uint32_t written, written_delta;
131 static uint32_t reused, reused_delta;
134 * Indexed commits
136 static struct commit **indexed_commits;
137 static unsigned int indexed_commits_nr;
138 static unsigned int indexed_commits_alloc;
140 static void index_commit_for_bitmap(struct commit *commit)
142 if (indexed_commits_nr >= indexed_commits_alloc) {
143 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
144 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
147 indexed_commits[indexed_commits_nr++] = commit;
150 static void *get_delta(struct object_entry *entry)
152 unsigned long size, base_size, delta_size;
153 void *buf, *base_buf, *delta_buf;
154 enum object_type type;
156 buf = read_object_file(&entry->idx.oid, &type, &size);
157 if (!buf)
158 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
159 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
160 &base_size);
161 if (!base_buf)
162 die("unable to read %s",
163 oid_to_hex(&DELTA(entry)->idx.oid));
164 delta_buf = diff_delta(base_buf, base_size,
165 buf, size, &delta_size, 0);
167 * We successfully computed this delta once but dropped it for
168 * memory reasons. Something is very wrong if this time we
169 * recompute and create a different delta.
171 if (!delta_buf || delta_size != DELTA_SIZE(entry))
172 BUG("delta size changed");
173 free(buf);
174 free(base_buf);
175 return delta_buf;
178 static unsigned long do_compress(void **pptr, unsigned long size)
180 git_zstream stream;
181 void *in, *out;
182 unsigned long maxsize;
184 git_deflate_init(&stream, pack_compression_level);
185 maxsize = git_deflate_bound(&stream, size);
187 in = *pptr;
188 out = xmalloc(maxsize);
189 *pptr = out;
191 stream.next_in = in;
192 stream.avail_in = size;
193 stream.next_out = out;
194 stream.avail_out = maxsize;
195 while (git_deflate(&stream, Z_FINISH) == Z_OK)
196 ; /* nothing */
197 git_deflate_end(&stream);
199 free(in);
200 return stream.total_out;
203 static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
204 const struct object_id *oid)
206 git_zstream stream;
207 unsigned char ibuf[1024 * 16];
208 unsigned char obuf[1024 * 16];
209 unsigned long olen = 0;
211 git_deflate_init(&stream, pack_compression_level);
213 for (;;) {
214 ssize_t readlen;
215 int zret = Z_OK;
216 readlen = read_istream(st, ibuf, sizeof(ibuf));
217 if (readlen == -1)
218 die(_("unable to read %s"), oid_to_hex(oid));
220 stream.next_in = ibuf;
221 stream.avail_in = readlen;
222 while ((stream.avail_in || readlen == 0) &&
223 (zret == Z_OK || zret == Z_BUF_ERROR)) {
224 stream.next_out = obuf;
225 stream.avail_out = sizeof(obuf);
226 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
227 hashwrite(f, obuf, stream.next_out - obuf);
228 olen += stream.next_out - obuf;
230 if (stream.avail_in)
231 die(_("deflate error (%d)"), zret);
232 if (readlen == 0) {
233 if (zret != Z_STREAM_END)
234 die(_("deflate error (%d)"), zret);
235 break;
238 git_deflate_end(&stream);
239 return olen;
243 * we are going to reuse the existing object data as is. make
244 * sure it is not corrupt.
246 static int check_pack_inflate(struct packed_git *p,
247 struct pack_window **w_curs,
248 off_t offset,
249 off_t len,
250 unsigned long expect)
252 git_zstream stream;
253 unsigned char fakebuf[4096], *in;
254 int st;
256 memset(&stream, 0, sizeof(stream));
257 git_inflate_init(&stream);
258 do {
259 in = use_pack(p, w_curs, offset, &stream.avail_in);
260 stream.next_in = in;
261 stream.next_out = fakebuf;
262 stream.avail_out = sizeof(fakebuf);
263 st = git_inflate(&stream, Z_FINISH);
264 offset += stream.next_in - in;
265 } while (st == Z_OK || st == Z_BUF_ERROR);
266 git_inflate_end(&stream);
267 return (st == Z_STREAM_END &&
268 stream.total_out == expect &&
269 stream.total_in == len) ? 0 : -1;
272 static void copy_pack_data(struct hashfile *f,
273 struct packed_git *p,
274 struct pack_window **w_curs,
275 off_t offset,
276 off_t len)
278 unsigned char *in;
279 unsigned long avail;
281 while (len) {
282 in = use_pack(p, w_curs, offset, &avail);
283 if (avail > len)
284 avail = (unsigned long)len;
285 hashwrite(f, in, avail);
286 offset += avail;
287 len -= avail;
291 /* Return 0 if we will bust the pack-size limit */
292 static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
293 unsigned long limit, int usable_delta)
295 unsigned long size, datalen;
296 unsigned char header[MAX_PACK_OBJECT_HEADER],
297 dheader[MAX_PACK_OBJECT_HEADER];
298 unsigned hdrlen;
299 enum object_type type;
300 void *buf;
301 struct git_istream *st = NULL;
302 const unsigned hashsz = the_hash_algo->rawsz;
304 if (!usable_delta) {
305 if (oe_type(entry) == OBJ_BLOB &&
306 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
307 (st = open_istream(the_repository, &entry->idx.oid, &type,
308 &size, NULL)) != NULL)
309 buf = NULL;
310 else {
311 buf = read_object_file(&entry->idx.oid, &type, &size);
312 if (!buf)
313 die(_("unable to read %s"),
314 oid_to_hex(&entry->idx.oid));
317 * make sure no cached delta data remains from a
318 * previous attempt before a pack split occurred.
320 FREE_AND_NULL(entry->delta_data);
321 entry->z_delta_size = 0;
322 } else if (entry->delta_data) {
323 size = DELTA_SIZE(entry);
324 buf = entry->delta_data;
325 entry->delta_data = NULL;
326 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
327 OBJ_OFS_DELTA : OBJ_REF_DELTA;
328 } else {
329 buf = get_delta(entry);
330 size = DELTA_SIZE(entry);
331 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
332 OBJ_OFS_DELTA : OBJ_REF_DELTA;
335 if (st) /* large blob case, just assume we don't compress well */
336 datalen = size;
337 else if (entry->z_delta_size)
338 datalen = entry->z_delta_size;
339 else
340 datalen = do_compress(&buf, size);
343 * The object header is a byte of 'type' followed by zero or
344 * more bytes of length.
346 hdrlen = encode_in_pack_object_header(header, sizeof(header),
347 type, size);
349 if (type == OBJ_OFS_DELTA) {
351 * Deltas with relative base contain an additional
352 * encoding of the relative offset for the delta
353 * base from this object's position in the pack.
355 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
356 unsigned pos = sizeof(dheader) - 1;
357 dheader[pos] = ofs & 127;
358 while (ofs >>= 7)
359 dheader[--pos] = 128 | (--ofs & 127);
360 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
361 if (st)
362 close_istream(st);
363 free(buf);
364 return 0;
366 hashwrite(f, header, hdrlen);
367 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
368 hdrlen += sizeof(dheader) - pos;
369 } else if (type == OBJ_REF_DELTA) {
371 * Deltas with a base reference contain
372 * additional bytes for the base object ID.
374 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
375 if (st)
376 close_istream(st);
377 free(buf);
378 return 0;
380 hashwrite(f, header, hdrlen);
381 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
382 hdrlen += hashsz;
383 } else {
384 if (limit && hdrlen + datalen + hashsz >= limit) {
385 if (st)
386 close_istream(st);
387 free(buf);
388 return 0;
390 hashwrite(f, header, hdrlen);
392 if (st) {
393 datalen = write_large_blob_data(st, f, &entry->idx.oid);
394 close_istream(st);
395 } else {
396 hashwrite(f, buf, datalen);
397 free(buf);
400 return hdrlen + datalen;
403 /* Return 0 if we will bust the pack-size limit */
404 static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
405 unsigned long limit, int usable_delta)
407 struct packed_git *p = IN_PACK(entry);
408 struct pack_window *w_curs = NULL;
409 struct revindex_entry *revidx;
410 off_t offset;
411 enum object_type type = oe_type(entry);
412 off_t datalen;
413 unsigned char header[MAX_PACK_OBJECT_HEADER],
414 dheader[MAX_PACK_OBJECT_HEADER];
415 unsigned hdrlen;
416 const unsigned hashsz = the_hash_algo->rawsz;
417 unsigned long entry_size = SIZE(entry);
419 if (DELTA(entry))
420 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
421 OBJ_OFS_DELTA : OBJ_REF_DELTA;
422 hdrlen = encode_in_pack_object_header(header, sizeof(header),
423 type, entry_size);
425 offset = entry->in_pack_offset;
426 revidx = find_pack_revindex(p, offset);
427 datalen = revidx[1].offset - offset;
428 if (!pack_to_stdout && p->index_version > 1 &&
429 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
430 error(_("bad packed object CRC for %s"),
431 oid_to_hex(&entry->idx.oid));
432 unuse_pack(&w_curs);
433 return write_no_reuse_object(f, entry, limit, usable_delta);
436 offset += entry->in_pack_header_size;
437 datalen -= entry->in_pack_header_size;
439 if (!pack_to_stdout && p->index_version == 1 &&
440 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
441 error(_("corrupt packed object for %s"),
442 oid_to_hex(&entry->idx.oid));
443 unuse_pack(&w_curs);
444 return write_no_reuse_object(f, entry, limit, usable_delta);
447 if (type == OBJ_OFS_DELTA) {
448 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
449 unsigned pos = sizeof(dheader) - 1;
450 dheader[pos] = ofs & 127;
451 while (ofs >>= 7)
452 dheader[--pos] = 128 | (--ofs & 127);
453 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
454 unuse_pack(&w_curs);
455 return 0;
457 hashwrite(f, header, hdrlen);
458 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
459 hdrlen += sizeof(dheader) - pos;
460 reused_delta++;
461 } else if (type == OBJ_REF_DELTA) {
462 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
463 unuse_pack(&w_curs);
464 return 0;
466 hashwrite(f, header, hdrlen);
467 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
468 hdrlen += hashsz;
469 reused_delta++;
470 } else {
471 if (limit && hdrlen + datalen + hashsz >= limit) {
472 unuse_pack(&w_curs);
473 return 0;
475 hashwrite(f, header, hdrlen);
477 copy_pack_data(f, p, &w_curs, offset, datalen);
478 unuse_pack(&w_curs);
479 reused++;
480 return hdrlen + datalen;
483 /* Return 0 if we will bust the pack-size limit */
484 static off_t write_object(struct hashfile *f,
485 struct object_entry *entry,
486 off_t write_offset)
488 unsigned long limit;
489 off_t len;
490 int usable_delta, to_reuse;
492 if (!pack_to_stdout)
493 crc32_begin(f);
495 /* apply size limit if limited packsize and not first object */
496 if (!pack_size_limit || !nr_written)
497 limit = 0;
498 else if (pack_size_limit <= write_offset)
500 * the earlier object did not fit the limit; avoid
501 * mistaking this with unlimited (i.e. limit = 0).
503 limit = 1;
504 else
505 limit = pack_size_limit - write_offset;
507 if (!DELTA(entry))
508 usable_delta = 0; /* no delta */
509 else if (!pack_size_limit)
510 usable_delta = 1; /* unlimited packfile */
511 else if (DELTA(entry)->idx.offset == (off_t)-1)
512 usable_delta = 0; /* base was written to another pack */
513 else if (DELTA(entry)->idx.offset)
514 usable_delta = 1; /* base already exists in this pack */
515 else
516 usable_delta = 0; /* base could end up in another pack */
518 if (!reuse_object)
519 to_reuse = 0; /* explicit */
520 else if (!IN_PACK(entry))
521 to_reuse = 0; /* can't reuse what we don't have */
522 else if (oe_type(entry) == OBJ_REF_DELTA ||
523 oe_type(entry) == OBJ_OFS_DELTA)
524 /* check_object() decided it for us ... */
525 to_reuse = usable_delta;
526 /* ... but pack split may override that */
527 else if (oe_type(entry) != entry->in_pack_type)
528 to_reuse = 0; /* pack has delta which is unusable */
529 else if (DELTA(entry))
530 to_reuse = 0; /* we want to pack afresh */
531 else
532 to_reuse = 1; /* we have it in-pack undeltified,
533 * and we do not need to deltify it.
536 if (!to_reuse)
537 len = write_no_reuse_object(f, entry, limit, usable_delta);
538 else
539 len = write_reuse_object(f, entry, limit, usable_delta);
540 if (!len)
541 return 0;
543 if (usable_delta)
544 written_delta++;
545 written++;
546 if (!pack_to_stdout)
547 entry->idx.crc32 = crc32_end(f);
548 return len;
551 enum write_one_status {
552 WRITE_ONE_SKIP = -1, /* already written */
553 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
554 WRITE_ONE_WRITTEN = 1, /* normal */
555 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
558 static enum write_one_status write_one(struct hashfile *f,
559 struct object_entry *e,
560 off_t *offset)
562 off_t size;
563 int recursing;
566 * we set offset to 1 (which is an impossible value) to mark
567 * the fact that this object is involved in "write its base
568 * first before writing a deltified object" recursion.
