Make git_check_attr() a void function
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "attr.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "tree.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
16 #include "diff.h"
17 #include "revision.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.h"
22 #include "progress.h"
23 #include "refs.h"
24 #include "streaming.h"
25 #include "thread-utils.h"
26 #include "pack-bitmap.h"
27 #include "reachable.h"
28 #include "sha1-array.h"
29 #include "argv-array.h"
30 #include "list.h"
31 #include "packfile.h"
32 #include "object-store.h"
33 #include "dir.h"
34
35 #define IN_PACK(obj) oe_in_pack(&to_pack, obj)
36 #define SIZE(obj) oe_size(&to_pack, obj)
37 #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
38 #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
39 #define DELTA(obj) oe_delta(&to_pack, obj)
40 #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
41 #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
42 #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
43 #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
44 #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
45 #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
46
47 static const char *pack_usage[] = {
48 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
49 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
50 NULL
51 };
52
53 /*
54 * Objects we are going to pack are collected in the `to_pack` structure.
55 * It contains an array (dynamically expanded) of the object data, and a map
56 * that can resolve SHA1s to their position in the array.
57 */
58 static struct packing_data to_pack;
59
60 static struct pack_idx_entry **written_list;
61 static uint32_t nr_result, nr_written, nr_seen;
62
63 static int non_empty;
64 static int reuse_delta = 1, reuse_object = 1;
65 static int keep_unreachable, unpack_unreachable, include_tag;
66 static timestamp_t unpack_unreachable_expiration;
67 static int pack_loose_unreachable;
68 static int local;
69 static int have_non_local_packs;
70 static int incremental;
71 static int ignore_packed_keep_on_disk;
72 static int ignore_packed_keep_in_core;
73 static int allow_ofs_delta;
74 static struct pack_idx_option pack_idx_opts;
75 static const char *base_name;
76 static int progress = 1;
77 static int window = 10;
78 static unsigned long pack_size_limit;
79 static int depth = 50;
80 static int delta_search_threads;
81 static int pack_to_stdout;
82 static int num_preferred_base;
83 static struct progress *progress_state;
84
85 static struct packed_git *reuse_packfile;
86 static uint32_t reuse_packfile_objects;
87 static off_t reuse_packfile_offset;
88
89 static int use_bitmap_index_default = 1;
90 static int use_bitmap_index = -1;
91 static int write_bitmap_index;
92 static uint16_t write_bitmap_options;
93
94 static int exclude_promisor_objects;
95
96 static unsigned long delta_cache_size = 0;
97 static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
98 static unsigned long cache_max_small_delta_size = 1000;
99
100 static unsigned long window_memory_limit = 0;
101
102 static struct list_objects_filter_options filter_options;
103
104 enum missing_action {
105 MA_ERROR = 0, /* fail if any missing objects are encountered */
106 MA_ALLOW_ANY, /* silently allow ALL missing objects */
107 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
108 };
109 static enum missing_action arg_missing_action;
110 static show_object_fn fn_show_object;
111
112 /*
113 * stats
114 */
115 static uint32_t written, written_delta;
116 static uint32_t reused, reused_delta;
117
118 /*
119 * Indexed commits
120 */
121 static struct commit **indexed_commits;
122 static unsigned int indexed_commits_nr;
123 static unsigned int indexed_commits_alloc;
124
125 static void index_commit_for_bitmap(struct commit *commit)
126 {
127 if (indexed_commits_nr >= indexed_commits_alloc) {
128 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
129 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
130 }
131
132 indexed_commits[indexed_commits_nr++] = commit;
133 }
134
135 static void *get_delta(struct object_entry *entry)
136 {
137 unsigned long size, base_size, delta_size;
138 void *buf, *base_buf, *delta_buf;
139 enum object_type type;
140
141 buf = read_object_file(&entry->idx.oid, &type, &size);
142 if (!buf)
143 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
144 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
145 &base_size);
146 if (!base_buf)
147 die("unable to read %s",
148 oid_to_hex(&DELTA(entry)->idx.oid));
149 delta_buf = diff_delta(base_buf, base_size,
150 buf, size, &delta_size, 0);
151 /*
152 * We succesfully computed this delta once but dropped it for
153 * memory reasons. Something is very wrong if this time we
154 * recompute and create a different delta.
155 */
156 if (!delta_buf || delta_size != DELTA_SIZE(entry))
157 BUG("delta size changed");
158 free(buf);
159 free(base_buf);
160 return delta_buf;
161 }
162
163 static unsigned long do_compress(void **pptr, unsigned long size)
164 {
165 git_zstream stream;
166 void *in, *out;
167 unsigned long maxsize;
168
169 git_deflate_init(&stream, pack_compression_level);
170 maxsize = git_deflate_bound(&stream, size);
171
172 in = *pptr;
173 out = xmalloc(maxsize);
174 *pptr = out;
175
176 stream.next_in = in;
177 stream.avail_in = size;
178 stream.next_out = out;
179 stream.avail_out = maxsize;
180 while (git_deflate(&stream, Z_FINISH) == Z_OK)
181 ; /* nothing */
182 git_deflate_end(&stream);
183
184 free(in);
185 return stream.total_out;
186 }
187
188 static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
189 const struct object_id *oid)
190 {
191 git_zstream stream;
192 unsigned char ibuf[1024 * 16];
193 unsigned char obuf[1024 * 16];
194 unsigned long olen = 0;
195
196 git_deflate_init(&stream, pack_compression_level);
197
198 for (;;) {
199 ssize_t readlen;
200 int zret = Z_OK;
201 readlen = read_istream(st, ibuf, sizeof(ibuf));
202 if (readlen == -1)
203 die(_("unable to read %s"), oid_to_hex(oid));
204
205 stream.next_in = ibuf;
206 stream.avail_in = readlen;
207 while ((stream.avail_in || readlen == 0) &&
208 (zret == Z_OK || zret == Z_BUF_ERROR)) {
209 stream.next_out = obuf;
210 stream.avail_out = sizeof(obuf);
211 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
212 hashwrite(f, obuf, stream.next_out - obuf);
213 olen += stream.next_out - obuf;
214 }
215 if (stream.avail_in)
216 die(_("deflate error (%d)"), zret);
217 if (readlen == 0) {
218 if (zret != Z_STREAM_END)
219 die(_("deflate error (%d)"), zret);
220 break;
221 }
222 }
223 git_deflate_end(&stream);
224 return olen;
225 }
226
227 /*
228 * we are going to reuse the existing object data as is. make
229 * sure it is not corrupt.
230 */
231 static int check_pack_inflate(struct packed_git *p,
232 struct pack_window **w_curs,
233 off_t offset,
234 off_t len,
235 unsigned long expect)
236 {
237 git_zstream stream;
238 unsigned char fakebuf[4096], *in;
239 int st;
240
241 memset(&stream, 0, sizeof(stream));
242 git_inflate_init(&stream);
243 do {
244 in = use_pack(p, w_curs, offset, &stream.avail_in);
245 stream.next_in = in;
246 stream.next_out = fakebuf;
247 stream.avail_out = sizeof(fakebuf);
248 st = git_inflate(&stream, Z_FINISH);
249 offset += stream.next_in - in;
250 } while (st == Z_OK || st == Z_BUF_ERROR);
251 git_inflate_end(&stream);
252 return (st == Z_STREAM_END &&
253 stream.total_out == expect &&
254 stream.total_in == len) ? 0 : -1;
255 }
256
257 static void copy_pack_data(struct hashfile *f,
258 struct packed_git *p,
259 struct pack_window **w_curs,
260 off_t offset,
261 off_t len)
262 {
263 unsigned char *in;
264 unsigned long avail;
265
266 while (len) {
267 in = use_pack(p, w_curs, offset, &avail);
268 if (avail > len)
269 avail = (unsigned long)len;
270 hashwrite(f, in, avail);
271 offset += avail;
272 len -= avail;
273 }
274 }
275
276 /* Return 0 if we will bust the pack-size limit */
277 static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
278 unsigned long limit, int usable_delta)
279 {
280 unsigned long size, datalen;
281 unsigned char header[MAX_PACK_OBJECT_HEADER],
282 dheader[MAX_PACK_OBJECT_HEADER];
283 unsigned hdrlen;
284 enum object_type type;
285 void *buf;
286 struct git_istream *st = NULL;
287 const unsigned hashsz = the_hash_algo->rawsz;
288
289 if (!usable_delta) {
290 if (oe_type(entry) == OBJ_BLOB &&
291 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
292 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
293 buf = NULL;
294 else {
295 buf = read_object_file(&entry->idx.oid, &type, &size);
296 if (!buf)
297 die(_("unable to read %s"),
298 oid_to_hex(&entry->idx.oid));
299 }
300 /*
301 * make sure no cached delta data remains from a
302 * previous attempt before a pack split occurred.
303 */
304 FREE_AND_NULL(entry->delta_data);
305 entry->z_delta_size = 0;
306 } else if (entry->delta_data) {
307 size = DELTA_SIZE(entry);
308 buf = entry->delta_data;
309 entry->delta_data = NULL;
310 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
311 OBJ_OFS_DELTA : OBJ_REF_DELTA;
312 } else {
313 buf = get_delta(entry);
314 size = DELTA_SIZE(entry);
315 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
316 OBJ_OFS_DELTA : OBJ_REF_DELTA;
317 }
318
319 if (st) /* large blob case, just assume we don't compress well */
320 datalen = size;
321 else if (entry->z_delta_size)
322 datalen = entry->z_delta_size;
323 else
324 datalen = do_compress(&buf, size);
325
326 /*
327 * The object header is a byte of 'type' followed by zero or
328 * more bytes of length.
329 */
330 hdrlen = encode_in_pack_object_header(header, sizeof(header),
331 type, size);
332
333 if (type == OBJ_OFS_DELTA) {
334 /*
335 * Deltas with relative base contain an additional
336 * encoding of the relative offset for the delta
337 * base from this object's position in the pack.
338 */
339 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
340 unsigned pos = sizeof(dheader) - 1;
341 dheader[pos] = ofs & 127;
342 while (ofs >>= 7)
343 dheader[--pos] = 128 | (--ofs & 127);
344 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
345 if (st)
346 close_istream(st);
347 free(buf);
348 return 0;
349 }
350 hashwrite(f, header, hdrlen);
351 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
352 hdrlen += sizeof(dheader) - pos;
353 } else if (type == OBJ_REF_DELTA) {
354 /*
355 * Deltas with a base reference contain
356 * additional bytes for the base object ID.