570 recursing = (e->idx.offset == 1);
571 if (recursing) {
572 warning(_("recursive delta detected for object %s"),
573 oid_to_hex(&e->idx.oid));
574 return WRITE_ONE_RECURSIVE;
575 } else if (e->idx.offset || e->preferred_base) {
576 /* offset is non zero if object is written already. */
577 return WRITE_ONE_SKIP;
580 /* if we are deltified, write out base object first. */
581 if (DELTA(e)) {
582 e->idx.offset = 1; /* now recurse */
583 switch (write_one(f, DELTA(e), offset)) {
584 case WRITE_ONE_RECURSIVE:
585 /* we cannot depend on this one */
586 SET_DELTA(e, NULL);
587 break;
588 default:
589 break;
590 case WRITE_ONE_BREAK:
591 e->idx.offset = recursing;
592 return WRITE_ONE_BREAK;
596 e->idx.offset = *offset;
597 size = write_object(f, e, *offset);
598 if (!size) {
599 e->idx.offset = recursing;
600 return WRITE_ONE_BREAK;
602 written_list[nr_written++] = &e->idx;
604 /* make sure off_t is sufficiently large not to wrap */
605 if (signed_add_overflows(*offset, size))
606 die(_("pack too large for current definition of off_t"));
607 *offset += size;
608 return WRITE_ONE_WRITTEN;
611 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
612 void *cb_data)
614 struct object_id peeled;
615 struct object_entry *entry = packlist_find(&to_pack, oid);
617 if (entry)
618 entry->tagged = 1;
619 if (!peel_ref(path, &peeled)) {
620 entry = packlist_find(&to_pack, &peeled);
621 if (entry)
622 entry->tagged = 1;
624 return 0;
627 static inline void add_to_write_order(struct object_entry **wo,
628 unsigned int *endp,
629 struct object_entry *e)
631 if (e->filled || oe_layer(&to_pack, e) != write_layer)
632 return;
633 wo[(*endp)++] = e;
634 e->filled = 1;
637 static void add_descendants_to_write_order(struct object_entry **wo,
638 unsigned int *endp,
639 struct object_entry *e)
641 int add_to_order = 1;
642 while (e) {
643 if (add_to_order) {
644 struct object_entry *s;
645 /* add this node... */
646 add_to_write_order(wo, endp, e);
647 /* all its siblings... */
648 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
649 add_to_write_order(wo, endp, s);
652 /* drop down a level to add left subtree nodes if possible */
653 if (DELTA_CHILD(e)) {
654 add_to_order = 1;
655 e = DELTA_CHILD(e);
656 } else {
657 add_to_order = 0;
658 /* our sibling might have some children, it is next */
659 if (DELTA_SIBLING(e)) {
660 e = DELTA_SIBLING(e);
661 continue;
663 /* go back to our parent node */
664 e = DELTA(e);
665 while (e && !DELTA_SIBLING(e)) {
666 /* we're on the right side of a subtree, keep
667 * going up until we can go right again */
668 e = DELTA(e);
670 if (!e) {
671 /* done- we hit our original root node */
672 return;
674 /* pass it off to sibling at this level */
675 e = DELTA_SIBLING(e);
680 static void add_family_to_write_order(struct object_entry **wo,
681 unsigned int *endp,
682 struct object_entry *e)
684 struct object_entry *root;
686 for (root = e; DELTA(root); root = DELTA(root))
687 ; /* nothing */
688 add_descendants_to_write_order(wo, endp, root);
691 static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
693 unsigned int i, last_untagged;
694 struct object_entry *objects = to_pack.objects;
696 for (i = 0; i < to_pack.nr_objects; i++) {
697 if (objects[i].tagged)
698 break;
699 add_to_write_order(wo, wo_end, &objects[i]);
701 last_untagged = i;
704 * Then fill all the tagged tips.
706 for (; i < to_pack.nr_objects; i++) {
707 if (objects[i].tagged)
708 add_to_write_order(wo, wo_end, &objects[i]);
712 * And then all remaining commits and tags.
714 for (i = last_untagged; i < to_pack.nr_objects; i++) {
715 if (oe_type(&objects[i]) != OBJ_COMMIT &&
716 oe_type(&objects[i]) != OBJ_TAG)
717 continue;
718 add_to_write_order(wo, wo_end, &objects[i]);
722 * And then all the trees.
724 for (i = last_untagged; i < to_pack.nr_objects; i++) {
725 if (oe_type(&objects[i]) != OBJ_TREE)
726 continue;
727 add_to_write_order(wo, wo_end, &objects[i]);
731 * Finally all the rest in really tight order
733 for (i = last_untagged; i < to_pack.nr_objects; i++) {
734 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
735 add_family_to_write_order(wo, wo_end, &objects[i]);
739 static struct object_entry **compute_write_order(void)
741 uint32_t max_layers = 1;
742 unsigned int i, wo_end;
744 struct object_entry **wo;
745 struct object_entry *objects = to_pack.objects;
747 for (i = 0; i < to_pack.nr_objects; i++) {
748 objects[i].tagged = 0;
749 objects[i].filled = 0;
750 SET_DELTA_CHILD(&objects[i], NULL);
751 SET_DELTA_SIBLING(&objects[i], NULL);
755 * Fully connect delta_child/delta_sibling network.
756 * Make sure delta_sibling is sorted in the original
757 * recency order.
759 for (i = to_pack.nr_objects; i > 0;) {
760 struct object_entry *e = &objects[--i];
761 if (!DELTA(e))
762 continue;
763 /* Mark me as the first child */
764 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
765 SET_DELTA_CHILD(DELTA(e), e);
769 * Mark objects that are at the tip of tags.
771 for_each_tag_ref(mark_tagged, NULL);
773 if (use_delta_islands)
774 max_layers = compute_pack_layers(&to_pack);
776 ALLOC_ARRAY(wo, to_pack.nr_objects);
777 wo_end = 0;
779 for (; write_layer < max_layers; ++write_layer)
780 compute_layer_order(wo, &wo_end);
782 if (wo_end != to_pack.nr_objects)
783 die(_("ordered %u objects, expected %"PRIu32),
784 wo_end, to_pack.nr_objects);
786 return wo;
791 * A reused set of objects. All objects in a chunk have the same
792 * relative position in the original packfile and the generated
793 * packfile.
796 static struct reused_chunk {
797 /* The offset of the first object of this chunk in the original
798 * packfile. */
799 off_t original;
800 /* The offset of the first object of this chunk in the generated
801 * packfile minus "original". */
802 off_t difference;
803 } *reused_chunks;
804 static int reused_chunks_nr;
805 static int reused_chunks_alloc;
807 static void record_reused_object(off_t where, off_t offset)
809 if (reused_chunks_nr && reused_chunks[reused_chunks_nr-1].difference == offset)
810 return;
812 ALLOC_GROW(reused_chunks, reused_chunks_nr + 1,
813 reused_chunks_alloc);
814 reused_chunks[reused_chunks_nr].original = where;
815 reused_chunks[reused_chunks_nr].difference = offset;
816 reused_chunks_nr++;
820 * Binary search to find the chunk that "where" is in. Note
821 * that we're not looking for an exact match, just the first
822 * chunk that contains it (which implicitly ends at the start
823 * of the next chunk.
825 static off_t find_reused_offset(off_t where)
827 int lo = 0, hi = reused_chunks_nr;
828 while (lo < hi) {
829 int mi = lo + ((hi - lo) / 2);
830 if (where == reused_chunks[mi].original)
831 return reused_chunks[mi].difference;
832 if (where < reused_chunks[mi].original)
833 hi = mi;
834 else
835 lo = mi + 1;
839 * The first chunk starts at zero, so we can't have gone below
840 * there.
842 assert(lo);
843 return reused_chunks[lo-1].difference;
846 static void write_reused_pack_one(size_t pos, struct hashfile *out,
847 struct pack_window **w_curs)
849 off_t offset, next, cur;
850 enum object_type type;
851 unsigned long size;
853 offset = reuse_packfile->revindex[pos].offset;
854 next = reuse_packfile->revindex[pos + 1].offset;
856 record_reused_object(offset, offset - hashfile_total(out));
858 cur = offset;
859 type = unpack_object_header(reuse_packfile, w_curs, &cur, &size);
860 assert(type >= 0);
862 if (type == OBJ_OFS_DELTA) {
863 off_t base_offset;
864 off_t fixup;
866 unsigned char header[MAX_PACK_OBJECT_HEADER];
867 unsigned len;
869 base_offset = get_delta_base(reuse_packfile, w_curs, &cur, type, offset);
870 assert(base_offset != 0);
872 /* Convert to REF_DELTA if we must... */
873 if (!allow_ofs_delta) {
874 int base_pos = find_revindex_position(reuse_packfile, base_offset);
875 const unsigned char *base_sha1 =
876 nth_packed_object_sha1(reuse_packfile,
877 reuse_packfile->revindex[base_pos].nr);
879 len = encode_in_pack_object_header(header, sizeof(header),
880 OBJ_REF_DELTA, size);
881 hashwrite(out, header, len);
882 hashwrite(out, base_sha1, 20);
883 copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
884 return;
887 /* Otherwise see if we need to rewrite the offset... */
888 fixup = find_reused_offset(offset) -
889 find_reused_offset(base_offset);
890 if (fixup) {
891 unsigned char ofs_header[10];
892 unsigned i, ofs_len;
893 off_t ofs = offset - base_offset - fixup;
895 len = encode_in_pack_object_header(header, sizeof(header),
896 OBJ_OFS_DELTA, size);
898 i = sizeof(ofs_header) - 1;
899 ofs_header[i] = ofs & 127;
900 while (ofs >>= 7)
901 ofs_header[--i] = 128 | (--ofs & 127);
903 ofs_len = sizeof(ofs_header) - i;
905 hashwrite(out, header, len);
906 hashwrite(out, ofs_header + sizeof(ofs_header) - ofs_len, ofs_len);
907 copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur);
908 return;
911 /* ...otherwise we have no fixup, and can write it verbatim */
914 copy_pack_data(out, reuse_packfile, w_curs, offset, next - offset);
917 static size_t write_reused_pack_verbatim(struct hashfile *out,
918 struct pack_window **w_curs)
920 size_t pos = 0;
922 while (pos < reuse_packfile_bitmap->word_alloc &&
923 reuse_packfile_bitmap->words[pos] == (eword_t)~0)
924 pos++;
926 if (pos) {
927 off_t to_write;
929 written = (pos * BITS_IN_EWORD);
930 to_write = reuse_packfile->revindex[written].offset
931 - sizeof(struct pack_header);
933 /* We're recording one chunk, not one object. */
934 record_reused_object(sizeof(struct pack_header), 0);
935 hashflush(out);
936 copy_pack_data(out, reuse_packfile, w_curs,
937 sizeof(struct pack_header), to_write);
939 display_progress(progress_state, written);
941 return pos;
944 static void write_reused_pack(struct hashfile *f)
946 size_t i = 0;
947 uint32_t offset;
948 struct pack_window *w_curs = NULL;
950 if (allow_ofs_delta)
951 i = write_reused_pack_verbatim(f, &w_curs);
953 for (; i < reuse_packfile_bitmap->word_alloc; ++i) {
954 eword_t word = reuse_packfile_bitmap->words[i];
955 size_t pos = (i * BITS_IN_EWORD);
957 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
958 if ((word >> offset) == 0)
959 break;
961 offset += ewah_bit_ctz64(word >> offset);
962 write_reused_pack_one(pos + offset, f, &w_curs);
963 display_progress(progress_state, ++written);
967 unuse_pack(&w_curs);
970 static const char no_split_warning[] = N_(
971 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
974 static void write_pack_file(void)
976 uint32_t i = 0, j;
977 struct hashfile *f;
978 off_t offset;
979 uint32_t nr_remaining = nr_result;
980 time_t last_mtime = 0;
981 struct object_entry **write_order;
983 if (progress > pack_to_stdout)
984 progress_state = start_progress(_("Writing objects"), nr_result);
985 ALLOC_ARRAY(written_list, to_pack.nr_objects);
986 write_order = compute_write_order();
988 do {
989 struct object_id oid;
990 char *pack_tmp_name = NULL;
992 if (pack_to_stdout)
993 f = hashfd_throughput(1, "<stdout>", progress_state);
994 else
995 f = create_tmp_packfile(&pack_tmp_name);
997 offset = write_pack_header(f, nr_remaining);
999 if (reuse_packfile) {
1000 assert(pack_to_stdout);
1001 write_reused_pack(f);
1002 offset = hashfile_total(f);
1005 nr_written = 0;
1006 for (; i < to_pack.nr_objects; i++) {
1007 struct object_entry *e = write_order[i];
1008 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
1009 break;
1010 display_progress(progress_state, written);
1014 * Did we write the wrong # entries in the header?