357 */
358 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
359 if (st)
360 close_istream(st);
361 free(buf);
362 return 0;
363 }
364 hashwrite(f, header, hdrlen);
365 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
366 hdrlen += hashsz;
367 } else {
368 if (limit && hdrlen + datalen + hashsz >= limit) {
369 if (st)
370 close_istream(st);
371 free(buf);
372 return 0;
373 }
374 hashwrite(f, header, hdrlen);
375 }
376 if (st) {
377 datalen = write_large_blob_data(st, f, &entry->idx.oid);
378 close_istream(st);
379 } else {
380 hashwrite(f, buf, datalen);
381 free(buf);
382 }
383
384 return hdrlen + datalen;
385 }
386
387 /* Return 0 if we will bust the pack-size limit */
388 static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
389 unsigned long limit, int usable_delta)
390 {
391 struct packed_git *p = IN_PACK(entry);
392 struct pack_window *w_curs = NULL;
393 struct revindex_entry *revidx;
394 off_t offset;
395 enum object_type type = oe_type(entry);
396 off_t datalen;
397 unsigned char header[MAX_PACK_OBJECT_HEADER],
398 dheader[MAX_PACK_OBJECT_HEADER];
399 unsigned hdrlen;
400 const unsigned hashsz = the_hash_algo->rawsz;
401 unsigned long entry_size = SIZE(entry);
402
403 if (DELTA(entry))
404 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
405 OBJ_OFS_DELTA : OBJ_REF_DELTA;
406 hdrlen = encode_in_pack_object_header(header, sizeof(header),
407 type, entry_size);
408
409 offset = entry->in_pack_offset;
410 revidx = find_pack_revindex(p, offset);
411 datalen = revidx[1].offset - offset;
412 if (!pack_to_stdout && p->index_version > 1 &&
413 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
414 error(_("bad packed object CRC for %s"),
415 oid_to_hex(&entry->idx.oid));
416 unuse_pack(&w_curs);
417 return write_no_reuse_object(f, entry, limit, usable_delta);
418 }
419
420 offset += entry->in_pack_header_size;
421 datalen -= entry->in_pack_header_size;
422
423 if (!pack_to_stdout && p->index_version == 1 &&
424 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
425 error(_("corrupt packed object for %s"),
426 oid_to_hex(&entry->idx.oid));
427 unuse_pack(&w_curs);
428 return write_no_reuse_object(f, entry, limit, usable_delta);
429 }
430
431 if (type == OBJ_OFS_DELTA) {
432 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
433 unsigned pos = sizeof(dheader) - 1;
434 dheader[pos] = ofs & 127;
435 while (ofs >>= 7)
436 dheader[--pos] = 128 | (--ofs & 127);
437 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
438 unuse_pack(&w_curs);
439 return 0;
440 }
441 hashwrite(f, header, hdrlen);
442 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
443 hdrlen += sizeof(dheader) - pos;
444 reused_delta++;
445 } else if (type == OBJ_REF_DELTA) {
446 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
447 unuse_pack(&w_curs);
448 return 0;
449 }
450 hashwrite(f, header, hdrlen);
451 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
452 hdrlen += hashsz;
453 reused_delta++;
454 } else {
455 if (limit && hdrlen + datalen + hashsz >= limit) {
456 unuse_pack(&w_curs);
457 return 0;
458 }
459 hashwrite(f, header, hdrlen);
460 }
461 copy_pack_data(f, p, &w_curs, offset, datalen);
462 unuse_pack(&w_curs);
463 reused++;
464 return hdrlen + datalen;
465 }
466
467 /* Return 0 if we will bust the pack-size limit */
468 static off_t write_object(struct hashfile *f,
469 struct object_entry *entry,
470 off_t write_offset)
471 {
472 unsigned long limit;
473 off_t len;
474 int usable_delta, to_reuse;
475
476 if (!pack_to_stdout)
477 crc32_begin(f);
478
479 /* apply size limit if limited packsize and not first object */
480 if (!pack_size_limit || !nr_written)
481 limit = 0;
482 else if (pack_size_limit <= write_offset)
483 /*
484 * the earlier object did not fit the limit; avoid
485 * mistaking this with unlimited (i.e. limit = 0).
486 */
487 limit = 1;
488 else
489 limit = pack_size_limit - write_offset;
490
491 if (!DELTA(entry))
492 usable_delta = 0; /* no delta */
493 else if (!pack_size_limit)
494 usable_delta = 1; /* unlimited packfile */
495 else if (DELTA(entry)->idx.offset == (off_t)-1)
496 usable_delta = 0; /* base was written to another pack */
497 else if (DELTA(entry)->idx.offset)
498 usable_delta = 1; /* base already exists in this pack */
499 else
500 usable_delta = 0; /* base could end up in another pack */
501
502 if (!reuse_object)
503 to_reuse = 0; /* explicit */
504 else if (!IN_PACK(entry))
505 to_reuse = 0; /* can't reuse what we don't have */
506 else if (oe_type(entry) == OBJ_REF_DELTA ||
507 oe_type(entry) == OBJ_OFS_DELTA)
508 /* check_object() decided it for us ... */
509 to_reuse = usable_delta;
510 /* ... but pack split may override that */
511 else if (oe_type(entry) != entry->in_pack_type)
512 to_reuse = 0; /* pack has delta which is unusable */
513 else if (DELTA(entry))
514 to_reuse = 0; /* we want to pack afresh */
515 else
516 to_reuse = 1; /* we have it in-pack undeltified,
517 * and we do not need to deltify it.
518 */
519
520 if (!to_reuse)
521 len = write_no_reuse_object(f, entry, limit, usable_delta);
522 else
523 len = write_reuse_object(f, entry, limit, usable_delta);
524 if (!len)
525 return 0;
526
527 if (usable_delta)
528 written_delta++;
529 written++;
530 if (!pack_to_stdout)
531 entry->idx.crc32 = crc32_end(f);
532 return len;
533 }
534
535 enum write_one_status {
536 WRITE_ONE_SKIP = -1, /* already written */
537 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
538 WRITE_ONE_WRITTEN = 1, /* normal */
539 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
540 };
541
542 static enum write_one_status write_one(struct hashfile *f,
543 struct object_entry *e,
544 off_t *offset)
545 {
546 off_t size;
547 int recursing;
548
549 /*
550 * we set offset to 1 (which is an impossible value) to mark
551 * the fact that this object is involved in "write its base
552 * first before writing a deltified object" recursion.
553 */
554 recursing = (e->idx.offset == 1);
555 if (recursing) {
556 warning(_("recursive delta detected for object %s"),
557 oid_to_hex(&e->idx.oid));
558 return WRITE_ONE_RECURSIVE;
559 } else if (e->idx.offset || e->preferred_base) {
560 /* offset is non zero if object is written already. */
561 return WRITE_ONE_SKIP;
562 }
563
564 /* if we are deltified, write out base object first. */
565 if (DELTA(e)) {
566 e->idx.offset = 1; /* now recurse */
567 switch (write_one(f, DELTA(e), offset)) {
568 case WRITE_ONE_RECURSIVE:
569 /* we cannot depend on this one */
570 SET_DELTA(e, NULL);
571 break;
572 default:
573 break;
574 case WRITE_ONE_BREAK:
575 e->idx.offset = recursing;
576 return WRITE_ONE_BREAK;
577 }
578 }
579
580 e->idx.offset = *offset;
581 size = write_object(f, e, *offset);
582 if (!size) {
583 e->idx.offset = recursing;
584 return WRITE_ONE_BREAK;
585 }
586 written_list[nr_written++] = &e->idx;
587
588 /* make sure off_t is sufficiently large not to wrap */
589 if (signed_add_overflows(*offset, size))
590 die(_("pack too large for current definition of off_t"));
591 *offset += size;
592 return WRITE_ONE_WRITTEN;
593 }
594
595 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
596 void *cb_data)
597 {
598 struct object_id peeled;
599 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
600
601 if (entry)
602 entry->tagged = 1;
603 if (!peel_ref(path, &peeled)) {
604 entry = packlist_find(&to_pack, peeled.hash, NULL);
605 if (entry)
606 entry->tagged = 1;
607 }
608 return 0;
609 }
610
611 static inline void add_to_write_order(struct object_entry **wo,
612 unsigned int *endp,
613 struct object_entry *e)
614 {
615 if (e->filled)
616 return;
617 wo[(*endp)++] = e;
618 e->filled = 1;
619 }
620
621 static void add_descendants_to_write_order(struct object_entry **wo,
622 unsigned int *endp,
623 struct object_entry *e)
624 {
625 int add_to_order = 1;
626 while (e) {
627 if (add_to_order) {
628 struct object_entry *s;
629 /* add this node... */
630 add_to_write_order(wo, endp, e);
631 /* all its siblings... */
632 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
633 add_to_write_order(wo, endp, s);
634 }
635 }
636 /* drop down a level to add left subtree nodes if possible */
637 if (DELTA_CHILD(e)) {
638 add_to_order = 1;
639 e = DELTA_CHILD(e);
640 } else {
641 add_to_order = 0;
642 /* our sibling might have some children, it is next */
643 if (DELTA_SIBLING(e)) {
644 e = DELTA_SIBLING(e);
645 continue;
646 }
647 /* go back to our parent node */
648 e = DELTA(e);
649 while (e && !DELTA_SIBLING(e)) {
650 /* we're on the right side of a subtree, keep
651 * going up until we can go right again */
652 e = DELTA(e);
653 }
654 if (!e) {
655 /* done- we hit our original root node */
656 return;
657 }
658 /* pass it off to sibling at this level */
659 e = DELTA_SIBLING(e);
660 }
661 };
662 }
663
664 static void add_family_to_write_order(struct object_entry **wo,
665 unsigned int *endp,
666 struct object_entry *e)
667 {
668 struct object_entry *root;
669
670 for (root = e; DELTA(root); root = DELTA(root))
671 ; /* nothing */
672 add_descendants_to_write_order(wo, endp, root);
673 }
674
675 static struct object_entry **compute_write_order(void)
676 {
677 unsigned int i, wo_end, last_untagged;
678
679 struct object_entry **wo;
680 struct object_entry *objects = to_pack.objects;
681
682 for (i = 0; i < to_pack.nr_objects; i++) {
683 objects[i].tagged = 0;
684 objects[i].filled = 0;
685 SET_DELTA_CHILD(&objects[i], NULL);
686 SET_DELTA_SIBLING(&objects[i], NULL);
687 }
688
689 /*
690 * Fully connect delta_child/delta_sibling network.