1015 * If so, rewrite it like in fast-import
1017 if (pack_to_stdout) {
1018 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
1019 } else if (nr_written == nr_remaining) {
1020 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
1021 } else {
1022 int fd = finalize_hashfile(f, oid.hash, 0);
1023 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
1024 nr_written, oid.hash, offset);
1025 close(fd);
1026 if (write_bitmap_index) {
1027 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1028 warning(_(no_split_warning));
1029 write_bitmap_index = 0;
1033 if (!pack_to_stdout) {
1034 struct stat st;
1035 struct strbuf tmpname = STRBUF_INIT;
1038 * Packs are runtime accessed in their mtime
1039 * order since newer packs are more likely to contain
1040 * younger objects. So if we are creating multiple
1041 * packs then we should modify the mtime of later ones
1042 * to preserve this property.
1044 if (stat(pack_tmp_name, &st) < 0) {
1045 warning_errno(_("failed to stat %s"), pack_tmp_name);
1046 } else if (!last_mtime) {
1047 last_mtime = st.st_mtime;
1048 } else {
1049 struct utimbuf utb;
1050 utb.actime = st.st_atime;
1051 utb.modtime = --last_mtime;
1052 if (utime(pack_tmp_name, &utb) < 0)
1053 warning_errno(_("failed utime() on %s"), pack_tmp_name);
1056 strbuf_addf(&tmpname, "%s-", base_name);
1058 if (write_bitmap_index) {
1059 bitmap_writer_set_checksum(oid.hash);
1060 bitmap_writer_build_type_index(
1061 &to_pack, written_list, nr_written);
1064 finish_tmp_packfile(&tmpname, pack_tmp_name,
1065 written_list, nr_written,
1066 &pack_idx_opts, oid.hash);
1068 if (write_bitmap_index) {
1069 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
1071 stop_progress(&progress_state);
1073 bitmap_writer_show_progress(progress);
1074 bitmap_writer_reuse_bitmaps(&to_pack);
1075 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
1076 bitmap_writer_build(&to_pack);
1077 bitmap_writer_finish(written_list, nr_written,
1078 tmpname.buf, write_bitmap_options);
1079 write_bitmap_index = 0;
1082 strbuf_release(&tmpname);
1083 free(pack_tmp_name);
1084 puts(oid_to_hex(&oid));
1087 /* mark written objects as written to previous pack */
1088 for (j = 0; j < nr_written; j++) {
1089 written_list[j]->offset = (off_t)-1;
1091 nr_remaining -= nr_written;
1092 } while (nr_remaining && i < to_pack.nr_objects);
1094 free(written_list);
1095 free(write_order);
1096 stop_progress(&progress_state);
1097 if (written != nr_result)
1098 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
1099 written, nr_result);
1100 trace2_data_intmax("pack-objects", the_repository,
1101 "write_pack_file/wrote", nr_result);
1104 static int no_try_delta(const char *path)
1106 static struct attr_check *check;
1108 if (!check)
1109 check = attr_check_initl("delta", NULL);
1110 git_check_attr(the_repository->index, path, check);
1111 if (ATTR_FALSE(check->items[0].value))
1112 return 1;
1113 return 0;
1117 * When adding an object, check whether we have already added it
1118 * to our packing list. If so, we can skip. However, if we are
1119 * being asked to excludei t, but the previous mention was to include
1120 * it, make sure to adjust its flags and tweak our numbers accordingly.
1122 * As an optimization, we pass out the index position where we would have
1123 * found the item, since that saves us from having to look it up again a
1124 * few lines later when we want to add the new entry.
1126 static int have_duplicate_entry(const struct object_id *oid,
1127 int exclude)
1129 struct object_entry *entry;
1131 if (reuse_packfile_bitmap &&
1132 bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid))
1133 return 1;
1135 entry = packlist_find(&to_pack, oid);
1136 if (!entry)
1137 return 0;
1139 if (exclude) {
1140 if (!entry->preferred_base)
1141 nr_result--;
1142 entry->preferred_base = 1;
1145 return 1;
1148 static int want_found_object(int exclude, struct packed_git *p)
1150 if (exclude)
1151 return 1;
1152 if (incremental)
1153 return 0;
1156 * When asked to do --local (do not include an object that appears in a
1157 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1158 * an object that appears in a pack marked with .keep), finding a pack
1159 * that matches the criteria is sufficient for us to decide to omit it.
1160 * However, even if this pack does not satisfy the criteria, we need to
1161 * make sure no copy of this object appears in _any_ pack that makes us
1162 * to omit the object, so we need to check all the packs.
1164 * We can however first check whether these options can possible matter;
1165 * if they do not matter we know we want the object in generated pack.
1166 * Otherwise, we signal "-1" at the end to tell the caller that we do
1167 * not know either way, and it needs to check more packs.
1169 if (!ignore_packed_keep_on_disk &&
1170 !ignore_packed_keep_in_core &&
1171 (!local || !have_non_local_packs))
1172 return 1;
1174 if (local && !p->pack_local)
1175 return 0;
1176 if (p->pack_local &&
1177 ((ignore_packed_keep_on_disk && p->pack_keep) ||
1178 (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1179 return 0;
1181 /* we don't know yet; keep looking for more packs */
1182 return -1;
1186 * Check whether we want the object in the pack (e.g., we do not want
1187 * objects found in non-local stores if the "--local" option was used).
1189 * If the caller already knows an existing pack it wants to take the object
1190 * from, that is passed in *found_pack and *found_offset; otherwise this
1191 * function finds if there is any pack that has the object and returns the pack
1192 * and its offset in these variables.
1194 static int want_object_in_pack(const struct object_id *oid,
1195 int exclude,
1196 struct packed_git **found_pack,
1197 off_t *found_offset)
1199 int want;
1200 struct list_head *pos;
1201 struct multi_pack_index *m;
1203 if (!exclude && local && has_loose_object_nonlocal(oid))
1204 return 0;
1207 * If we already know the pack object lives in, start checks from that
1208 * pack - in the usual case when neither --local was given nor .keep files
1209 * are present we will determine the answer right now.
1211 if (*found_pack) {
1212 want = want_found_object(exclude, *found_pack);
1213 if (want != -1)
1214 return want;
1217 for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1218 struct pack_entry e;
1219 if (fill_midx_entry(the_repository, oid, &e, m)) {
1220 struct packed_git *p = e.p;
1221 off_t offset;
1223 if (p == *found_pack)
1224 offset = *found_offset;
1225 else
1226 offset = find_pack_entry_one(oid->hash, p);
1228 if (offset) {
1229 if (!*found_pack) {
1230 if (!is_pack_valid(p))
1231 continue;
1232 *found_offset = offset;
1233 *found_pack = p;
1235 want = want_found_object(exclude, p);
1236 if (want != -1)
1237 return want;
1242 list_for_each(pos, get_packed_git_mru(the_repository)) {
1243 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1244 off_t offset;
1246 if (p == *found_pack)
1247 offset = *found_offset;
1248 else
1249 offset = find_pack_entry_one(oid->hash, p);
1251 if (offset) {
1252 if (!*found_pack) {
1253 if (!is_pack_valid(p))
1254 continue;
1255 *found_offset = offset;
1256 *found_pack = p;
1258 want = want_found_object(exclude, p);
1259 if (!exclude && want > 0)
1260 list_move(&p->mru,
1261 get_packed_git_mru(the_repository));
1262 if (want != -1)
1263 return want;
1267 return 1;
1270 static void create_object_entry(const struct object_id *oid,
1271 enum object_type type,
1272 uint32_t hash,
1273 int exclude,
1274 int no_try_delta,
1275 struct packed_git *found_pack,
1276 off_t found_offset)
1278 struct object_entry *entry;
1280 entry = packlist_alloc(&to_pack, oid);
1281 entry->hash = hash;
1282 oe_set_type(entry, type);
1283 if (exclude)
1284 entry->preferred_base = 1;
1285 else
1286 nr_result++;
1287 if (found_pack) {
1288 oe_set_in_pack(&to_pack, entry, found_pack);
1289 entry->in_pack_offset = found_offset;
1292 entry->no_try_delta = no_try_delta;
1295 static const char no_closure_warning[] = N_(
1296 "disabling bitmap writing, as some objects are not being packed"
1299 static int add_object_entry(const struct object_id *oid, enum object_type type,
1300 const char *name, int exclude)
1302 struct packed_git *found_pack = NULL;
1303 off_t found_offset = 0;
1305 display_progress(progress_state, ++nr_seen);
1307 if (have_duplicate_entry(oid, exclude))
1308 return 0;
1310 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1311 /* The pack is missing an object, so it will not have closure */
1312 if (write_bitmap_index) {
1313 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1314 warning(_(no_closure_warning));
1315 write_bitmap_index = 0;
1317 return 0;
1320 create_object_entry(oid, type, pack_name_hash(name),
1321 exclude, name && no_try_delta(name),
1322 found_pack, found_offset);
1323 return 1;
1326 static int add_object_entry_from_bitmap(const struct object_id *oid,
1327 enum object_type type,
1328 int flags, uint32_t name_hash,
1329 struct packed_git *pack, off_t offset)
1331 display_progress(progress_state, ++nr_seen);
1333 if (have_duplicate_entry(oid, 0))
1334 return 0;
1336 if (!want_object_in_pack(oid, 0, &pack, &offset))
1337 return 0;
1339 create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
1340 return 1;
1343 struct pbase_tree_cache {
1344 struct object_id oid;
1345 int ref;
1346 int temporary;
1347 void *tree_data;
1348 unsigned long tree_size;
1351 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1352 static int pbase_tree_cache_ix(const struct object_id *oid)
1354 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1356 static int pbase_tree_cache_ix_incr(int ix)
1358 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1361 static struct pbase_tree {
1362 struct pbase_tree *next;
1363 /* This is a phony "cache" entry; we are not
1364 * going to evict it or find it through _get()
1365 * mechanism -- this is for the toplevel node that
1366 * would almost always change with any commit.
1368 struct pbase_tree_cache pcache;
1369 } *pbase_tree;
1371 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1373 struct pbase_tree_cache *ent, *nent;
1374 void *data;
1375 unsigned long size;
1376 enum object_type type;
1377 int neigh;
1378 int my_ix = pbase_tree_cache_ix(oid);
1379 int available_ix = -1;
1381 /* pbase-tree-cache acts as a limited hashtable.
1382 * your object will be found at your index or within a few
1383 * slots after that slot if it is cached.
1385 for (neigh = 0; neigh < 8; neigh++) {
1386 ent = pbase_tree_cache[my_ix];
1387 if (ent && oideq(&ent->oid, oid)) {
1388 ent->ref++;
1389 return ent;
1391 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1392 ((0 <= available_ix) &&
1393 (!ent && pbase_tree_cache[available_ix])))
1394 available_ix = my_ix;
1395 if (!ent)
1396 break;
1397 my_ix = pbase_tree_cache_ix_incr(my_ix);
1400 /* Did not find one. Either we got a bogus request or
1401 * we need to read and perhaps cache.