691 * Make sure delta_sibling is sorted in the original
692 * recency order.
693 */
694 for (i = to_pack.nr_objects; i > 0;) {
695 struct object_entry *e = &objects[--i];
696 if (!DELTA(e))
697 continue;
698 /* Mark me as the first child */
699 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
700 SET_DELTA_CHILD(DELTA(e), e);
701 }
702
703 /*
704 * Mark objects that are at the tip of tags.
705 */
706 for_each_tag_ref(mark_tagged, NULL);
707
708 /*
709 * Give the objects in the original recency order until
710 * we see a tagged tip.
711 */
712 ALLOC_ARRAY(wo, to_pack.nr_objects);
713 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
714 if (objects[i].tagged)
715 break;
716 add_to_write_order(wo, &wo_end, &objects[i]);
717 }
718 last_untagged = i;
719
720 /*
721 * Then fill all the tagged tips.
722 */
723 for (; i < to_pack.nr_objects; i++) {
724 if (objects[i].tagged)
725 add_to_write_order(wo, &wo_end, &objects[i]);
726 }
727
728 /*
729 * And then all remaining commits and tags.
730 */
731 for (i = last_untagged; i < to_pack.nr_objects; i++) {
732 if (oe_type(&objects[i]) != OBJ_COMMIT &&
733 oe_type(&objects[i]) != OBJ_TAG)
734 continue;
735 add_to_write_order(wo, &wo_end, &objects[i]);
736 }
737
738 /*
739 * And then all the trees.
740 */
741 for (i = last_untagged; i < to_pack.nr_objects; i++) {
742 if (oe_type(&objects[i]) != OBJ_TREE)
743 continue;
744 add_to_write_order(wo, &wo_end, &objects[i]);
745 }
746
747 /*
748 * Finally all the rest in really tight order
749 */
750 for (i = last_untagged; i < to_pack.nr_objects; i++) {
751 if (!objects[i].filled)
752 add_family_to_write_order(wo, &wo_end, &objects[i]);
753 }
754
755 if (wo_end != to_pack.nr_objects)
756 die(_("ordered %u objects, expected %"PRIu32),
757 wo_end, to_pack.nr_objects);
758
759 return wo;
760 }
761
762 static off_t write_reused_pack(struct hashfile *f)
763 {
764 unsigned char buffer[8192];
765 off_t to_write, total;
766 int fd;
767
768 if (!is_pack_valid(reuse_packfile))
769 die(_("packfile is invalid: %s"), reuse_packfile->pack_name);
770
771 fd = git_open(reuse_packfile->pack_name);
772 if (fd < 0)
773 die_errno(_("unable to open packfile for reuse: %s"),
774 reuse_packfile->pack_name);
775
776 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
777 die_errno(_("unable to seek in reused packfile"));
778
779 if (reuse_packfile_offset < 0)
780 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
781
782 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
783
784 while (to_write) {
785 int read_pack = xread(fd, buffer, sizeof(buffer));
786
787 if (read_pack <= 0)
788 die_errno(_("unable to read from reused packfile"));
789
790 if (read_pack > to_write)
791 read_pack = to_write;
792
793 hashwrite(f, buffer, read_pack);
794 to_write -= read_pack;
795
796 /*
797 * We don't know the actual number of objects written,
798 * only how many bytes written, how many bytes total, and
799 * how many objects total. So we can fake it by pretending all
800 * objects we are writing are the same size. This gives us a
801 * smooth progress meter, and at the end it matches the true
802 * answer.
803 */
804 written = reuse_packfile_objects *
805 (((double)(total - to_write)) / total);
806 display_progress(progress_state, written);
807 }
808
809 close(fd);
810 written = reuse_packfile_objects;
811 display_progress(progress_state, written);
812 return reuse_packfile_offset - sizeof(struct pack_header);
813 }
814
815 static const char no_split_warning[] = N_(
816 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
817 );
818
819 static void write_pack_file(void)
820 {
821 uint32_t i = 0, j;
822 struct hashfile *f;
823 off_t offset;
824 uint32_t nr_remaining = nr_result;
825 time_t last_mtime = 0;
826 struct object_entry **write_order;
827
828 if (progress > pack_to_stdout)
829 progress_state = start_progress(_("Writing objects"), nr_result);
830 ALLOC_ARRAY(written_list, to_pack.nr_objects);
831 write_order = compute_write_order();
832
833 do {
834 struct object_id oid;
835 char *pack_tmp_name = NULL;
836
837 if (pack_to_stdout)
838 f = hashfd_throughput(1, "<stdout>", progress_state);
839 else
840 f = create_tmp_packfile(&pack_tmp_name);
841
842 offset = write_pack_header(f, nr_remaining);
843
844 if (reuse_packfile) {
845 off_t packfile_size;
846 assert(pack_to_stdout);
847
848 packfile_size = write_reused_pack(f);
849 offset += packfile_size;
850 }
851
852 nr_written = 0;
853 for (; i < to_pack.nr_objects; i++) {
854 struct object_entry *e = write_order[i];
855 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
856 break;
857 display_progress(progress_state, written);
858 }
859
860 /*
861 * Did we write the wrong # entries in the header?
862 * If so, rewrite it like in fast-import
863 */
864 if (pack_to_stdout) {
865 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
866 } else if (nr_written == nr_remaining) {
867 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
868 } else {
869 int fd = finalize_hashfile(f, oid.hash, 0);
870 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
871 nr_written, oid.hash, offset);
872 close(fd);
873 if (write_bitmap_index) {
874 warning(_(no_split_warning));
875 write_bitmap_index = 0;
876 }
877 }
878
879 if (!pack_to_stdout) {
880 struct stat st;
881 struct strbuf tmpname = STRBUF_INIT;
882
883 /*
884 * Packs are runtime accessed in their mtime
885 * order since newer packs are more likely to contain
886 * younger objects. So if we are creating multiple
887 * packs then we should modify the mtime of later ones
888 * to preserve this property.
889 */
890 if (stat(pack_tmp_name, &st) < 0) {
891 warning_errno(_("failed to stat %s"), pack_tmp_name);
892 } else if (!last_mtime) {
893 last_mtime = st.st_mtime;
894 } else {
895 struct utimbuf utb;
896 utb.actime = st.st_atime;
897 utb.modtime = --last_mtime;
898 if (utime(pack_tmp_name, &utb) < 0)
899 warning_errno(_("failed utime() on %s"), pack_tmp_name);
900 }
901
902 strbuf_addf(&tmpname, "%s-", base_name);
903
904 if (write_bitmap_index) {
905 bitmap_writer_set_checksum(oid.hash);
906 bitmap_writer_build_type_index(
907 &to_pack, written_list, nr_written);
908 }
909
910 finish_tmp_packfile(&tmpname, pack_tmp_name,
911 written_list, nr_written,
912 &pack_idx_opts, oid.hash);
913
914 if (write_bitmap_index) {
915 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
916
917 stop_progress(&progress_state);
918
919 bitmap_writer_show_progress(progress);
920 bitmap_writer_reuse_bitmaps(&to_pack);
921 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
922 bitmap_writer_build(&to_pack);
923 bitmap_writer_finish(written_list, nr_written,
924 tmpname.buf, write_bitmap_options);
925 write_bitmap_index = 0;
926 }
927
928 strbuf_release(&tmpname);
929 free(pack_tmp_name);
930 puts(oid_to_hex(&oid));
931 }
932
933 /* mark written objects as written to previous pack */
934 for (j = 0; j < nr_written; j++) {
935 written_list[j]->offset = (off_t)-1;
936 }
937 nr_remaining -= nr_written;
938 } while (nr_remaining && i < to_pack.nr_objects);
939
940 free(written_list);
941 free(write_order);
942 stop_progress(&progress_state);
943 if (written != nr_result)
944 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
945 written, nr_result);
946 }
947
948 static int no_try_delta(const char *path)
949 {
950 static struct attr_check *check;
951
952 if (!check)
953 check = attr_check_initl("delta", NULL);
954 git_check_attr(&the_index, path, check);
955 if (ATTR_FALSE(check->items[0].value))
956 return 1;
957 return 0;
958 }
959
960 /*
961 * When adding an object, check whether we have already added it
962 * to our packing list. If so, we can skip. However, if we are
963 * being asked to excludei t, but the previous mention was to include
964 * it, make sure to adjust its flags and tweak our numbers accordingly.
965 *
966 * As an optimization, we pass out the index position where we would have
967 * found the item, since that saves us from having to look it up again a
968 * few lines later when we want to add the new entry.
969 */
970 static int have_duplicate_entry(const struct object_id *oid,
971 int exclude,
972 uint32_t *index_pos)
973 {
974 struct object_entry *entry;
975
976 entry = packlist_find(&to_pack, oid->hash, index_pos);
977 if (!entry)
978 return 0;
979
980 if (exclude) {
981 if (!entry->preferred_base)
982 nr_result--;
983 entry->preferred_base = 1;
984 }
985
986 return 1;
987 }
988
989 static int want_found_object(int exclude, struct packed_git *p)
990 {
991 if (exclude)
992 return 1;
993 if (incremental)
994 return 0;
995
996 /*
997 * When asked to do --local (do not include an object that appears in a
998 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
999 * an object that appears in a pack marked with .keep), finding a pack
1000 * that matches the criteria is sufficient for us to decide to omit it.
1001 * However, even if this pack does not satisfy the criteria, we need to
1002 * make sure no copy of this object appears in _any_ pack that makes us
1003 * to omit the object, so we need to check all the packs.
1004 *
1005 * We can however first check whether these options can possible matter;
1006 * if they do not matter we know we want the object in generated pack.
1007 * Otherwise, we signal "-1" at the end to tell the caller that we do
1008 * not know either way, and it needs to check more packs.