1403 data = read_object_file(oid, &type, &size);
1404 if (!data)
1405 return NULL;
1406 if (type != OBJ_TREE) {
1407 free(data);
1408 return NULL;
1411 /* We need to either cache or return a throwaway copy */
1413 if (available_ix < 0)
1414 ent = NULL;
1415 else {
1416 ent = pbase_tree_cache[available_ix];
1417 my_ix = available_ix;
1420 if (!ent) {
1421 nent = xmalloc(sizeof(*nent));
1422 nent->temporary = (available_ix < 0);
1424 else {
1425 /* evict and reuse */
1426 free(ent->tree_data);
1427 nent = ent;
1429 oidcpy(&nent->oid, oid);
1430 nent->tree_data = data;
1431 nent->tree_size = size;
1432 nent->ref = 1;
1433 if (!nent->temporary)
1434 pbase_tree_cache[my_ix] = nent;
1435 return nent;
1438 static void pbase_tree_put(struct pbase_tree_cache *cache)
1440 if (!cache->temporary) {
1441 cache->ref--;
1442 return;
1444 free(cache->tree_data);
1445 free(cache);
1448 static int name_cmp_len(const char *name)
1450 int i;
1451 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1453 return i;
1456 static void add_pbase_object(struct tree_desc *tree,
1457 const char *name,
1458 int cmplen,
1459 const char *fullname)
1461 struct name_entry entry;
1462 int cmp;
1464 while (tree_entry(tree,&entry)) {
1465 if (S_ISGITLINK(entry.mode))
1466 continue;
1467 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1468 memcmp(name, entry.path, cmplen);
1469 if (cmp > 0)
1470 continue;
1471 if (cmp < 0)
1472 return;
1473 if (name[cmplen] != '/') {
1474 add_object_entry(&entry.oid,
1475 object_type(entry.mode),
1476 fullname, 1);
1477 return;
1479 if (S_ISDIR(entry.mode)) {
1480 struct tree_desc sub;
1481 struct pbase_tree_cache *tree;
1482 const char *down = name+cmplen+1;
1483 int downlen = name_cmp_len(down);
1485 tree = pbase_tree_get(&entry.oid);
1486 if (!tree)
1487 return;
1488 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1490 add_pbase_object(&sub, down, downlen, fullname);
1491 pbase_tree_put(tree);
1496 static unsigned *done_pbase_paths;
1497 static int done_pbase_paths_num;
1498 static int done_pbase_paths_alloc;
1499 static int done_pbase_path_pos(unsigned hash)
1501 int lo = 0;
1502 int hi = done_pbase_paths_num;
1503 while (lo < hi) {
1504 int mi = lo + (hi - lo) / 2;
1505 if (done_pbase_paths[mi] == hash)
1506 return mi;
1507 if (done_pbase_paths[mi] < hash)
1508 hi = mi;
1509 else
1510 lo = mi + 1;
1512 return -lo-1;
1515 static int check_pbase_path(unsigned hash)
1517 int pos = done_pbase_path_pos(hash);
1518 if (0 <= pos)
1519 return 1;
1520 pos = -pos - 1;
1521 ALLOC_GROW(done_pbase_paths,
1522 done_pbase_paths_num + 1,
1523 done_pbase_paths_alloc);
1524 done_pbase_paths_num++;
1525 if (pos < done_pbase_paths_num)
1526 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1527 done_pbase_paths_num - pos - 1);
1528 done_pbase_paths[pos] = hash;
1529 return 0;
1532 static void add_preferred_base_object(const char *name)
1534 struct pbase_tree *it;
1535 int cmplen;
1536 unsigned hash = pack_name_hash(name);
1538 if (!num_preferred_base || check_pbase_path(hash))
1539 return;
1541 cmplen = name_cmp_len(name);
1542 for (it = pbase_tree; it; it = it->next) {
1543 if (cmplen == 0) {
1544 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1546 else {
1547 struct tree_desc tree;
1548 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1549 add_pbase_object(&tree, name, cmplen, name);
1554 static void add_preferred_base(struct object_id *oid)
1556 struct pbase_tree *it;
1557 void *data;
1558 unsigned long size;
1559 struct object_id tree_oid;
1561 if (window <= num_preferred_base++)
1562 return;
1564 data = read_object_with_reference(the_repository, oid,
1565 tree_type, &size, &tree_oid);
1566 if (!data)
1567 return;
1569 for (it = pbase_tree; it; it = it->next) {
1570 if (oideq(&it->pcache.oid, &tree_oid)) {
1571 free(data);
1572 return;
1576 it = xcalloc(1, sizeof(*it));
1577 it->next = pbase_tree;
1578 pbase_tree = it;
1580 oidcpy(&it->pcache.oid, &tree_oid);
1581 it->pcache.tree_data = data;
1582 it->pcache.tree_size = size;
1585 static void cleanup_preferred_base(void)
1587 struct pbase_tree *it;
1588 unsigned i;
1590 it = pbase_tree;
1591 pbase_tree = NULL;
1592 while (it) {
1593 struct pbase_tree *tmp = it;
1594 it = tmp->next;
1595 free(tmp->pcache.tree_data);
1596 free(tmp);
1599 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1600 if (!pbase_tree_cache[i])
1601 continue;
1602 free(pbase_tree_cache[i]->tree_data);
1603 FREE_AND_NULL(pbase_tree_cache[i]);
1606 FREE_AND_NULL(done_pbase_paths);
1607 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1611 * Return 1 iff the object specified by "delta" can be sent
1612 * literally as a delta against the base in "base_sha1". If
1613 * so, then *base_out will point to the entry in our packing
1614 * list, or NULL if we must use the external-base list.
1616 * Depth value does not matter - find_deltas() will
1617 * never consider reused delta as the base object to
1618 * deltify other objects against, in order to avoid
1619 * circular deltas.
1621 static int can_reuse_delta(const unsigned char *base_sha1,
1622 struct object_entry *delta,
1623 struct object_entry **base_out)
1625 struct object_entry *base;
1626 struct object_id base_oid;
1628 if (!base_sha1)
1629 return 0;
1631 oidread(&base_oid, base_sha1);
1634 * First see if we're already sending the base (or it's explicitly in
1635 * our "excluded" list).
1637 base = packlist_find(&to_pack, &base_oid);
1638 if (base) {
1639 if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1640 return 0;
1641 *base_out = base;
1642 return 1;
1646 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1647 * even if it was buried too deep in history to make it into the
1648 * packing list.
1650 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {
1651 if (use_delta_islands) {
1652 if (!in_same_island(&delta->idx.oid, &base_oid))
1653 return 0;
1655 *base_out = NULL;
1656 return 1;
1659 return 0;
1662 static void check_object(struct object_entry *entry)
1664 unsigned long canonical_size;
1666 if (IN_PACK(entry)) {
1667 struct packed_git *p = IN_PACK(entry);
1668 struct pack_window *w_curs = NULL;
1669 const unsigned char *base_ref = NULL;
1670 struct object_entry *base_entry;
1671 unsigned long used, used_0;
1672 unsigned long avail;
1673 off_t ofs;
1674 unsigned char *buf, c;
1675 enum object_type type;
1676 unsigned long in_pack_size;
1678 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1681 * We want in_pack_type even if we do not reuse delta
1682 * since non-delta representations could still be reused.
1684 used = unpack_object_header_buffer(buf, avail,
1685 &type,
1686 &in_pack_size);
1687 if (used == 0)
1688 goto give_up;
1690 if (type < 0)
1691 BUG("invalid type %d", type);
1692 entry->in_pack_type = type;
1695 * Determine if this is a delta and if so whether we can
1696 * reuse it or not. Otherwise let's find out as cheaply as
1697 * possible what the actual type and size for this object is.
1699 switch (entry->in_pack_type) {
1700 default:
1701 /* Not a delta hence we've already got all we need. */
1702 oe_set_type(entry, entry->in_pack_type);
1703 SET_SIZE(entry, in_pack_size);
1704 entry->in_pack_header_size = used;
1705 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1706 goto give_up;
1707 unuse_pack(&w_curs);
1708 return;
1709 case OBJ_REF_DELTA:
1710 if (reuse_delta && !entry->preferred_base)
1711 base_ref = use_pack(p, &w_curs,
1712 entry->in_pack_offset + used, NULL);
1713 entry->in_pack_header_size = used + the_hash_algo->rawsz;
1714 break;
1715 case OBJ_OFS_DELTA:
1716 buf = use_pack(p, &w_curs,
1717 entry->in_pack_offset + used, NULL);
1718 used_0 = 0;
1719 c = buf[used_0++];
1720 ofs = c & 127;
1721 while (c & 128) {
1722 ofs += 1;
1723 if (!ofs || MSB(ofs, 7)) {
1724 error(_("delta base offset overflow in pack for %s"),
1725 oid_to_hex(&entry->idx.oid));
1726 goto give_up;
1728 c = buf[used_0++];
1729 ofs = (ofs << 7) + (c & 127);
1731 ofs = entry->in_pack_offset - ofs;
1732 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1733 error(_("delta base offset out of bound for %s"),
1734 oid_to_hex(&entry->idx.oid));
1735 goto give_up;
1737 if (reuse_delta && !entry->preferred_base) {
1738 struct revindex_entry *revidx;
1739 revidx = find_pack_revindex(p, ofs);
1740 if (!revidx)
1741 goto give_up;
1742 base_ref = nth_packed_object_sha1(p, revidx->nr);
1744 entry->in_pack_header_size = used + used_0;
1745 break;
1748 if (can_reuse_delta(base_ref, entry, &base_entry)) {
1749 oe_set_type(entry, entry->in_pack_type);
1750 SET_SIZE(entry, in_pack_size); /* delta size */
1751 SET_DELTA_SIZE(entry, in_pack_size);
1753 if (base_entry) {
1754 SET_DELTA(entry, base_entry);
1755 entry->delta_sibling_idx = base_entry->delta_child_idx;
1756 SET_DELTA_CHILD(base_entry, entry);
1757 } else {
1758 SET_DELTA_EXT(entry, base_ref);
1761 unuse_pack(&w_curs);
1762 return;
1765 if (oe_type(entry)) {
1766 off_t delta_pos;
1769 * This must be a delta and we already know what the
1770 * final object type is. Let's extract the actual
1771 * object size from the delta header.
1773 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1774 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1775 if (canonical_size == 0)
1776 goto give_up;
1777 SET_SIZE(entry, canonical_size);
1778 unuse_pack(&w_curs);
1779 return;
1783 * No choice but to fall back to the recursive delta walk
1784 * with oid_object_info() to find about the object type
1785 * at this point...
1787 give_up:
1788 unuse_pack(&w_curs);
1791 oe_set_type(entry,
1792 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1793 if (entry->type_valid) {
1794 SET_SIZE(entry, canonical_size);
1795 } else {
1797 * Bad object type is checked in prepare_pack(). This is
1798 * to permit a missing preferred base object to be ignored
1799 * as a preferred base. Doing so can result in a larger
1800 * pack file, but the transfer will still take place.
1805 static int pack_offset_sort(const void *_a, const void *_b)
1807 const struct object_entry *a = *(struct object_entry **)_a;
1808 const struct object_entry *b = *(struct object_entry **)_b;
1809 const struct packed_git *a_in_pack = IN_PACK(a);
1810 const struct packed_git *b_in_pack = IN_PACK(b);
1812 /* avoid filesystem trashing with loose objects */
1813 if (!a_in_pack && !b_in_pack)
1814 return oidcmp(&a->idx.oid, &b->idx.oid);
1816 if (a_in_pack < b_in_pack)
1817 return -1;
1818 if (a_in_pack > b_in_pack)
1819 return 1;
1820 return a->in_pack_offset < b->in_pack_offset ? -1 :
1821 (a->in_pack_offset > b->in_pack_offset);
1825 * Drop an on-disk delta we were planning to reuse. Naively, this would
1826 * just involve blanking out the "delta" field, but we have to deal
1827 * with some extra book-keeping:
1829 * 1. Removing ourselves from the delta_sibling linked list.
1831 * 2. Updating our size/type to the non-delta representation. These were
1832 * either not recorded initially (size) or overwritten with the delta type
1833 * (type) when check_object() decided to reuse the delta.
1835 * 3. Resetting our delta depth, as we are now a base object.
1837 static void drop_reused_delta(struct object_entry *entry)
1839 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1840 struct object_info oi = OBJECT_INFO_INIT;
1841 enum object_type type;
1842 unsigned long size;
1844 while (*idx) {
1845 struct object_entry *oe = &to_pack.objects[*idx - 1];
1847 if (oe == entry)
1848 *idx = oe->delta_sibling_idx;
1849 else
1850 idx = &oe->delta_sibling_idx;
1852 SET_DELTA(entry, NULL);
1853 entry->depth = 0;
1855 oi.sizep = &size;
1856 oi.typep = &type;
1857 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1859 * We failed to get the info from this pack for some reason;
1860 * fall back to oid_object_info, which may find another copy.
1861 * And if that fails, the error will be recorded in oe_type(entry)
1862 * and dealt with in prepare_pack().
1864 oe_set_type(entry,
1865 oid_object_info(the_repository, &entry->idx.oid, &size));
1866 } else {
1867 oe_set_type(entry, type);
1869 SET_SIZE(entry, size);
1873 * Follow the chain of deltas from this entry onward, throwing away any links
1874 * that cause us to hit a cycle (as determined by the DFS state flags in
1875 * the entries).
1877 * We also detect too-long reused chains that would violate our --depth
1878 * limit.
1880 static void break_delta_chains(struct object_entry *entry)
1883 * The actual depth of each object we will write is stored as an int,
1884 * as it cannot exceed our int "depth" limit. But before we break
1885 * changes based no that limit, we may potentially go as deep as the
1886 * number of objects, which is elsewhere bounded to a uint32_t.
1888 uint32_t total_depth;
1889 struct object_entry *cur, *next;
1891 for (cur = entry, total_depth = 0;
1892 cur;
1893 cur = DELTA(cur), total_depth++) {
1894 if (cur->dfs_state == DFS_DONE) {
1896 * We've already seen this object and know it isn't
1897 * part of a cycle. We do need to append its depth
1898 * to our count.
1900 total_depth += cur->depth;
1901 break;
1905 * We break cycles before looping, so an ACTIVE state (or any
1906 * other cruft which made its way into the state variable)
1907 * is a bug.
1909 if (cur->dfs_state != DFS_NONE)
1910 BUG("confusing delta dfs state in first pass: %d",
1911 cur->dfs_state);
1914 * Now we know this is the first time we've seen the object. If
1915 * it's not a delta, we're done traversing, but we'll mark it
1916 * done to save time on future traversals.
1918 if (!DELTA(cur)) {
1919 cur->dfs_state = DFS_DONE;
1920 break;
1924 * Mark ourselves as active and see if the next step causes
1925 * us to cycle to another active object. It's important to do
1926 * this _before_ we loop, because it impacts where we make the
1927 * cut, and thus how our total_depth counter works.
1928 * E.g., We may see a partial loop like:
1930 * A -> B -> C -> D -> B
1932 * Cutting B->C breaks the cycle. But now the depth of A is
1933 * only 1, and our total_depth counter is at 3. The size of the
1934 * error is always one less than the size of the cycle we
1935 * broke. Commits C and D were "lost" from A's chain.
1937 * If we instead cut D->B, then the depth of A is correct at 3.
1938 * We keep all commits in the chain that we examined.
1940 cur->dfs_state = DFS_ACTIVE;
1941 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1942 drop_reused_delta(cur);
1943 cur->dfs_state = DFS_DONE;
1944 break;
1949 * And now that we've gone all the way to the bottom of the chain, we
1950 * need to clear the active flags and set the depth fields as
1951 * appropriate. Unlike the loop above, which can quit when it drops a
1952 * delta, we need to keep going to look for more depth cuts. So we need
1953 * an extra "next" pointer to keep going after we reset cur->delta.
1955 for (cur = entry; cur; cur = next) {
1956 next = DELTA(cur);
1959 * We should have a chain of zero or more ACTIVE states down to
1960 * a final DONE. We can quit after the DONE, because either it
1961 * has no bases, or we've already handled them in a previous
1962 * call.
1964 if (cur->dfs_state == DFS_DONE)
1965 break;
1966 else if (cur->dfs_state != DFS_ACTIVE)
1967 BUG("confusing delta dfs state in second pass: %d",
1968 cur->dfs_state);
1971 * If the total_depth is more than depth, then we need to snip
1972 * the chain into two or more smaller chains that don't exceed
1973 * the maximum depth. Most of the resulting chains will contain
1974 * (depth + 1) entries (i.e., depth deltas plus one base), and
1975 * the last chain (i.e., the one containing entry) will contain
1976 * whatever entries are left over, namely
1977 * (total_depth % (depth + 1)) of them.
1979 * Since we are iterating towards decreasing depth, we need to
1980 * decrement total_depth as we go, and we need to write to the
1981 * entry what its final depth will be after all of the
1982 * snipping. Since we're snipping into chains of length (depth
1983 * + 1) entries, the final depth of an entry will be its
1984 * original depth modulo (depth + 1). Any time we encounter an
1985 * entry whose final depth is supposed to be zero, we snip it
1986 * from its delta base, thereby making it so.
1988 cur->depth = (total_depth--) % (depth + 1);
1989 if (!cur->depth)
1990 drop_reused_delta(cur);
1992 cur->dfs_state = DFS_DONE;
1996 static void get_object_details(void)
1998 uint32_t i;
1999 struct object_entry **sorted_by_offset;
2001 if (progress)
2002 progress_state = start_progress(_("Counting objects"),
2003 to_pack.nr_objects);
2005 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
2006 for (i = 0; i < to_pack.nr_objects; i++)
2007 sorted_by_offset[i] = to_pack.objects + i;
2008 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
2010 for (i = 0; i < to_pack.nr_objects; i++) {
2011 struct object_entry *entry = sorted_by_offset[i];
2012 check_object(entry);
2013 if (entry->type_valid &&
2014 oe_size_greater_than(&to_pack, entry, big_file_threshold))
2015 entry->no_try_delta = 1;
2016 display_progress(progress_state, i + 1);
2018 stop_progress(&progress_state);
2021 * This must happen in a second pass, since we rely on the delta
2022 * information for the whole list being completed.
2024 for (i = 0; i < to_pack.nr_objects; i++)
2025 break_delta_chains(&to_pack.objects[i]);
2027 free(sorted_by_offset);
2031 * We search for deltas in a list sorted by type, by filename hash, and then
2032 * by size, so that we see progressively smaller and smaller files.
2033 * That's because we prefer deltas to be from the bigger file
2034 * to the smaller -- deletes are potentially cheaper, but perhaps
2035 * more importantly, the bigger file is likely the more recent
2036 * one. The deepest deltas are therefore the oldest objects which are
2037 * less susceptible to be accessed often.
2039 static int type_size_sort(const void *_a, const void *_b)
2041 const struct object_entry *a = *(struct object_entry **)_a;
2042 const struct object_entry *b = *(struct object_entry **)_b;
2043 const enum object_type a_type = oe_type(a);
2044 const enum object_type b_type = oe_type(b);
2045 const unsigned long a_size = SIZE(a);
2046 const unsigned long b_size = SIZE(b);
2048 if (a_type > b_type)
2049 return -1;
2050 if (a_type < b_type)
2051 return 1;
2052 if (a->hash > b->hash)
2053 return -1;
2054 if (a->hash < b->hash)
2055 return 1;
2056 if (a->preferred_base > b->preferred_base)
2057 return -1;
2058 if (a->preferred_base < b->preferred_base)
2059 return 1;
2060 if (use_delta_islands) {
2061 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
2062 if (island_cmp)
2063 return island_cmp;
2065 if (a_size > b_size)
2066 return -1;
2067 if (a_size < b_size)
2068 return 1;
2069 return a < b ? -1 : (a > b); /* newest first */
2072 struct unpacked {
2073 struct object_entry *entry;
2074 void *data;
2075 struct delta_index *index;
2076 unsigned depth;
2079 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
2080 unsigned long delta_size)
2082 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
2083 return 0;
2085 if (delta_size < cache_max_small_delta_size)
2086 return 1;
2088 /* cache delta, if objects are large enough compared to delta size */
2089 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
2090 return 1;
2092 return 0;
2095 /* Protect delta_cache_size */
2096 static pthread_mutex_t cache_mutex;
2097 #define cache_lock() pthread_mutex_lock(&cache_mutex)
2098 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
2101 * Protect object list partitioning (e.g. struct thread_param) and
2102 * progress_state
2104 static pthread_mutex_t progress_mutex;
2105 #define progress_lock() pthread_mutex_lock(&progress_mutex)
2106 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
2109 * Access to struct object_entry is unprotected since each thread owns
2110 * a portion of the main object list. Just don't access object entries
2111 * ahead in the list because they can be stolen and would need
2112 * progress_mutex for protection.
2116 * Return the size of the object without doing any delta
2117 * reconstruction (so non-deltas are true object sizes, but deltas
2118 * return the size of the delta data).
2120 unsigned long oe_get_size_slow(struct packing_data *pack,
2121 const struct object_entry *e)
2123 struct packed_git *p;
2124 struct pack_window *w_curs;
2125 unsigned char *buf;
2126 enum object_type type;
2127 unsigned long used, avail, size;
2129 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
2130 packing_data_lock(&to_pack);
2131 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
2132 die(_("unable to get size of %s"),
2133 oid_to_hex(&e->idx.oid));
2134 packing_data_unlock(&to_pack);
2135 return size;
2138 p = oe_in_pack(pack, e);
2139 if (!p)
2140 BUG("when e->type is a delta, it must belong to a pack");
2142 packing_data_lock(&to_pack);
2143 w_curs = NULL;
2144 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2145 used = unpack_object_header_buffer(buf, avail, &type, &size);
2146 if (used == 0)
2147 die(_("unable to parse object header of %s"),
2148 oid_to_hex(&e->idx.oid));
2150 unuse_pack(&w_curs);
2151 packing_data_unlock(&to_pack);
2152 return size;
2155 static int try_delta(struct unpacked *trg, struct unpacked *src,
2156 unsigned max_depth, unsigned long *mem_usage)
2158 struct object_entry *trg_entry = trg->entry;
2159 struct object_entry *src_entry = src->entry;
2160 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
2161 unsigned ref_depth;
2162 enum object_type type;
2163 void *delta_buf;
2165 /* Don't bother doing diffs between different types */
2166 if (oe_type(trg_entry) != oe_type(src_entry))
2167 return -1;
2170 * We do not bother to try a delta that we discarded on an
2171 * earlier try, but only when reusing delta data. Note that
2172 * src_entry that is marked as the preferred_base should always
2173 * be considered, as even if we produce a suboptimal delta against
2174 * it, we will still save the transfer cost, as we already know
2175 * the other side has it and we won't send src_entry at all.
2177 if (reuse_delta && IN_PACK(trg_entry) &&
2178 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2179 !src_entry->preferred_base &&
2180 trg_entry->in_pack_type != OBJ_REF_DELTA &&
2181 trg_entry->in_pack_type != OBJ_OFS_DELTA)
2182 return 0;
2184 /* Let's not bust the allowed depth. */
2185 if (src->depth >= max_depth)
2186 return 0;
2188 /* Now some size filtering heuristics. */
2189 trg_size = SIZE(trg_entry);
2190 if (!DELTA(trg_entry)) {
2191 max_size = trg_size/2 - the_hash_algo->rawsz;
2192 ref_depth = 1;
2193 } else {
2194 max_size = DELTA_SIZE(trg_entry);
2195 ref_depth = trg->depth;
2197 max_size = (uint64_t)max_size * (max_depth - src->depth) /
2198 (max_depth - ref_depth + 1);
2199 if (max_size == 0)
2200 return 0;
2201 src_size = SIZE(src_entry);
2202 sizediff = src_size < trg_size ? trg_size - src_size : 0;
2203 if (sizediff >= max_size)
2204 return 0;
2205 if (trg_size < src_size / 32)
2206 return 0;
2208 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2209 return 0;
2211 /* Load data if not already done */
2212 if (!trg->data) {
2213 packing_data_lock(&to_pack);
2214 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
2215 packing_data_unlock(&to_pack);
2216 if (!trg->data)
2217 die(_("object %s cannot be read"),
2218 oid_to_hex(&trg_entry->idx.oid));
2219 if (sz != trg_size)
2220 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2221 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2222 (uintmax_t)trg_size);
2223 *mem_usage += sz;
2225 if (!src->data) {
2226 packing_data_lock(&to_pack);
2227 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
2228 packing_data_unlock(&to_pack);
2229 if (!src->data) {
2230 if (src_entry->preferred_base) {
2231 static int warned = 0;
2232 if (!warned++)
2233 warning(_("object %s cannot be read"),
2234 oid_to_hex(&src_entry->idx.oid));
2236 * Those objects are not included in the
2237 * resulting pack. Be resilient and ignore
2238 * them if they can't be read, in case the
2239 * pack could be created nevertheless.
2241 return 0;
2243 die(_("object %s cannot be read"),
2244 oid_to_hex(&src_entry->idx.oid));
2246 if (sz != src_size)
2247 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2248 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2249 (uintmax_t)src_size);
2250 *mem_usage += sz;
2252 if (!src->index) {
2253 src->index = create_delta_index(src->data, src_size);
2254 if (!src->index) {
2255 static int warned = 0;
2256 if (!warned++)
2257 warning(_("suboptimal pack - out of memory"));
2258 return 0;
2260 *mem_usage += sizeof_delta_index(src->index);
2263 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2264 if (!delta_buf)
2265 return 0;
2267 if (DELTA(trg_entry)) {
2268 /* Prefer only shallower same-sized deltas. */
2269 if (delta_size == DELTA_SIZE(trg_entry) &&
2270 src->depth + 1 >= trg->depth) {
2271 free(delta_buf);
2272 return 0;
2277 * Handle memory allocation outside of the cache
2278 * accounting lock. Compiler will optimize the strangeness
2279 * away when NO_PTHREADS is defined.
2281 free(trg_entry->delta_data);
2282 cache_lock();
2283 if (trg_entry->delta_data) {
2284 delta_cache_size -= DELTA_SIZE(trg_entry);
2285 trg_entry->delta_data = NULL;
2287 if (delta_cacheable(src_size, trg_size, delta_size)) {
2288 delta_cache_size += delta_size;
2289 cache_unlock();
2290 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2291 } else {
2292 cache_unlock();
2293 free(delta_buf);
2296 SET_DELTA(trg_entry, src_entry);
2297 SET_DELTA_SIZE(trg_entry, delta_size);
2298 trg->depth = src->depth + 1;
2300 return 1;
2303 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2305 struct object_entry *child = DELTA_CHILD(me);
2306 unsigned int m = n;
2307 while (child) {
2308 const unsigned int c = check_delta_limit(child, n + 1);
2309 if (m < c)
2310 m = c;
2311 child = DELTA_SIBLING(child);
2313 return m;
2316 static unsigned long free_unpacked(struct unpacked *n)
2318 unsigned long freed_mem = sizeof_delta_index(n->index);
2319 free_delta_index(n->index);
2320 n->index = NULL;
2321 if (n->data) {
2322 freed_mem += SIZE(n->entry);
2323 FREE_AND_NULL(n->data);
2325 n->entry = NULL;
2326 n->depth = 0;
2327 return freed_mem;
2330 static void find_deltas(struct object_entry **list, unsigned *list_size,
2331 int window, int depth, unsigned *processed)
2333 uint32_t i, idx = 0, count = 0;
2334 struct unpacked *array;
2335 unsigned long mem_usage = 0;
2337 array = xcalloc(window, sizeof(struct unpacked));
2339 for (;;) {
2340 struct object_entry *entry;
2341 struct unpacked *n = array + idx;
2342 int j, max_depth, best_base = -1;
2344 progress_lock();
2345 if (!*list_size) {
2346 progress_unlock();
2347 break;
2349 entry = *list++;
2350 (*list_size)--;
2351 if (!entry->preferred_base) {
2352 (*processed)++;
2353 display_progress(progress_state, *processed);
2355 progress_unlock();
2357 mem_usage -= free_unpacked(n);
2358 n->entry = entry;
2360 while (window_memory_limit &&
2361 mem_usage > window_memory_limit &&
2362 count > 1) {
2363 const uint32_t tail = (idx + window - count) % window;
2364 mem_usage -= free_unpacked(array + tail);
2365 count--;
2368 /* We do not compute delta to *create* objects we are not
2369 * going to pack.