1009 */
1010 if (!ignore_packed_keep_on_disk &&
1011 !ignore_packed_keep_in_core &&
1012 (!local || !have_non_local_packs))
1013 return 1;
1014
1015 if (local && !p->pack_local)
1016 return 0;
1017 if (p->pack_local &&
1018 ((ignore_packed_keep_on_disk && p->pack_keep) ||
1019 (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1020 return 0;
1021
1022 /* we don't know yet; keep looking for more packs */
1023 return -1;
1024 }
1025
1026 /*
1027 * Check whether we want the object in the pack (e.g., we do not want
1028 * objects found in non-local stores if the "--local" option was used).
1029 *
1030 * If the caller already knows an existing pack it wants to take the object
1031 * from, that is passed in *found_pack and *found_offset; otherwise this
1032 * function finds if there is any pack that has the object and returns the pack
1033 * and its offset in these variables.
1034 */
1035 static int want_object_in_pack(const struct object_id *oid,
1036 int exclude,
1037 struct packed_git **found_pack,
1038 off_t *found_offset)
1039 {
1040 int want;
1041 struct list_head *pos;
1042
1043 if (!exclude && local && has_loose_object_nonlocal(oid))
1044 return 0;
1045
1046 /*
1047 * If we already know the pack object lives in, start checks from that
1048 * pack - in the usual case when neither --local was given nor .keep files
1049 * are present we will determine the answer right now.
1050 */
1051 if (*found_pack) {
1052 want = want_found_object(exclude, *found_pack);
1053 if (want != -1)
1054 return want;
1055 }
1056 list_for_each(pos, get_packed_git_mru(the_repository)) {
1057 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1058 off_t offset;
1059
1060 if (p == *found_pack)
1061 offset = *found_offset;
1062 else
1063 offset = find_pack_entry_one(oid->hash, p);
1064
1065 if (offset) {
1066 if (!*found_pack) {
1067 if (!is_pack_valid(p))
1068 continue;
1069 *found_offset = offset;
1070 *found_pack = p;
1071 }
1072 want = want_found_object(exclude, p);
1073 if (!exclude && want > 0)
1074 list_move(&p->mru,
1075 get_packed_git_mru(the_repository));
1076 if (want != -1)
1077 return want;
1078 }
1079 }
1080
1081 return 1;
1082 }
1083
1084 static void create_object_entry(const struct object_id *oid,
1085 enum object_type type,
1086 uint32_t hash,
1087 int exclude,
1088 int no_try_delta,
1089 uint32_t index_pos,
1090 struct packed_git *found_pack,
1091 off_t found_offset)
1092 {
1093 struct object_entry *entry;
1094
1095 entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1096 entry->hash = hash;
1097 oe_set_type(entry, type);
1098 if (exclude)
1099 entry->preferred_base = 1;
1100 else
1101 nr_result++;
1102 if (found_pack) {
1103 oe_set_in_pack(&to_pack, entry, found_pack);
1104 entry->in_pack_offset = found_offset;
1105 }
1106
1107 entry->no_try_delta = no_try_delta;
1108 }
1109
1110 static const char no_closure_warning[] = N_(
1111 "disabling bitmap writing, as some objects are not being packed"
1112 );
1113
1114 static int add_object_entry(const struct object_id *oid, enum object_type type,
1115 const char *name, int exclude)
1116 {
1117 struct packed_git *found_pack = NULL;
1118 off_t found_offset = 0;
1119 uint32_t index_pos;
1120
1121 display_progress(progress_state, ++nr_seen);
1122
1123 if (have_duplicate_entry(oid, exclude, &index_pos))
1124 return 0;
1125
1126 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1127 /* The pack is missing an object, so it will not have closure */
1128 if (write_bitmap_index) {
1129 warning(_(no_closure_warning));
1130 write_bitmap_index = 0;
1131 }
1132 return 0;
1133 }
1134
1135 create_object_entry(oid, type, pack_name_hash(name),
1136 exclude, name && no_try_delta(name),
1137 index_pos, found_pack, found_offset);
1138 return 1;
1139 }
1140
1141 static int add_object_entry_from_bitmap(const struct object_id *oid,
1142 enum object_type type,
1143 int flags, uint32_t name_hash,
1144 struct packed_git *pack, off_t offset)
1145 {
1146 uint32_t index_pos;
1147
1148 display_progress(progress_state, ++nr_seen);
1149
1150 if (have_duplicate_entry(oid, 0, &index_pos))
1151 return 0;
1152
1153 if (!want_object_in_pack(oid, 0, &pack, &offset))
1154 return 0;
1155
1156 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1157 return 1;
1158 }
1159
1160 struct pbase_tree_cache {
1161 struct object_id oid;
1162 int ref;
1163 int temporary;
1164 void *tree_data;
1165 unsigned long tree_size;
1166 };
1167
1168 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1169 static int pbase_tree_cache_ix(const struct object_id *oid)
1170 {
1171 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1172 }
1173 static int pbase_tree_cache_ix_incr(int ix)
1174 {
1175 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1176 }
1177
1178 static struct pbase_tree {
1179 struct pbase_tree *next;
1180 /* This is a phony "cache" entry; we are not
1181 * going to evict it or find it through _get()
1182 * mechanism -- this is for the toplevel node that
1183 * would almost always change with any commit.
1184 */
1185 struct pbase_tree_cache pcache;
1186 } *pbase_tree;
1187
1188 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1189 {
1190 struct pbase_tree_cache *ent, *nent;
1191 void *data;
1192 unsigned long size;
1193 enum object_type type;
1194 int neigh;
1195 int my_ix = pbase_tree_cache_ix(oid);
1196 int available_ix = -1;
1197
1198 /* pbase-tree-cache acts as a limited hashtable.
1199 * your object will be found at your index or within a few
1200 * slots after that slot if it is cached.
1201 */
1202 for (neigh = 0; neigh < 8; neigh++) {
1203 ent = pbase_tree_cache[my_ix];
1204 if (ent && !oidcmp(&ent->oid, oid)) {
1205 ent->ref++;
1206 return ent;
1207 }
1208 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1209 ((0 <= available_ix) &&
1210 (!ent && pbase_tree_cache[available_ix])))
1211 available_ix = my_ix;
1212 if (!ent)
1213 break;
1214 my_ix = pbase_tree_cache_ix_incr(my_ix);
1215 }
1216
1217 /* Did not find one. Either we got a bogus request or
1218 * we need to read and perhaps cache.
1219 */
1220 data = read_object_file(oid, &type, &size);
1221 if (!data)
1222 return NULL;
1223 if (type != OBJ_TREE) {
1224 free(data);
1225 return NULL;
1226 }
1227
1228 /* We need to either cache or return a throwaway copy */
1229
1230 if (available_ix < 0)
1231 ent = NULL;
1232 else {
1233 ent = pbase_tree_cache[available_ix];
1234 my_ix = available_ix;
1235 }
1236
1237 if (!ent) {
1238 nent = xmalloc(sizeof(*nent));
1239 nent->temporary = (available_ix < 0);
1240 }
1241 else {
1242 /* evict and reuse */
1243 free(ent->tree_data);
1244 nent = ent;
1245 }
1246 oidcpy(&nent->oid, oid);
1247 nent->tree_data = data;
1248 nent->tree_size = size;
1249 nent->ref = 1;
1250 if (!nent->temporary)
1251 pbase_tree_cache[my_ix] = nent;
1252 return nent;
1253 }
1254
1255 static void pbase_tree_put(struct pbase_tree_cache *cache)
1256 {
1257 if (!cache->temporary) {
1258 cache->ref--;
1259 return;
1260 }
1261 free(cache->tree_data);
1262 free(cache);
1263 }
1264
1265 static int name_cmp_len(const char *name)
1266 {
1267 int i;
1268 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1269 ;
1270 return i;
1271 }
1272
1273 static void add_pbase_object(struct tree_desc *tree,
1274 const char *name,
1275 int cmplen,
1276 const char *fullname)
1277 {
1278 struct name_entry entry;
1279 int cmp;
1280
1281 while (tree_entry(tree,&entry)) {
1282 if (S_ISGITLINK(entry.mode))
1283 continue;
1284 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1285 memcmp(name, entry.path, cmplen);
1286 if (cmp > 0)
1287 continue;
1288 if (cmp < 0)
1289 return;
1290 if (name[cmplen] != '/') {
1291 add_object_entry(entry.oid,
1292 object_type(entry.mode),
1293 fullname, 1);
1294 return;
1295 }
1296 if (S_ISDIR(entry.mode)) {
1297 struct tree_desc sub;
1298 struct pbase_tree_cache *tree;
1299 const char *down = name+cmplen+1;
1300 int downlen = name_cmp_len(down);
1301
1302 tree = pbase_tree_get(entry.oid);
1303 if (!tree)
1304 return;
1305 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1306
1307 add_pbase_object(&sub, down, downlen, fullname);
1308 pbase_tree_put(tree);
1309 }
1310 }
1311 }
1312
1313 static unsigned *done_pbase_paths;
1314 static int done_pbase_paths_num;
1315 static int done_pbase_paths_alloc;
1316 static int done_pbase_path_pos(unsigned hash)
1317 {
1318 int lo = 0;
1319 int hi = done_pbase_paths_num;
1320 while (lo < hi) {
1321 int mi = lo + (hi - lo) / 2;
1322 if (done_pbase_paths[mi] == hash)
1323 return mi;
1324 if (done_pbase_paths[mi] < hash)
1325 hi = mi;
1326 else
1327 lo = mi + 1;
1328 }
1329 return -lo-1;
1330 }
1331
1332 static int check_pbase_path(unsigned hash)
1333 {
1334 int pos = done_pbase_path_pos(hash);
1335 if (0 <= pos)
1336 return 1;
1337 pos = -pos - 1;
1338 ALLOC_GROW(done_pbase_paths,
1339 done_pbase_paths_num + 1,
1340 done_pbase_paths_alloc);
1341 done_pbase_paths_num++;
1342 if (pos < done_pbase_paths_num)
1343 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1344 done_pbase_paths_num - pos - 1);
1345 done_pbase_paths[pos] = hash;
1346 return 0;
1347 }
1348
1349 static void add_preferred_base_object(const char *name)
1350 {
1351 struct pbase_tree *it;
1352 int cmplen;
1353 unsigned hash = pack_name_hash(name);
1354
1355 if (!num_preferred_base || check_pbase_path(hash))
1356 return;
1357
1358 cmplen = name_cmp_len(name);
1359 for (it = pbase_tree; it; it = it->next) {
1360 if (cmplen == 0) {
1361 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1362 }
1363 else {
1364 struct tree_desc tree;
1365 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1366 add_pbase_object(&tree, name, cmplen, name);
1367 }
1368 }
1369 }
1370
1371 static void add_preferred_base(struct object_id *oid)
1372 {
1373 struct pbase_tree *it;
1374 void *data;
1375 unsigned long size;
1376 struct object_id tree_oid;
1377
1378 if (window <= num_preferred_base++)
1379 return;
1380
1381 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);
1382 if (!data)
1383 return;
1384
1385 for (it = pbase_tree; it; it = it->next) {
1386 if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1387 free(data);
1388 return;
1389 }
1390 }
1391
1392 it = xcalloc(1, sizeof(*it));
1393 it->next = pbase_tree;
1394 pbase_tree = it;
1395
1396 oidcpy(&it->pcache.oid, &tree_oid);
1397 it->pcache.tree_data = data;
1398 it->pcache.tree_size = size;
1399 }
1400
1401 static void cleanup_preferred_base(void)
1402 {
1403 struct pbase_tree *it;
1404 unsigned i;
1405
1406 it = pbase_tree;
1407 pbase_tree = NULL;
1408 while (it) {
1409 struct pbase_tree *tmp = it;
1410 it = tmp->next;
1411 free(tmp->pcache.tree_data);
1412 free(tmp);
1413 }
1414
1415 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1416 if (!pbase_tree_cache[i])
1417 continue;
1418 free(pbase_tree_cache[i]->tree_data);
1419 FREE_AND_NULL(pbase_tree_cache[i]);
1420 }
1421
1422 FREE_AND_NULL(done_pbase_paths);
1423 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1424 }
1425
1426 static void check_object(struct object_entry *entry)
1427 {
1428 unsigned long canonical_size;
1429
1430 if (IN_PACK(entry)) {
1431 struct packed_git *p = IN_PACK(entry);
1432 struct pack_window *w_curs = NULL;
1433 const unsigned char *base_ref = NULL;
1434 struct object_entry *base_entry;
1435 unsigned long used, used_0;
1436 unsigned long avail;
1437 off_t ofs;
1438 unsigned char *buf, c;
1439 enum object_type type;
1440 unsigned long in_pack_size;
1441
1442 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1443
1444 /*
1445 * We want in_pack_type even if we do not reuse delta
1446 * since non-delta representations could still be reused.