2371 if (entry->preferred_base)
2372 goto next;
2375 * If the current object is at pack edge, take the depth the
2376 * objects that depend on the current object into account
2377 * otherwise they would become too deep.
2379 max_depth = depth;
2380 if (DELTA_CHILD(entry)) {
2381 max_depth -= check_delta_limit(entry, 0);
2382 if (max_depth <= 0)
2383 goto next;
2386 j = window;
2387 while (--j > 0) {
2388 int ret;
2389 uint32_t other_idx = idx + j;
2390 struct unpacked *m;
2391 if (other_idx >= window)
2392 other_idx -= window;
2393 m = array + other_idx;
2394 if (!m->entry)
2395 break;
2396 ret = try_delta(n, m, max_depth, &mem_usage);
2397 if (ret < 0)
2398 break;
2399 else if (ret > 0)
2400 best_base = other_idx;
2404 * If we decided to cache the delta data, then it is best
2405 * to compress it right away. First because we have to do
2406 * it anyway, and doing it here while we're threaded will
2407 * save a lot of time in the non threaded write phase,
2408 * as well as allow for caching more deltas within
2409 * the same cache size limit.
2410 * ...
2411 * But only if not writing to stdout, since in that case
2412 * the network is most likely throttling writes anyway,
2413 * and therefore it is best to go to the write phase ASAP
2414 * instead, as we can afford spending more time compressing
2415 * between writes at that moment.
2417 if (entry->delta_data && !pack_to_stdout) {
2418 unsigned long size;
2420 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2421 if (size < (1U << OE_Z_DELTA_BITS)) {
2422 entry->z_delta_size = size;
2423 cache_lock();
2424 delta_cache_size -= DELTA_SIZE(entry);
2425 delta_cache_size += entry->z_delta_size;
2426 cache_unlock();
2427 } else {
2428 FREE_AND_NULL(entry->delta_data);
2429 entry->z_delta_size = 0;
2433 /* if we made n a delta, and if n is already at max
2434 * depth, leaving it in the window is pointless. we
2435 * should evict it first.
2437 if (DELTA(entry) && max_depth <= n->depth)
2438 continue;
2441 * Move the best delta base up in the window, after the
2442 * currently deltified object, to keep it longer. It will
2443 * be the first base object to be attempted next.
2445 if (DELTA(entry)) {
2446 struct unpacked swap = array[best_base];
2447 int dist = (window + idx - best_base) % window;
2448 int dst = best_base;
2449 while (dist--) {
2450 int src = (dst + 1) % window;
2451 array[dst] = array[src];
2452 dst = src;
2454 array[dst] = swap;
2457 next:
2458 idx++;
2459 if (count + 1 < window)
2460 count++;
2461 if (idx >= window)
2462 idx = 0;
2465 for (i = 0; i < window; ++i) {
2466 free_delta_index(array[i].index);
2467 free(array[i].data);
2469 free(array);
2473 * The main object list is split into smaller lists, each is handed to
2474 * one worker.
2476 * The main thread waits on the condition that (at least) one of the workers
2477 * has stopped working (which is indicated in the .working member of
2478 * struct thread_params).
2480 * When a work thread has completed its work, it sets .working to 0 and
2481 * signals the main thread and waits on the condition that .data_ready
2482 * becomes 1.
2484 * The main thread steals half of the work from the worker that has
2485 * most work left to hand it to the idle worker.
2488 struct thread_params {
2489 pthread_t thread;
2490 struct object_entry **list;
2491 unsigned list_size;
2492 unsigned remaining;
2493 int window;
2494 int depth;
2495 int working;
2496 int data_ready;
2497 pthread_mutex_t mutex;
2498 pthread_cond_t cond;
2499 unsigned *processed;
2502 static pthread_cond_t progress_cond;
2505 * Mutex and conditional variable can't be statically-initialized on Windows.
2507 static void init_threaded_search(void)
2509 pthread_mutex_init(&cache_mutex, NULL);
2510 pthread_mutex_init(&progress_mutex, NULL);
2511 pthread_cond_init(&progress_cond, NULL);
2514 static void cleanup_threaded_search(void)
2516 pthread_cond_destroy(&progress_cond);
2517 pthread_mutex_destroy(&cache_mutex);
2518 pthread_mutex_destroy(&progress_mutex);
2521 static void *threaded_find_deltas(void *arg)
2523 struct thread_params *me = arg;
2525 progress_lock();
2526 while (me->remaining) {
2527 progress_unlock();
2529 find_deltas(me->list, &me->remaining,
2530 me->window, me->depth, me->processed);
2532 progress_lock();
2533 me->working = 0;
2534 pthread_cond_signal(&progress_cond);
2535 progress_unlock();
2538 * We must not set ->data_ready before we wait on the
2539 * condition because the main thread may have set it to 1
2540 * before we get here. In order to be sure that new
2541 * work is available if we see 1 in ->data_ready, it
2542 * was initialized to 0 before this thread was spawned
2543 * and we reset it to 0 right away.
2545 pthread_mutex_lock(&me->mutex);
2546 while (!me->data_ready)
2547 pthread_cond_wait(&me->cond, &me->mutex);
2548 me->data_ready = 0;
2549 pthread_mutex_unlock(&me->mutex);
2551 progress_lock();
2553 progress_unlock();
2554 /* leave ->working 1 so that this doesn't get more work assigned */
2555 return NULL;
2558 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2559 int window, int depth, unsigned *processed)
2561 struct thread_params *p;
2562 int i, ret, active_threads = 0;
2564 init_threaded_search();
2566 if (delta_search_threads <= 1) {
2567 find_deltas(list, &list_size, window, depth, processed);
2568 cleanup_threaded_search();
2569 return;
2571 if (progress > pack_to_stdout)
2572 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
2573 delta_search_threads);
2574 p = xcalloc(delta_search_threads, sizeof(*p));
2576 /* Partition the work amongst work threads. */
2577 for (i = 0; i < delta_search_threads; i++) {
2578 unsigned sub_size = list_size / (delta_search_threads - i);
2580 /* don't use too small segments or no deltas will be found */
2581 if (sub_size < 2*window && i+1 < delta_search_threads)
2582 sub_size = 0;
2584 p[i].window = window;
2585 p[i].depth = depth;
2586 p[i].processed = processed;
2587 p[i].working = 1;
2588 p[i].data_ready = 0;
2590 /* try to split chunks on "path" boundaries */
2591 while (sub_size && sub_size < list_size &&
2592 list[sub_size]->hash &&
2593 list[sub_size]->hash == list[sub_size-1]->hash)
2594 sub_size++;
2596 p[i].list = list;
2597 p[i].list_size = sub_size;
2598 p[i].remaining = sub_size;
2600 list += sub_size;
2601 list_size -= sub_size;
2604 /* Start work threads. */
2605 for (i = 0; i < delta_search_threads; i++) {
2606 if (!p[i].list_size)
2607 continue;
2608 pthread_mutex_init(&p[i].mutex, NULL);
2609 pthread_cond_init(&p[i].cond, NULL);
2610 ret = pthread_create(&p[i].thread, NULL,
2611 threaded_find_deltas, &p[i]);
2612 if (ret)
2613 die(_("unable to create thread: %s"), strerror(ret));
2614 active_threads++;
2618 * Now let's wait for work completion. Each time a thread is done
2619 * with its work, we steal half of the remaining work from the
2620 * thread with the largest number of unprocessed objects and give
2621 * it to that newly idle thread. This ensure good load balancing
2622 * until the remaining object list segments are simply too short
2623 * to be worth splitting anymore.
2625 while (active_threads) {
2626 struct thread_params *target = NULL;
2627 struct thread_params *victim = NULL;
2628 unsigned sub_size = 0;
2630 progress_lock();
2631 for (;;) {
2632 for (i = 0; !target && i < delta_search_threads; i++)
2633 if (!p[i].working)
2634 target = &p[i];
2635 if (target)
2636 break;
2637 pthread_cond_wait(&progress_cond, &progress_mutex);
2640 for (i = 0; i < delta_search_threads; i++)
2641 if (p[i].remaining > 2*window &&
2642 (!victim || victim->remaining < p[i].remaining))
2643 victim = &p[i];
2644 if (victim) {
2645 sub_size = victim->remaining / 2;
2646 list = victim->list + victim->list_size - sub_size;
2647 while (sub_size && list[0]->hash &&
2648 list[0]->hash == list[-1]->hash) {
2649 list++;
2650 sub_size--;
2652 if (!sub_size) {
2654 * It is possible for some "paths" to have
2655 * so many objects that no hash boundary
2656 * might be found. Let's just steal the
2657 * exact half in that case.
2659 sub_size = victim->remaining / 2;
2660 list -= sub_size;
2662 target->list = list;
2663 victim->list_size -= sub_size;
2664 victim->remaining -= sub_size;
2666 target->list_size = sub_size;
2667 target->remaining = sub_size;
2668 target->working = 1;
2669 progress_unlock();
2671 pthread_mutex_lock(&target->mutex);
2672 target->data_ready = 1;
2673 pthread_cond_signal(&target->cond);
2674 pthread_mutex_unlock(&target->mutex);
2676 if (!sub_size) {
2677 pthread_join(target->thread, NULL);
2678 pthread_cond_destroy(&target->cond);
2679 pthread_mutex_destroy(&target->mutex);
2680 active_threads--;
2683 cleanup_threaded_search();
2684 free(p);
2687 static int obj_is_packed(const struct object_id *oid)
2689 return packlist_find(&to_pack, oid) ||
2690 (reuse_packfile_bitmap &&
2691 bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid));
2694 static void add_tag_chain(const struct object_id *oid)
2696 struct tag *tag;
2699 * We catch duplicates already in add_object_entry(), but we'd
2700 * prefer to do this extra check to avoid having to parse the
2701 * tag at all if we already know that it's being packed (e.g., if
2702 * it was included via bitmaps, we would not have parsed it
2703 * previously).
2705 if (obj_is_packed(oid))
2706 return;
2708 tag = lookup_tag(the_repository, oid);
2709 while (1) {
2710 if (!tag || parse_tag(tag) || !tag->tagged)
2711 die(_("unable to pack objects reachable from tag %s"),
2712 oid_to_hex(oid));
2714 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2716 if (tag->tagged->type != OBJ_TAG)
2717 return;
2719 tag = (struct tag *)tag->tagged;
2723 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2725 struct object_id peeled;
2727 if (starts_with(path, "refs/tags/") && /* is a tag? */
2728 !peel_ref(path, &peeled) && /* peelable? */
2729 obj_is_packed(&peeled)) /* object packed? */
2730 add_tag_chain(oid);
2731 return 0;
2734 static void prepare_pack(int window, int depth)
2736 struct object_entry **delta_list;
2737 uint32_t i, nr_deltas;
2738 unsigned n;
2740 if (use_delta_islands)
2741 resolve_tree_islands(the_repository, progress, &to_pack);
2743 get_object_details();
2746 * If we're locally repacking then we need to be doubly careful
2747 * from now on in order to make sure no stealth corruption gets
2748 * propagated to the new pack. Clients receiving streamed packs
2749 * should validate everything they get anyway so no need to incur
2750 * the additional cost here in that case.
2752 if (!pack_to_stdout)
2753 do_check_packed_object_crc = 1;
2755 if (!to_pack.nr_objects || !window || !depth)
2756 return;
2758 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2759 nr_deltas = n = 0;
2761 for (i = 0; i < to_pack.nr_objects; i++) {
2762 struct object_entry *entry = to_pack.objects + i;
2764 if (DELTA(entry))
2765 /* This happens if we decided to reuse existing
2766 * delta from a pack. "reuse_delta &&" is implied.
2768 continue;
2770 if (!entry->type_valid ||
2771 oe_size_less_than(&to_pack, entry, 50))
2772 continue;
2774 if (entry->no_try_delta)
2775 continue;
2777 if (!entry->preferred_base) {
2778 nr_deltas++;
2779 if (oe_type(entry) < 0)
2780 die(_("unable to get type of object %s"),
2781 oid_to_hex(&entry->idx.oid));
2782 } else {
2783 if (oe_type(entry) < 0) {
2785 * This object is not found, but we
2786 * don't have to include it anyway.