1447 */
1448 used = unpack_object_header_buffer(buf, avail,
1449 &type,
1450 &in_pack_size);
1451 if (used == 0)
1452 goto give_up;
1453
1454 if (type < 0)
1455 BUG("invalid type %d", type);
1456 entry->in_pack_type = type;
1457
1458 /*
1459 * Determine if this is a delta and if so whether we can
1460 * reuse it or not. Otherwise let's find out as cheaply as
1461 * possible what the actual type and size for this object is.
1462 */
1463 switch (entry->in_pack_type) {
1464 default:
1465 /* Not a delta hence we've already got all we need. */
1466 oe_set_type(entry, entry->in_pack_type);
1467 SET_SIZE(entry, in_pack_size);
1468 entry->in_pack_header_size = used;
1469 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1470 goto give_up;
1471 unuse_pack(&w_curs);
1472 return;
1473 case OBJ_REF_DELTA:
1474 if (reuse_delta && !entry->preferred_base)
1475 base_ref = use_pack(p, &w_curs,
1476 entry->in_pack_offset + used, NULL);
1477 entry->in_pack_header_size = used + the_hash_algo->rawsz;
1478 break;
1479 case OBJ_OFS_DELTA:
1480 buf = use_pack(p, &w_curs,
1481 entry->in_pack_offset + used, NULL);
1482 used_0 = 0;
1483 c = buf[used_0++];
1484 ofs = c & 127;
1485 while (c & 128) {
1486 ofs += 1;
1487 if (!ofs || MSB(ofs, 7)) {
1488 error(_("delta base offset overflow in pack for %s"),
1489 oid_to_hex(&entry->idx.oid));
1490 goto give_up;
1491 }
1492 c = buf[used_0++];
1493 ofs = (ofs << 7) + (c & 127);
1494 }
1495 ofs = entry->in_pack_offset - ofs;
1496 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1497 error(_("delta base offset out of bound for %s"),
1498 oid_to_hex(&entry->idx.oid));
1499 goto give_up;
1500 }
1501 if (reuse_delta && !entry->preferred_base) {
1502 struct revindex_entry *revidx;
1503 revidx = find_pack_revindex(p, ofs);
1504 if (!revidx)
1505 goto give_up;
1506 base_ref = nth_packed_object_sha1(p, revidx->nr);
1507 }
1508 entry->in_pack_header_size = used + used_0;
1509 break;
1510 }
1511
1512 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1513 /*
1514 * If base_ref was set above that means we wish to
1515 * reuse delta data, and we even found that base
1516 * in the list of objects we want to pack. Goodie!
1517 *
1518 * Depth value does not matter - find_deltas() will
1519 * never consider reused delta as the base object to
1520 * deltify other objects against, in order to avoid
1521 * circular deltas.
1522 */
1523 oe_set_type(entry, entry->in_pack_type);
1524 SET_SIZE(entry, in_pack_size); /* delta size */
1525 SET_DELTA(entry, base_entry);
1526 SET_DELTA_SIZE(entry, in_pack_size);
1527 entry->delta_sibling_idx = base_entry->delta_child_idx;
1528 SET_DELTA_CHILD(base_entry, entry);
1529 unuse_pack(&w_curs);
1530 return;
1531 }
1532
1533 if (oe_type(entry)) {
1534 off_t delta_pos;
1535
1536 /*
1537 * This must be a delta and we already know what the
1538 * final object type is. Let's extract the actual
1539 * object size from the delta header.
1540 */
1541 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1542 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1543 if (canonical_size == 0)
1544 goto give_up;
1545 SET_SIZE(entry, canonical_size);
1546 unuse_pack(&w_curs);
1547 return;
1548 }
1549
1550 /*
1551 * No choice but to fall back to the recursive delta walk
1552 * with sha1_object_info() to find about the object type
1553 * at this point...
1554 */
1555 give_up:
1556 unuse_pack(&w_curs);
1557 }
1558
1559 oe_set_type(entry,
1560 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1561 if (entry->type_valid) {
1562 SET_SIZE(entry, canonical_size);
1563 } else {
1564 /*
1565 * Bad object type is checked in prepare_pack(). This is
1566 * to permit a missing preferred base object to be ignored
1567 * as a preferred base. Doing so can result in a larger
1568 * pack file, but the transfer will still take place.
1569 */
1570 }
1571 }
1572
1573 static int pack_offset_sort(const void *_a, const void *_b)
1574 {
1575 const struct object_entry *a = *(struct object_entry **)_a;
1576 const struct object_entry *b = *(struct object_entry **)_b;
1577 const struct packed_git *a_in_pack = IN_PACK(a);
1578 const struct packed_git *b_in_pack = IN_PACK(b);
1579
1580 /* avoid filesystem trashing with loose objects */
1581 if (!a_in_pack && !b_in_pack)
1582 return oidcmp(&a->idx.oid, &b->idx.oid);
1583
1584 if (a_in_pack < b_in_pack)
1585 return -1;
1586 if (a_in_pack > b_in_pack)
1587 return 1;
1588 return a->in_pack_offset < b->in_pack_offset ? -1 :
1589 (a->in_pack_offset > b->in_pack_offset);
1590 }
1591
1592 /*
1593 * Drop an on-disk delta we were planning to reuse. Naively, this would
1594 * just involve blanking out the "delta" field, but we have to deal
1595 * with some extra book-keeping:
1596 *
1597 * 1. Removing ourselves from the delta_sibling linked list.
1598 *
1599 * 2. Updating our size/type to the non-delta representation. These were
1600 * either not recorded initially (size) or overwritten with the delta type
1601 * (type) when check_object() decided to reuse the delta.
1602 *
1603 * 3. Resetting our delta depth, as we are now a base object.
1604 */
1605 static void drop_reused_delta(struct object_entry *entry)
1606 {
1607 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1608 struct object_info oi = OBJECT_INFO_INIT;
1609 enum object_type type;
1610 unsigned long size;
1611
1612 while (*idx) {
1613 struct object_entry *oe = &to_pack.objects[*idx - 1];
1614
1615 if (oe == entry)
1616 *idx = oe->delta_sibling_idx;
1617 else
1618 idx = &oe->delta_sibling_idx;
1619 }
1620 SET_DELTA(entry, NULL);
1621 entry->depth = 0;
1622
1623 oi.sizep = &size;
1624 oi.typep = &type;
1625 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1626 /*
1627 * We failed to get the info from this pack for some reason;
1628 * fall back to sha1_object_info, which may find another copy.
1629 * And if that fails, the error will be recorded in oe_type(entry)
1630 * and dealt with in prepare_pack().
1631 */
1632 oe_set_type(entry,
1633 oid_object_info(the_repository, &entry->idx.oid, &size));
1634 } else {
1635 oe_set_type(entry, type);
1636 }
1637 SET_SIZE(entry, size);
1638 }
1639
1640 /*
1641 * Follow the chain of deltas from this entry onward, throwing away any links
1642 * that cause us to hit a cycle (as determined by the DFS state flags in
1643 * the entries).
1644 *
1645 * We also detect too-long reused chains that would violate our --depth
1646 * limit.
1647 */
1648 static void break_delta_chains(struct object_entry *entry)
1649 {
1650 /*
1651 * The actual depth of each object we will write is stored as an int,
1652 * as it cannot exceed our int "depth" limit. But before we break
1653 * changes based no that limit, we may potentially go as deep as the
1654 * number of objects, which is elsewhere bounded to a uint32_t.
1655 */
1656 uint32_t total_depth;
1657 struct object_entry *cur, *next;
1658
1659 for (cur = entry, total_depth = 0;
1660 cur;
1661 cur = DELTA(cur), total_depth++) {
1662 if (cur->dfs_state == DFS_DONE) {
1663 /*
1664 * We've already seen this object and know it isn't
1665 * part of a cycle. We do need to append its depth
1666 * to our count.
1667 */
1668 total_depth += cur->depth;
1669 break;
1670 }
1671
1672 /*
1673 * We break cycles before looping, so an ACTIVE state (or any
1674 * other cruft which made its way into the state variable)
1675 * is a bug.