2788 continue;
2792 delta_list[n++] = entry;
2795 if (nr_deltas && n > 1) {
2796 unsigned nr_done = 0;
2798 if (progress)
2799 progress_state = start_progress(_("Compressing objects"),
2800 nr_deltas);
2801 QSORT(delta_list, n, type_size_sort);
2802 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2803 stop_progress(&progress_state);
2804 if (nr_done != nr_deltas)
2805 die(_("inconsistency with delta count"));
2807 free(delta_list);
2810 static int git_pack_config(const char *k, const char *v, void *cb)
2812 if (!strcmp(k, "pack.window")) {
2813 window = git_config_int(k, v);
2814 return 0;
2816 if (!strcmp(k, "pack.windowmemory")) {
2817 window_memory_limit = git_config_ulong(k, v);
2818 return 0;
2820 if (!strcmp(k, "pack.depth")) {
2821 depth = git_config_int(k, v);
2822 return 0;
2824 if (!strcmp(k, "pack.deltacachesize")) {
2825 max_delta_cache_size = git_config_int(k, v);
2826 return 0;
2828 if (!strcmp(k, "pack.deltacachelimit")) {
2829 cache_max_small_delta_size = git_config_int(k, v);
2830 return 0;
2832 if (!strcmp(k, "pack.writebitmaphashcache")) {
2833 if (git_config_bool(k, v))
2834 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2835 else
2836 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2838 if (!strcmp(k, "pack.usebitmaps")) {
2839 use_bitmap_index_default = git_config_bool(k, v);
2840 return 0;
2842 if (!strcmp(k, "pack.allowpackreuse")) {
2843 allow_pack_reuse = git_config_bool(k, v);
2844 return 0;
2846 if (!strcmp(k, "pack.threads")) {
2847 delta_search_threads = git_config_int(k, v);
2848 if (delta_search_threads < 0)
2849 die(_("invalid number of threads specified (%d)"),
2850 delta_search_threads);
2851 if (!HAVE_THREADS && delta_search_threads != 1) {
2852 warning(_("no threads support, ignoring %s"), k);
2853 delta_search_threads = 0;
2855 return 0;
2857 if (!strcmp(k, "pack.indexversion")) {
2858 pack_idx_opts.version = git_config_int(k, v);
2859 if (pack_idx_opts.version > 2)
2860 die(_("bad pack.indexversion=%"PRIu32),
2861 pack_idx_opts.version);
2862 return 0;
2864 return git_default_config(k, v, cb);
2867 static void read_object_list_from_stdin(void)
2869 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2870 struct object_id oid;
2871 const char *p;
2873 for (;;) {
2874 if (!fgets(line, sizeof(line), stdin)) {
2875 if (feof(stdin))
2876 break;
2877 if (!ferror(stdin))
2878 die("BUG: fgets returned NULL, not EOF, not error!");
2879 if (errno != EINTR)
2880 die_errno("fgets");
2881 clearerr(stdin);
2882 continue;
2884 if (line[0] == '-') {
2885 if (get_oid_hex(line+1, &oid))
2886 die(_("expected edge object ID, got garbage:\n %s"),
2887 line);
2888 add_preferred_base(&oid);
2889 continue;
2891 if (parse_oid_hex(line, &oid, &p))
2892 die(_("expected object ID, got garbage:\n %s"), line);
2894 add_preferred_base_object(p + 1);
2895 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2899 /* Remember to update object flag allocation in object.h */
2900 #define OBJECT_ADDED (1u<<20)
2902 static void show_commit(struct commit *commit, void *data)
2904 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2905 commit->object.flags |= OBJECT_ADDED;
2907 if (write_bitmap_index)
2908 index_commit_for_bitmap(commit);
2910 if (use_delta_islands)
2911 propagate_island_marks(commit);
2914 static void show_object(struct object *obj, const char *name, void *data)
2916 add_preferred_base_object(name);
2917 add_object_entry(&obj->oid, obj->type, name, 0);
2918 obj->flags |= OBJECT_ADDED;
2920 if (use_delta_islands) {
2921 const char *p;
2922 unsigned depth;
2923 struct object_entry *ent;
2925 /* the empty string is a root tree, which is depth 0 */
2926 depth = *name ? 1 : 0;
2927 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2928 depth++;
2930 ent = packlist_find(&to_pack, &obj->oid);
2931 if (ent && depth > oe_tree_depth(&to_pack, ent))
2932 oe_set_tree_depth(&to_pack, ent, depth);
2936 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2938 assert(arg_missing_action == MA_ALLOW_ANY);
2941 * Quietly ignore ALL missing objects. This avoids problems with
2942 * staging them now and getting an odd error later.
2944 if (!has_object_file(&obj->oid))
2945 return;
2947 show_object(obj, name, data);
2950 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2952 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2955 * Quietly ignore EXPECTED missing objects. This avoids problems with
2956 * staging them now and getting an odd error later.
2958 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2959 return;
2961 show_object(obj, name, data);
2964 static int option_parse_missing_action(const struct option *opt,
2965 const char *arg, int unset)
2967 assert(arg);
2968 assert(!unset);
2970 if (!strcmp(arg, "error")) {
2971 arg_missing_action = MA_ERROR;
2972 fn_show_object = show_object;
2973 return 0;
2976 if (!strcmp(arg, "allow-any")) {
2977 arg_missing_action = MA_ALLOW_ANY;
2978 fetch_if_missing = 0;
2979 fn_show_object = show_object__ma_allow_any;
2980 return 0;
2983 if (!strcmp(arg, "allow-promisor")) {
2984 arg_missing_action = MA_ALLOW_PROMISOR;
2985 fetch_if_missing = 0;
2986 fn_show_object = show_object__ma_allow_promisor;
2987 return 0;
2990 die(_("invalid value for --missing"));
2991 return 0;
2994 static void show_edge(struct commit *commit)
2996 add_preferred_base(&commit->object.oid);
2999 struct in_pack_object {
3000 off_t offset;
3001 struct object *object;
3004 struct in_pack {
3005 unsigned int alloc;
3006 unsigned int nr;
3007 struct in_pack_object *array;
3010 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
3012 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
3013 in_pack->array[in_pack->nr].object = object;
3014 in_pack->nr++;
3018 * Compare the objects in the offset order, in order to emulate the
3019 * "git rev-list --objects" output that produced the pack originally.
3021 static int ofscmp(const void *a_, const void *b_)
3023 struct in_pack_object *a = (struct in_pack_object *)a_;
3024 struct in_pack_object *b = (struct in_pack_object *)b_;
3026 if (a->offset < b->offset)
3027 return -1;
3028 else if (a->offset > b->offset)
3029 return 1;
3030 else
3031 return oidcmp(&a->object->oid, &b->object->oid);
3034 static void add_objects_in_unpacked_packs(void)
3036 struct packed_git *p;
3037 struct in_pack in_pack;
3038 uint32_t i;
3040 memset(&in_pack, 0, sizeof(in_pack));
3042 for (p = get_all_packs(the_repository); p; p = p->next) {
3043 struct object_id oid;
3044 struct object *o;
3046 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3047 continue;
3048 if (open_pack_index(p))
3049 die(_("cannot open pack index"));
3051 ALLOC_GROW(in_pack.array,
3052 in_pack.nr + p->num_objects,
3053 in_pack.alloc);
3055 for (i = 0; i < p->num_objects; i++) {
3056 nth_packed_object_oid(&oid, p, i);
3057 o = lookup_unknown_object(&oid);
3058 if (!(o->flags & OBJECT_ADDED))
3059 mark_in_pack_object(o, p, &in_pack);
3060 o->flags |= OBJECT_ADDED;
3064 if (in_pack.nr) {
3065 QSORT(in_pack.array, in_pack.nr, ofscmp);
3066 for (i = 0; i < in_pack.nr; i++) {
3067 struct object *o = in_pack.array[i].object;
3068 add_object_entry(&o->oid, o->type, "", 0);
3071 free(in_pack.array);
3074 static int add_loose_object(const struct object_id *oid, const char *path,
3075 void *data)
3077 enum object_type type = oid_object_info(the_repository, oid, NULL);
3079 if (type < 0) {
3080 warning(_("loose object at %s could not be examined"), path);
3081 return 0;
3084 add_object_entry(oid, type, "", 0);
3085 return 0;
3089 * We actually don't even have to worry about reachability here.
3090 * add_object_entry will weed out duplicates, so we just add every
3091 * loose object we find.
3093 static void add_unreachable_loose_objects(void)
3095 for_each_loose_file_in_objdir(get_object_directory(),
3096 add_loose_object,
3097 NULL, NULL, NULL);
3100 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
3102 static struct packed_git *last_found = (void *)1;
3103 struct packed_git *p;
3105 p = (last_found != (void *)1) ? last_found :
3106 get_all_packs(the_repository);
3108 while (p) {
3109 if ((!p->pack_local || p->pack_keep ||
3110 p->pack_keep_in_core) &&
3111 find_pack_entry_one(oid->hash, p)) {
3112 last_found = p;
3113 return 1;
3115 if (p == last_found)
3116 p = get_all_packs(the_repository);
3117 else
3118 p = p->next;
3119 if (p == last_found)
3120 p = p->next;
3122 return 0;
3126 * Store a list of sha1s that are should not be discarded
3127 * because they are either written too recently, or are
3128 * reachable from another object that was.
3130 * This is filled by get_object_list.
3132 static struct oid_array recent_objects;
3134 static int loosened_object_can_be_discarded(const struct object_id *oid,
3135 timestamp_t mtime)
3137 if (!unpack_unreachable_expiration)
3138 return 0;
3139 if (mtime > unpack_unreachable_expiration)
3140 return 0;
3141 if (oid_array_lookup(&recent_objects, oid) >= 0)
3142 return 0;
3143 return 1;
3146 static void loosen_unused_packed_objects(void)
3148 struct packed_git *p;
3149 uint32_t i;
3150 struct object_id oid;
3152 for (p = get_all_packs(the_repository); p; p = p->next) {
3153 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3154 continue;
3156 if (open_pack_index(p))
3157 die(_("cannot open pack index"));
3159 for (i = 0; i < p->num_objects; i++) {
3160 nth_packed_object_oid(&oid, p, i);
3161 if (!packlist_find(&to_pack, &oid) &&
3162 !has_sha1_pack_kept_or_nonlocal(&oid) &&
3163 !loosened_object_can_be_discarded(&oid, p->mtime))
3164 if (force_object_loose(&oid, p->mtime))
3165 die(_("unable to force loose object"));
3171 * This tracks any options which pack-reuse code expects to be on, or which a
3172 * reader of the pack might not understand, and which would therefore prevent
3173 * blind reuse of what we have on disk.