1676 */
1677 if (cur->dfs_state != DFS_NONE)
1678 BUG("confusing delta dfs state in first pass: %d",
1679 cur->dfs_state);
1680
1681 /*
1682 * Now we know this is the first time we've seen the object. If
1683 * it's not a delta, we're done traversing, but we'll mark it
1684 * done to save time on future traversals.
1685 */
1686 if (!DELTA(cur)) {
1687 cur->dfs_state = DFS_DONE;
1688 break;
1689 }
1690
1691 /*
1692 * Mark ourselves as active and see if the next step causes
1693 * us to cycle to another active object. It's important to do
1694 * this _before_ we loop, because it impacts where we make the
1695 * cut, and thus how our total_depth counter works.
1696 * E.g., We may see a partial loop like:
1697 *
1698 * A -> B -> C -> D -> B
1699 *
1700 * Cutting B->C breaks the cycle. But now the depth of A is
1701 * only 1, and our total_depth counter is at 3. The size of the
1702 * error is always one less than the size of the cycle we
1703 * broke. Commits C and D were "lost" from A's chain.
1704 *
1705 * If we instead cut D->B, then the depth of A is correct at 3.
1706 * We keep all commits in the chain that we examined.
1707 */
1708 cur->dfs_state = DFS_ACTIVE;
1709 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1710 drop_reused_delta(cur);
1711 cur->dfs_state = DFS_DONE;
1712 break;
1713 }
1714 }
1715
1716 /*
1717 * And now that we've gone all the way to the bottom of the chain, we
1718 * need to clear the active flags and set the depth fields as
1719 * appropriate. Unlike the loop above, which can quit when it drops a
1720 * delta, we need to keep going to look for more depth cuts. So we need
1721 * an extra "next" pointer to keep going after we reset cur->delta.
1722 */
1723 for (cur = entry; cur; cur = next) {
1724 next = DELTA(cur);
1725
1726 /*
1727 * We should have a chain of zero or more ACTIVE states down to
1728 * a final DONE. We can quit after the DONE, because either it
1729 * has no bases, or we've already handled them in a previous
1730 * call.
1731 */
1732 if (cur->dfs_state == DFS_DONE)
1733 break;
1734 else if (cur->dfs_state != DFS_ACTIVE)
1735 BUG("confusing delta dfs state in second pass: %d",
1736 cur->dfs_state);
1737
1738 /*
1739 * If the total_depth is more than depth, then we need to snip
1740 * the chain into two or more smaller chains that don't exceed
1741 * the maximum depth. Most of the resulting chains will contain
1742 * (depth + 1) entries (i.e., depth deltas plus one base), and
1743 * the last chain (i.e., the one containing entry) will contain
1744 * whatever entries are left over, namely
1745 * (total_depth % (depth + 1)) of them.
1746 *
1747 * Since we are iterating towards decreasing depth, we need to
1748 * decrement total_depth as we go, and we need to write to the
1749 * entry what its final depth will be after all of the
1750 * snipping. Since we're snipping into chains of length (depth
1751 * + 1) entries, the final depth of an entry will be its
1752 * original depth modulo (depth + 1). Any time we encounter an
1753 * entry whose final depth is supposed to be zero, we snip it
1754 * from its delta base, thereby making it so.
1755 */
1756 cur->depth = (total_depth--) % (depth + 1);
1757 if (!cur->depth)
1758 drop_reused_delta(cur);
1759
1760 cur->dfs_state = DFS_DONE;
1761 }
1762 }
1763
1764 static void get_object_details(void)
1765 {
1766 uint32_t i;
1767 struct object_entry **sorted_by_offset;
1768
1769 if (progress)
1770 progress_state = start_progress(_("Counting objects"),
1771 to_pack.nr_objects);
1772
1773 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1774 for (i = 0; i < to_pack.nr_objects; i++)
1775 sorted_by_offset[i] = to_pack.objects + i;
1776 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1777
1778 for (i = 0; i < to_pack.nr_objects; i++) {
1779 struct object_entry *entry = sorted_by_offset[i];
1780 check_object(entry);
1781 if (entry->type_valid &&
1782 oe_size_greater_than(&to_pack, entry, big_file_threshold))
1783 entry->no_try_delta = 1;
1784 display_progress(progress_state, i + 1);
1785 }
1786 stop_progress(&progress_state);
1787
1788 /*
1789 * This must happen in a second pass, since we rely on the delta
1790 * information for the whole list being completed.
1791 */
1792 for (i = 0; i < to_pack.nr_objects; i++)
1793 break_delta_chains(&to_pack.objects[i]);
1794
1795 free(sorted_by_offset);
1796 }
1797
1798 /*
1799 * We search for deltas in a list sorted by type, by filename hash, and then
1800 * by size, so that we see progressively smaller and smaller files.
1801 * That's because we prefer deltas to be from the bigger file
1802 * to the smaller -- deletes are potentially cheaper, but perhaps
1803 * more importantly, the bigger file is likely the more recent
1804 * one. The deepest deltas are therefore the oldest objects which are
1805 * less susceptible to be accessed often.
1806 */
1807 static int type_size_sort(const void *_a, const void *_b)
1808 {
1809 const struct object_entry *a = *(struct object_entry **)_a;
1810 const struct object_entry *b = *(struct object_entry **)_b;
1811 enum object_type a_type = oe_type(a);
1812 enum object_type b_type = oe_type(b);
1813 unsigned long a_size = SIZE(a);
1814 unsigned long b_size = SIZE(b);
1815
1816 if (a_type > b_type)
1817 return -1;
1818 if (a_type < b_type)
1819 return 1;
1820 if (a->hash > b->hash)
1821 return -1;
1822 if (a->hash < b->hash)
1823 return 1;
1824 if (a->preferred_base > b->preferred_base)
1825 return -1;
1826 if (a->preferred_base < b->preferred_base)
1827 return 1;
1828 if (a_size > b_size)
1829 return -1;
1830 if (a_size < b_size)
1831 return 1;
1832 return a < b ? -1 : (a > b); /* newest first */
1833 }
1834
1835 struct unpacked {
1836 struct object_entry *entry;
1837 void *data;
1838 struct delta_index *index;
1839 unsigned depth;
1840 };
1841
1842 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1843 unsigned long delta_size)
1844 {
1845 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1846 return 0;
1847
1848 if (delta_size < cache_max_small_delta_size)
1849 return 1;
1850
1851 /* cache delta, if objects are large enough compared to delta size */
1852 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1853 return 1;
1854
1855 return 0;
1856 }
1857
1858 #ifndef NO_PTHREADS
1859
1860 /* Protect access to object database */
1861 static pthread_mutex_t read_mutex;
1862 #define read_lock() pthread_mutex_lock(&read_mutex)
1863 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1864
1865 /* Protect delta_cache_size */
1866 static pthread_mutex_t cache_mutex;
1867 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1868 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1869
1870 /*
1871 * Protect object list partitioning (e.g. struct thread_param) and
1872 * progress_state
1873 */
1874 static pthread_mutex_t progress_mutex;
1875 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1876 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1877
1878 /*
1879 * Access to struct object_entry is unprotected since each thread owns
1880 * a portion of the main object list. Just don't access object entries
1881 * ahead in the list because they can be stolen and would need
1882 * progress_mutex for protection.
1883 */
1884 #else
1885
1886 #define read_lock() (void)0
1887 #define read_unlock() (void)0
1888 #define cache_lock() (void)0
1889 #define cache_unlock() (void)0
1890 #define progress_lock() (void)0
1891 #define progress_unlock() (void)0
1892
1893 #endif
1894
1895 /*
1896 * Return the size of the object without doing any delta
1897 * reconstruction (so non-deltas are true object sizes, but deltas
1898 * return the size of the delta data).
1899 */
1900 unsigned long oe_get_size_slow(struct packing_data *pack,
1901 const struct object_entry *e)
1902 {
1903 struct packed_git *p;
1904 struct pack_window *w_curs;
1905 unsigned char *buf;
1906 enum object_type type;
1907 unsigned long used, avail, size;
1908
1909 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
1910 read_lock();
1911 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
1912 die(_("unable to get size of %s"),
1913 oid_to_hex(&e->idx.oid));
1914 read_unlock();
1915 return size;
1916 }
1917
1918 p = oe_in_pack(pack, e);
1919 if (!p)
1920 BUG("when e->type is a delta, it must belong to a pack");
1921
1922 read_lock();
1923 w_curs = NULL;
1924 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
1925 used = unpack_object_header_buffer(buf, avail, &type, &size);
1926 if (used == 0)
1927 die(_("unable to parse object header of %s"),
1928 oid_to_hex(&e->idx.oid));
1929
1930 unuse_pack(&w_curs);
1931 read_unlock();
1932 return size;
1933 }
1934
1935 static int try_delta(struct unpacked *trg, struct unpacked *src,
1936 unsigned max_depth, unsigned long *mem_usage)
1937 {
1938 struct object_entry *trg_entry = trg->entry;
1939 struct object_entry *src_entry = src->entry;
1940 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1941 unsigned ref_depth;
1942 enum object_type type;
1943 void *delta_buf;
1944
1945 /* Don't bother doing diffs between different types */
1946 if (oe_type(trg_entry) != oe_type(src_entry))
1947 return -1;
1948
1949 /*
1950 * We do not bother to try a delta that we discarded on an
1951 * earlier try, but only when reusing delta data. Note that
1952 * src_entry that is marked as the preferred_base should always
1953 * be considered, as even if we produce a suboptimal delta against
1954 * it, we will still save the transfer cost, as we already know
1955 * the other side has it and we won't send src_entry at all.