3175 static int pack_options_allow_reuse(void)
3177 return allow_pack_reuse &&
3178 pack_to_stdout &&
3179 !ignore_packed_keep_on_disk &&
3180 !ignore_packed_keep_in_core &&
3181 (!local || !have_non_local_packs) &&
3182 !incremental;
3185 static int get_object_list_from_bitmap(struct rev_info *revs)
3187 if (!(bitmap_git = prepare_bitmap_walk(revs)))
3188 return -1;
3190 if (pack_options_allow_reuse() &&
3191 !reuse_partial_packfile_from_bitmap(
3192 bitmap_git,
3193 &reuse_packfile,
3194 &reuse_packfile_objects,
3195 &reuse_packfile_bitmap)) {
3196 assert(reuse_packfile_objects);
3197 nr_result += reuse_packfile_objects;
3198 display_progress(progress_state, nr_result);
3201 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
3202 return 0;
3205 static void record_recent_object(struct object *obj,
3206 const char *name,
3207 void *data)
3209 oid_array_append(&recent_objects, &obj->oid);
3212 static void record_recent_commit(struct commit *commit, void *data)
3214 oid_array_append(&recent_objects, &commit->object.oid);
3217 static void get_object_list(int ac, const char **av)
3219 struct rev_info revs;
3220 struct setup_revision_opt s_r_opt = {
3221 .allow_exclude_promisor_objects = 1,
3223 char line[1000];
3224 int flags = 0;
3225 int save_warning;
3227 repo_init_revisions(the_repository, &revs, NULL);
3228 save_commit_buffer = 0;
3229 setup_revisions(ac, av, &revs, &s_r_opt);
3231 /* make sure shallows are read */
3232 is_repository_shallow(the_repository);
3234 save_warning = warn_on_object_refname_ambiguity;
3235 warn_on_object_refname_ambiguity = 0;
3237 while (fgets(line, sizeof(line), stdin) != NULL) {
3238 int len = strlen(line);
3239 if (len && line[len - 1] == '\n')
3240 line[--len] = 0;
3241 if (!len)
3242 break;
3243 if (*line == '-') {
3244 if (!strcmp(line, "--not")) {
3245 flags ^= UNINTERESTING;
3246 write_bitmap_index = 0;
3247 continue;
3249 if (starts_with(line, "--shallow ")) {
3250 struct object_id oid;
3251 if (get_oid_hex(line + 10, &oid))
3252 die("not an SHA-1 '%s'", line + 10);
3253 register_shallow(the_repository, &oid);
3254 use_bitmap_index = 0;
3255 continue;
3257 die(_("not a rev '%s'"), line);
3259 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3260 die(_("bad revision '%s'"), line);
3263 warn_on_object_refname_ambiguity = save_warning;
3265 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3266 return;
3268 if (use_delta_islands)
3269 load_delta_islands(the_repository, progress);
3271 if (prepare_revision_walk(&revs))
3272 die(_("revision walk setup failed"));
3273 mark_edges_uninteresting(&revs, show_edge, sparse);
3275 if (!fn_show_object)
3276 fn_show_object = show_object;
3277 traverse_commit_list_filtered(&filter_options, &revs,
3278 show_commit, fn_show_object, NULL,
3279 NULL);
3281 if (unpack_unreachable_expiration) {
3282 revs.ignore_missing_links = 1;
3283 if (add_unseen_recent_objects_to_traversal(&revs,
3284 unpack_unreachable_expiration))
3285 die(_("unable to add recent objects"));
3286 if (prepare_revision_walk(&revs))
3287 die(_("revision walk setup failed"));
3288 traverse_commit_list(&revs, record_recent_commit,
3289 record_recent_object, NULL);
3292 if (keep_unreachable)
3293 add_objects_in_unpacked_packs();
3294 if (pack_loose_unreachable)
3295 add_unreachable_loose_objects();
3296 if (unpack_unreachable)
3297 loosen_unused_packed_objects();
3299 oid_array_clear(&recent_objects);
3302 static void add_extra_kept_packs(const struct string_list *names)
3304 struct packed_git *p;
3306 if (!names->nr)
3307 return;
3309 for (p = get_all_packs(the_repository); p; p = p->next) {
3310 const char *name = basename(p->pack_name);
3311 int i;
3313 if (!p->pack_local)
3314 continue;
3316 for (i = 0; i < names->nr; i++)
3317 if (!fspathcmp(name, names->items[i].string))
3318 break;
3320 if (i < names->nr) {
3321 p->pack_keep_in_core = 1;
3322 ignore_packed_keep_in_core = 1;
3323 continue;
3328 static int option_parse_index_version(const struct option *opt,
3329 const char *arg, int unset)
3331 char *c;
3332 const char *val = arg;
3334 BUG_ON_OPT_NEG(unset);
3336 pack_idx_opts.version = strtoul(val, &c, 10);
3337 if (pack_idx_opts.version > 2)
3338 die(_("unsupported index version %s"), val);
3339 if (*c == ',' && c[1])
3340 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3341 if (*c || pack_idx_opts.off32_limit & 0x80000000)
3342 die(_("bad index version '%s'"), val);
3343 return 0;
3346 static int option_parse_unpack_unreachable(const struct option *opt,
3347 const char *arg, int unset)
3349 if (unset) {
3350 unpack_unreachable = 0;
3351 unpack_unreachable_expiration = 0;
3353 else {
3354 unpack_unreachable = 1;
3355 if (arg)
3356 unpack_unreachable_expiration = approxidate(arg);
3358 return 0;
3361 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3363 int use_internal_rev_list = 0;
3364 int shallow = 0;
3365 int all_progress_implied = 0;
3366 struct argv_array rp = ARGV_ARRAY_INIT;
3367 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3368 int rev_list_index = 0;
3369 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3370 struct option pack_objects_options[] = {
3371 OPT_SET_INT('q', "quiet", &progress,
3372 N_("do not show progress meter"), 0),
3373 OPT_SET_INT(0, "progress", &progress,
3374 N_("show progress meter"), 1),
3375 OPT_SET_INT(0, "all-progress", &progress,
3376 N_("show progress meter during object writing phase"), 2),
3377 OPT_BOOL(0, "all-progress-implied",
3378 &all_progress_implied,
3379 N_("similar to --all-progress when progress meter is shown")),
3380 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),
3381 N_("write the pack index file in the specified idx format version"),
3382 PARSE_OPT_NONEG, option_parse_index_version },
3383 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3384 N_("maximum size of each output pack file")),
3385 OPT_BOOL(0, "local", &local,
3386 N_("ignore borrowed objects from alternate object store")),
3387 OPT_BOOL(0, "incremental", &incremental,
3388 N_("ignore packed objects")),
3389 OPT_INTEGER(0, "window", &window,
3390 N_("limit pack window by objects")),
3391 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3392 N_("limit pack window by memory in addition to object limit")),
3393 OPT_INTEGER(0, "depth", &depth,
3394 N_("maximum length of delta chain allowed in the resulting pack")),
3395 OPT_BOOL(0, "reuse-delta", &reuse_delta,
3396 N_("reuse existing deltas")),
3397 OPT_BOOL(0, "reuse-object", &reuse_object,
3398 N_("reuse existing objects")),
3399 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3400 N_("use OFS_DELTA objects")),
3401 OPT_INTEGER(0, "threads", &delta_search_threads,
3402 N_("use threads when searching for best delta matches")),
3403 OPT_BOOL(0, "non-empty", &non_empty,
3404 N_("do not create an empty pack output")),
3405 OPT_BOOL(0, "revs", &use_internal_rev_list,
3406 N_("read revision arguments from standard input")),
3407 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3408 N_("limit the objects to those that are not yet packed"),
3409 1, PARSE_OPT_NONEG),
3410 OPT_SET_INT_F(0, "all", &rev_list_all,
3411 N_("include objects reachable from any reference"),
3412 1, PARSE_OPT_NONEG),
3413 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3414 N_("include objects referred by reflog entries"),
3415 1, PARSE_OPT_NONEG),
3416 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3417 N_("include objects referred to by the index"),
3418 1, PARSE_OPT_NONEG),
3419 OPT_BOOL(0, "stdout", &pack_to_stdout,
3420 N_("output pack to stdout")),
3421 OPT_BOOL(0, "include-tag", &include_tag,
3422 N_("include tag objects that refer to objects to be packed")),
3423 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3424 N_("keep unreachable objects")),
3425 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3426 N_("pack loose unreachable objects")),
3427 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3428 N_("unpack unreachable objects newer than <time>"),
3429 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3430 OPT_BOOL(0, "sparse", &sparse,
3431 N_("use the sparse reachability algorithm")),
3432 OPT_BOOL(0, "thin", &thin,
3433 N_("create thin packs")),
3434 OPT_BOOL(0, "shallow", &shallow,
3435 N_("create packs suitable for shallow fetches")),
3436 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3437 N_("ignore packs that have companion .keep file")),
3438 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3439 N_("ignore this pack")),
3440 OPT_INTEGER(0, "compression", &pack_compression_level,
3441 N_("pack compression level")),
3442 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3443 N_("do not hide commits by grafts"), 0),
3444 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3445 N_("use a bitmap index if available to speed up counting objects")),
3446 OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index,
3447 N_("write a bitmap index together with the pack index"),
3448 WRITE_BITMAP_TRUE),
3449 OPT_SET_INT_F(0, "write-bitmap-index-quiet",
3450 &write_bitmap_index,
3451 N_("write a bitmap index if possible"),
3452 WRITE_BITMAP_QUIET, PARSE_OPT_HIDDEN),
3453 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3454 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3455 N_("handling for missing objects"), PARSE_OPT_NONEG,
3456 option_parse_missing_action },
3457 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3458 N_("do not pack objects in promisor packfiles")),
3459 OPT_BOOL(0, "delta-islands", &use_delta_islands,
3460 N_("respect islands during delta compression")),
3461 OPT_END(),
3464 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3465 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3467 read_replace_refs = 0;
3469 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);
3470 prepare_repo_settings(the_repository);
3471 if (!sparse && the_repository->settings.pack_use_sparse != -1)
3472 sparse = the_repository->settings.pack_use_sparse;
3474 reset_pack_idx_option(&pack_idx_opts);
3475 git_config(git_pack_config, NULL);
3477 progress = isatty(2);
3478 argc = parse_options(argc, argv, prefix, pack_objects_options,
3479 pack_usage, 0);
3481 if (argc) {
3482 base_name = argv[0];
3483 argc--;
3485 if (pack_to_stdout != !base_name || argc)
3486 usage_with_options(pack_usage, pack_objects_options);
3488 if (depth >= (1 << OE_DEPTH_BITS)) {
3489 warning(_("delta chain depth %d is too deep, forcing %d"),
3490 depth, (1 << OE_DEPTH_BITS) - 1);
3491 depth = (1 << OE_DEPTH_BITS) - 1;
3493 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3494 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3495 (1U << OE_Z_DELTA_BITS) - 1);
3496 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3499 argv_array_push(&rp, "pack-objects");
3500 if (thin) {
3501 use_internal_rev_list = 1;
3502 argv_array_push(&rp, shallow
3503 ? "--objects-edge-aggressive"
3504 : "--objects-edge");
3505 } else
3506 argv_array_push(&rp, "--objects");
3508 if (rev_list_all) {
3509 use_internal_rev_list = 1;
3510 argv_array_push(&rp, "--all");
3512 if (rev_list_reflog) {
3513 use_internal_rev_list = 1;
3514 argv_array_push(&rp, "--reflog");
3516 if (rev_list_index) {
3517 use_internal_rev_list = 1;
3518 argv_array_push(&rp, "--indexed-objects");
3520 if (rev_list_unpacked) {
3521 use_internal_rev_list = 1;
3522 argv_array_push(&rp, "--unpacked");
3525 if (exclude_promisor_objects) {
3526 use_internal_rev_list = 1;
3527 fetch_if_missing = 0;
3528 argv_array_push(&rp, "--exclude-promisor-objects");
3530 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3531 use_internal_rev_list = 1;
3533 if (!reuse_object)
3534 reuse_delta = 0;
3535 if (pack_compression_level == -1)
3536 pack_compression_level = Z_DEFAULT_COMPRESSION;
3537 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3538 die(_("bad pack compression level %d"), pack_compression_level);
3540 if (!delta_search_threads) /* --threads=0 means autodetect */
3541 delta_search_threads = online_cpus();
3543 if (!HAVE_THREADS && delta_search_threads != 1)
3544 warning(_("no threads support, ignoring --threads"));
3545 if (!pack_to_stdout && !pack_size_limit)
3546 pack_size_limit = pack_size_limit_cfg;
3547 if (pack_to_stdout && pack_size_limit)
3548 die(_("--max-pack-size cannot be used to build a pack for transfer"));
3549 if (pack_size_limit && pack_size_limit < 1024*1024) {
3550 warning(_("minimum pack size limit is 1 MiB"));
3551 pack_size_limit = 1024*1024;
3554 if (!pack_to_stdout && thin)
3555 die(_("--thin cannot be used to build an indexable pack"));
3557 if (keep_unreachable && unpack_unreachable)
3558 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));
3559 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3560 unpack_unreachable_expiration = 0;
3562 if (filter_options.choice) {
3563 if (!pack_to_stdout)
3564 die(_("cannot use --filter without --stdout"));
3565 use_bitmap_index = 0;
3569 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3571 * - to produce good pack (with bitmap index not-yet-packed objects are
3572 * packed in suboptimal order).
3574 * - to use more robust pack-generation codepath (avoiding possible
3575 * bugs in bitmap code and possible bitmap index corruption).
3577 if (!pack_to_stdout)
3578 use_bitmap_index_default = 0;
3580 if (use_bitmap_index < 0)
3581 use_bitmap_index = use_bitmap_index_default;
3583 /* "hard" reasons not to use bitmaps; these just won't work at all */
3584 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))
3585 use_bitmap_index = 0;
3587 if (pack_to_stdout || !rev_list_all)
3588 write_bitmap_index = 0;
3590 if (use_delta_islands)
3591 argv_array_push(&rp, "--topo-order");
3593 if (progress && all_progress_implied)
3594 progress = 2;
3596 add_extra_kept_packs(&keep_pack_list);
3597 if (ignore_packed_keep_on_disk) {
3598 struct packed_git *p;
3599 for (p = get_all_packs(the_repository); p; p = p->next)
3600 if (p->pack_local && p->pack_keep)
3601 break;
3602 if (!p) /* no keep-able packs found */
3603 ignore_packed_keep_on_disk = 0;
3605 if (local) {
3607 * unlike ignore_packed_keep_on_disk above, we do not
3608 * want to unset "local" based on looking at packs, as
3609 * it also covers non-local objects
3611 struct packed_git *p;
3612 for (p = get_all_packs(the_repository); p; p = p->next) {
3613 if (!p->pack_local) {
3614 have_non_local_packs = 1;
3615 break;
3620 trace2_region_enter("pack-objects", "enumerate-objects",
3621 the_repository);
3622 prepare_packing_data(the_repository, &to_pack);
3624 if (progress)
3625 progress_state = start_progress(_("Enumerating objects"), 0);
3626 if (!use_internal_rev_list)
3627 read_object_list_from_stdin();
3628 else {
3629 get_object_list(rp.argc, rp.argv);
3630 argv_array_clear(&rp);
3632 cleanup_preferred_base();
3633 if (include_tag && nr_result)
3634 for_each_ref(add_ref_tag, NULL);
3635 stop_progress(&progress_state);
3636 trace2_region_leave("pack-objects", "enumerate-objects",
3637 the_repository);
3639 if (non_empty && !nr_result)
3640 return 0;
3641 if (nr_result) {
3642 trace2_region_enter("pack-objects", "prepare-pack",
3643 the_repository);
3644 prepare_pack(window, depth);
3645 trace2_region_leave("pack-objects", "prepare-pack",
3646 the_repository);
3649 trace2_region_enter("pack-objects", "write-pack-file", the_repository);
3650 write_pack_file();
3651 trace2_region_leave("pack-objects", "write-pack-file", the_repository);
3653 if (progress)
3654 fprintf_ln(stderr,
3655 _("Total %"PRIu32" (delta %"PRIu32"),"
3656 " reused %"PRIu32" (delta %"PRIu32"),"
3657 " pack-reused %"PRIu32),
3658 written, written_delta, reused, reused_delta,
3659 reuse_packfile_objects);
3660 return 0;