1956 */
1957 if (reuse_delta && IN_PACK(trg_entry) &&
1958 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
1959 !src_entry->preferred_base &&
1960 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1961 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1962 return 0;
1963
1964 /* Let's not bust the allowed depth. */
1965 if (src->depth >= max_depth)
1966 return 0;
1967
1968 /* Now some size filtering heuristics. */
1969 trg_size = SIZE(trg_entry);
1970 if (!DELTA(trg_entry)) {
1971 max_size = trg_size/2 - the_hash_algo->rawsz;
1972 ref_depth = 1;
1973 } else {
1974 max_size = DELTA_SIZE(trg_entry);
1975 ref_depth = trg->depth;
1976 }
1977 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1978 (max_depth - ref_depth + 1);
1979 if (max_size == 0)
1980 return 0;
1981 src_size = SIZE(src_entry);
1982 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1983 if (sizediff >= max_size)
1984 return 0;
1985 if (trg_size < src_size / 32)
1986 return 0;
1987
1988 /* Load data if not already done */
1989 if (!trg->data) {
1990 read_lock();
1991 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
1992 read_unlock();
1993 if (!trg->data)
1994 die(_("object %s cannot be read"),
1995 oid_to_hex(&trg_entry->idx.oid));
1996 if (sz != trg_size)
1997 die(_("object %s inconsistent object length (%lu vs %lu)"),
1998 oid_to_hex(&trg_entry->idx.oid), sz,
1999 trg_size);
2000 *mem_usage += sz;
2001 }
2002 if (!src->data) {
2003 read_lock();
2004 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
2005 read_unlock();
2006 if (!src->data) {
2007 if (src_entry->preferred_base) {
2008 static int warned = 0;
2009 if (!warned++)
2010 warning(_("object %s cannot be read"),
2011 oid_to_hex(&src_entry->idx.oid));
2012 /*
2013 * Those objects are not included in the
2014 * resulting pack. Be resilient and ignore
2015 * them if they can't be read, in case the
2016 * pack could be created nevertheless.
2017 */
2018 return 0;
2019 }
2020 die(_("object %s cannot be read"),
2021 oid_to_hex(&src_entry->idx.oid));
2022 }
2023 if (sz != src_size)
2024 die(_("object %s inconsistent object length (%lu vs %lu)"),
2025 oid_to_hex(&src_entry->idx.oid), sz,
2026 src_size);
2027 *mem_usage += sz;
2028 }
2029 if (!src->index) {
2030 src->index = create_delta_index(src->data, src_size);
2031 if (!src->index) {
2032 static int warned = 0;
2033 if (!warned++)
2034 warning(_("suboptimal pack - out of memory"));
2035 return 0;
2036 }
2037 *mem_usage += sizeof_delta_index(src->index);
2038 }
2039
2040 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2041 if (!delta_buf)
2042 return 0;
2043
2044 if (DELTA(trg_entry)) {
2045 /* Prefer only shallower same-sized deltas. */
2046 if (delta_size == DELTA_SIZE(trg_entry) &&
2047 src->depth + 1 >= trg->depth) {
2048 free(delta_buf);
2049 return 0;
2050 }
2051 }
2052
2053 /*
2054 * Handle memory allocation outside of the cache
2055 * accounting lock. Compiler will optimize the strangeness
2056 * away when NO_PTHREADS is defined.
2057 */
2058 free(trg_entry->delta_data);
2059 cache_lock();
2060 if (trg_entry->delta_data) {
2061 delta_cache_size -= DELTA_SIZE(trg_entry);
2062 trg_entry->delta_data = NULL;
2063 }
2064 if (delta_cacheable(src_size, trg_size, delta_size)) {
2065 delta_cache_size += delta_size;
2066 cache_unlock();
2067 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2068 } else {
2069 cache_unlock();
2070 free(delta_buf);
2071 }
2072
2073 SET_DELTA(trg_entry, src_entry);
2074 SET_DELTA_SIZE(trg_entry, delta_size);
2075 trg->depth = src->depth + 1;
2076
2077 return 1;
2078 }
2079
2080 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2081 {
2082 struct object_entry *child = DELTA_CHILD(me);
2083 unsigned int m = n;
2084 while (child) {
2085 unsigned int c = check_delta_limit(child, n + 1);
2086 if (m < c)
2087 m = c;
2088 child = DELTA_SIBLING(child);
2089 }
2090 return m;
2091 }
2092
2093 static unsigned long free_unpacked(struct unpacked *n)
2094 {
2095 unsigned long freed_mem = sizeof_delta_index(n->index);
2096 free_delta_index(n->index);
2097 n->index = NULL;
2098 if (n->data) {
2099 freed_mem += SIZE(n->entry);
2100 FREE_AND_NULL(n->data);
2101 }
2102 n->entry = NULL;
2103 n->depth = 0;
2104 return freed_mem;
2105 }
2106
2107 static void find_deltas(struct object_entry **list, unsigned *list_size,
2108 int window, int depth, unsigned *processed)
2109 {
2110 uint32_t i, idx = 0, count = 0;
2111 struct unpacked *array;
2112 unsigned long mem_usage = 0;
2113
2114 array = xcalloc(window, sizeof(struct unpacked));
2115
2116 for (;;) {
2117 struct object_entry *entry;
2118 struct unpacked *n = array + idx;
2119 int j, max_depth, best_base = -1;
2120
2121 progress_lock();
2122 if (!*list_size) {
2123 progress_unlock();
2124 break;
2125 }
2126 entry = *list++;
2127 (*list_size)--;
2128 if (!entry->preferred_base) {
2129 (*processed)++;
2130 display_progress(progress_state, *processed);
2131 }
2132 progress_unlock();
2133
2134 mem_usage -= free_unpacked(n);
2135 n->entry = entry;
2136
2137 while (window_memory_limit &&
2138 mem_usage > window_memory_limit &&
2139 count > 1) {
2140 uint32_t tail = (idx + window - count) % window;
2141 mem_usage -= free_unpacked(array + tail);
2142 count--;
2143 }
2144
2145 /* We do not compute delta to *create* objects we are not
2146 * going to pack.
2147 */
2148 if (entry->preferred_base)
2149 goto next;
2150
2151 /*
2152 * If the current object is at pack edge, take the depth the
2153 * objects that depend on the current object into account
2154 * otherwise they would become too deep.
2155 */
2156 max_depth = depth;
2157 if (DELTA_CHILD(entry)) {
2158 max_depth -= check_delta_limit(entry, 0);
2159 if (max_depth <= 0)
2160 goto next;
2161 }
2162
2163 j = window;
2164 while (--j > 0) {
2165 int ret;
2166 uint32_t other_idx = idx + j;
2167 struct unpacked *m;
2168 if (other_idx >= window)
2169 other_idx -= window;
2170 m = array + other_idx;
2171 if (!m->entry)
2172 break;
2173 ret = try_delta(n, m, max_depth, &mem_usage);
2174 if (ret < 0)
2175 break;
2176 else if (ret > 0)
2177 best_base = other_idx;
2178 }
2179
2180 /*
2181 * If we decided to cache the delta data, then it is best
2182 * to compress it right away. First because we have to do
2183 * it anyway, and doing it here while we're threaded will
2184 * save a lot of time in the non threaded write phase,
2185 * as well as allow for caching more deltas within
2186 * the same cache size limit.
2187 * ...
2188 * But only if not writing to stdout, since in that case
2189 * the network is most likely throttling writes anyway,
2190 * and therefore it is best to go to the write phase ASAP
2191 * instead, as we can afford spending more time compressing
2192 * between writes at that moment.
2193 */
2194 if (entry->delta_data && !pack_to_stdout) {
2195 unsigned long size;
2196
2197 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2198 if (size < (1U << OE_Z_DELTA_BITS)) {
2199 entry->z_delta_size = size;
2200 cache_lock();
2201 delta_cache_size -= DELTA_SIZE(entry);
2202 delta_cache_size += entry->z_delta_size;
2203 cache_unlock();
2204 } else {
2205 FREE_AND_NULL(entry->delta_data);
2206 entry->z_delta_size = 0;
2207 }
2208 }
2209
2210 /* if we made n a delta, and if n is already at max
2211 * depth, leaving it in the window is pointless. we
2212 * should evict it first.
2213 */
2214 if (DELTA(entry) && max_depth <= n->depth)
2215 continue;
2216
2217 /*
2218 * Move the best delta base up in the window, after the
2219 * currently deltified object, to keep it longer. It will
2220 * be the first base object to be attempted next.
2221 */
2222 if (DELTA(entry)) {
2223 struct unpacked swap = array[best_base];
2224 int dist = (window + idx - best_base) % window;
2225 int dst = best_base;
2226 while (dist--) {
2227 int src = (dst + 1) % window;
2228 array[dst] = array[src];
2229 dst = src;
2230 }
2231 array[dst] = swap;
2232 }
2233
2234 next:
2235 idx++;
2236 if (count + 1 < window)
2237 count++;
2238 if (idx >= window)
2239 idx = 0;
2240 }
2241
2242 for (i = 0; i < window; ++i) {
2243 free_delta_index(array[i].index);
2244 free(array[i].data);
2245 }
2246 free(array);
2247 }
2248
2249 #ifndef NO_PTHREADS
2250
2251 static void try_to_free_from_threads(size_t size)
2252 {
2253 read_lock();
2254 release_pack_memory(size);
2255 read_unlock();
2256 }
2257
2258 static try_to_free_t old_try_to_free_routine;
2259
2260 /*
2261 * The main object list is split into smaller lists, each is handed to
2262 * one worker.
2263 *
2264 * The main thread waits on the condition that (at least) one of the workers
2265 * has stopped working (which is indicated in the .working member of
2266 * struct thread_params).
2267 *
2268 * When a work thread has completed its work, it sets .working to 0 and
2269 * signals the main thread and waits on the condition that .data_ready
2270 * becomes 1.
2271 *
2272 * The main thread steals half of the work from the worker that has
2273 * most work left to hand it to the idle worker.
2274 */
2275
2276 struct thread_params {
2277 pthread_t thread;
2278 struct object_entry **list;
2279 unsigned list_size;
2280 unsigned remaining;
2281 int window;
2282 int depth;
2283 int working;
2284 int data_ready;
2285 pthread_mutex_t mutex;
2286 pthread_cond_t cond;
2287 unsigned *processed;
2288 };
2289
2290 static pthread_cond_t progress_cond;
2291
2292 /*
2293 * Mutex and conditional variable can't be statically-initialized on Windows.
2294 */
2295 static void init_threaded_search(void)
2296 {
2297 init_recursive_mutex(&read_mutex);
2298 pthread_mutex_init(&cache_mutex, NULL);
2299 pthread_mutex_init(&progress_mutex, NULL);
2300 pthread_cond_init(&progress_cond, NULL);
2301 pthread_mutex_init(&to_pack.lock, NULL);
2302 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2303 }
2304
2305 static void cleanup_threaded_search(void)
2306 {
2307 set_try_to_free_routine(old_try_to_free_routine);
2308 pthread_cond_destroy(&progress_cond);
2309 pthread_mutex_destroy(&read_mutex);
2310 pthread_mutex_destroy(&cache_mutex);
2311 pthread_mutex_destroy(&progress_mutex);
2312 }
2313
2314 static void *threaded_find_deltas(void *arg)
2315 {
2316 struct thread_params *me = arg;
2317
2318 progress_lock();
2319 while (me->remaining) {
2320 progress_unlock();
2321
2322 find_deltas(me->list, &me->remaining,
2323 me->window, me->depth, me->processed);
2324
2325 progress_lock();
2326 me->working = 0;
2327 pthread_cond_signal(&progress_cond);
2328 progress_unlock();
2329
2330 /*
2331 * We must not set ->data_ready before we wait on the
2332 * condition because the main thread may have set it to 1
2333 * before we get here. In order to be sure that new
2334 * work is available if we see 1 in ->data_ready, it
2335 * was initialized to 0 before this thread was spawned
2336 * and we reset it to 0 right away.
2337 */
2338 pthread_mutex_lock(&me->mutex);
2339 while (!me->data_ready)
2340 pthread_cond_wait(&me->cond, &me->mutex);
2341 me->data_ready = 0;
2342 pthread_mutex_unlock(&me->mutex);
2343
2344 progress_lock();
2345 }
2346 progress_unlock();
2347 /* leave ->working 1 so that this doesn't get more work assigned */
2348 return NULL;
2349 }
2350
2351 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2352 int window, int depth, unsigned *processed)
2353 {
2354 struct thread_params *p;
2355 int i, ret, active_threads = 0;
2356
2357 init_threaded_search();
2358
2359 if (delta_search_threads <= 1) {
2360 find_deltas(list, &list_size, window, depth, processed);
2361 cleanup_threaded_search();
2362 return;
2363 }
2364 if (progress > pack_to_stdout)
2365 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
2366 delta_search_threads);
2367 p = xcalloc(delta_search_threads, sizeof(*p));
2368
2369 /* Partition the work amongst work threads. */
2370 for (i = 0; i < delta_search_threads; i++) {
2371 unsigned sub_size = list_size / (delta_search_threads - i);
2372
2373 /* don't use too small segments or no deltas will be found */
2374 if (sub_size < 2*window && i+1 < delta_search_threads)
2375 sub_size = 0;
2376
2377 p[i].window = window;
2378 p[i].depth = depth;
2379 p[i].processed = processed;
2380 p[i].working = 1;
2381 p[i].data_ready = 0;
2382
2383 /* try to split chunks on "path" boundaries */
2384 while (sub_size && sub_size < list_size &&
2385 list[sub_size]->hash &&
2386 list[sub_size]->hash == list[sub_size-1]->hash)
2387 sub_size++;
2388
2389 p[i].list = list;
2390 p[i].list_size = sub_size;
2391 p[i].remaining = sub_size;
2392
2393 list += sub_size;
2394 list_size -= sub_size;
2395 }
2396
2397 /* Start work threads. */
2398 for (i = 0; i < delta_search_threads; i++) {
2399 if (!p[i].list_size)
2400 continue;
2401 pthread_mutex_init(&p[i].mutex, NULL);
2402 pthread_cond_init(&p[i].cond, NULL);
2403 ret = pthread_create(&p[i].thread, NULL,
2404 threaded_find_deltas, &p[i]);
2405 if (ret)
2406 die(_("unable to create thread: %s"), strerror(ret));
2407 active_threads++;
2408 }
2409
2410 /*
2411 * Now let's wait for work completion. Each time a thread is done
2412 * with its work, we steal half of the remaining work from the
2413 * thread with the largest number of unprocessed objects and give
2414 * it to that newly idle thread. This ensure good load balancing
2415 * until the remaining object list segments are simply too short
2416 * to be worth splitting anymore.
2417 */
2418 while (active_threads) {
2419 struct thread_params *target = NULL;
2420 struct thread_params *victim = NULL;
2421 unsigned sub_size = 0;
2422
2423 progress_lock();
2424 for (;;) {
2425 for (i = 0; !target && i < delta_search_threads; i++)
2426 if (!p[i].working)
2427 target = &p[i];
2428 if (target)
2429 break;
2430 pthread_cond_wait(&progress_cond, &progress_mutex);
2431 }
2432
2433 for (i = 0; i < delta_search_threads; i++)
2434 if (p[i].remaining > 2*window &&
2435 (!victim || victim->remaining < p[i].remaining))
2436 victim = &p[i];
2437 if (victim) {
2438 sub_size = victim->remaining / 2;
2439 list = victim->list + victim->list_size - sub_size;
2440 while (sub_size && list[0]->hash &&
2441 list[0]->hash == list[-1]->hash) {
2442 list++;
2443 sub_size--;
2444 }
2445 if (!sub_size) {
2446 /*
2447 * It is possible for some "paths" to have
2448 * so many objects that no hash boundary
2449 * might be found. Let's just steal the
2450 * exact half in that case.
2451 */
2452 sub_size = victim->remaining / 2;
2453 list -= sub_size;
2454 }
2455 target->list = list;
2456 victim->list_size -= sub_size;
2457 victim->remaining -= sub_size;
2458 }
2459 target->list_size = sub_size;
2460 target->remaining = sub_size;
2461 target->working = 1;
2462 progress_unlock();
2463
2464 pthread_mutex_lock(&target->mutex);
2465 target->data_ready = 1;
2466 pthread_cond_signal(&target->cond);
2467 pthread_mutex_unlock(&target->mutex);
2468
2469 if (!sub_size) {
2470 pthread_join(target->thread, NULL);
2471 pthread_cond_destroy(&target->cond);
2472 pthread_mutex_destroy(&target->mutex);
2473 active_threads--;
2474 }
2475 }
2476 cleanup_threaded_search();
2477 free(p);
2478 }
2479
2480 #else
2481 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2482 #endif
2483
2484 static void add_tag_chain(const struct object_id *oid)
2485 {
2486 struct tag *tag;
2487
2488 /*
2489 * We catch duplicates already in add_object_entry(), but we'd
2490 * prefer to do this extra check to avoid having to parse the
2491 * tag at all if we already know that it's being packed (e.g., if
2492 * it was included via bitmaps, we would not have parsed it
2493 * previously).
2494 */
2495 if (packlist_find(&to_pack, oid->hash, NULL))
2496 return;
2497
2498 tag = lookup_tag(the_repository, oid);
2499 while (1) {
2500 if (!tag || parse_tag(tag) || !tag->tagged)
2501 die(_("unable to pack objects reachable from tag %s"),
2502 oid_to_hex(oid));
2503
2504 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2505
2506 if (tag->tagged->type != OBJ_TAG)
2507 return;
2508
2509 tag = (struct tag *)tag->tagged;
2510 }
2511 }
2512
2513 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2514 {
2515 struct object_id peeled;
2516
2517 if (starts_with(path, "refs/tags/") && /* is a tag? */
2518 !peel_ref(path, &peeled) && /* peelable? */
2519 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2520 add_tag_chain(oid);
2521 return 0;
2522 }
2523
2524 static void prepare_pack(int window, int depth)
2525 {
2526 struct object_entry **delta_list;
2527 uint32_t i, nr_deltas;
2528 unsigned n;
2529
2530 get_object_details();
2531
2532 /*
2533 * If we're locally repacking then we need to be doubly careful
2534 * from now on in order to make sure no stealth corruption gets
2535 * propagated to the new pack. Clients receiving streamed packs
2536 * should validate everything they get anyway so no need to incur
2537 * the additional cost here in that case.
2538 */
2539 if (!pack_to_stdout)
2540 do_check_packed_object_crc = 1;
2541
2542 if (!to_pack.nr_objects || !window || !depth)
2543 return;
2544
2545 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2546 nr_deltas = n = 0;
2547
2548 for (i = 0; i < to_pack.nr_objects; i++) {
2549 struct object_entry *entry = to_pack.objects + i;
2550
2551 if (DELTA(entry))
2552 /* This happens if we decided to reuse existing
2553 * delta from a pack. "reuse_delta &&" is implied.
2554 */
2555 continue;
2556
2557 if (!entry->type_valid ||
2558 oe_size_less_than(&to_pack, entry, 50))
2559 continue;
2560
2561 if (entry->no_try_delta)
2562 continue;
2563
2564 if (!entry->preferred_base) {
2565 nr_deltas++;
2566 if (oe_type(entry) < 0)
2567 die(_("unable to get type of object %s"),
2568 oid_to_hex(&entry->idx.oid));
2569 } else {
2570 if (oe_type(entry) < 0) {
2571 /*
2572 * This object is not found, but we
2573 * don't have to include it anyway.
2574 */
2575 continue;
2576 }
2577 }
2578
2579 delta_list[n++] = entry;
2580 }
2581
2582 if (nr_deltas && n > 1) {
2583 unsigned nr_done = 0;
2584 if (progress)
2585 progress_state = start_progress(_("Compressing objects"),
2586 nr_deltas);
2587 QSORT(delta_list, n, type_size_sort);
2588 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2589 stop_progress(&progress_state);
2590 if (nr_done != nr_deltas)
2591 die(_("inconsistency with delta count"));
2592 }
2593 free(delta_list);
2594 }
2595
2596 static int git_pack_config(const char *k, const char *v, void *cb)
2597 {
2598 if (!strcmp(k, "pack.window")) {
2599 window = git_config_int(k, v);
2600 return 0;
2601 }
2602 if (!strcmp(k, "pack.windowmemory")) {
2603 window_memory_limit = git_config_ulong(k, v);
2604 return 0;
2605 }
2606 if (!strcmp(k, "pack.depth")) {
2607 depth = git_config_int(k, v);
2608 return 0;
2609 }
2610 if (!strcmp(k, "pack.deltacachesize")) {
2611 max_delta_cache_size = git_config_int(k, v);
2612 return 0;
2613 }
2614 if (!strcmp(k, "pack.deltacachelimit")) {
2615 cache_max_small_delta_size = git_config_int(k, v);
2616 return 0;
2617 }
2618 if (!strcmp(k, "pack.writebitmaphashcache")) {
2619 if (git_config_bool(k, v))
2620 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2621 else
2622 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2623 }
2624 if (!strcmp(k, "pack.usebitmaps")) {
2625 use_bitmap_index_default = git_config_bool(k, v);
2626 return 0;
2627 }
2628 if (!strcmp(k, "pack.threads")) {
2629 delta_search_threads = git_config_int(k, v);
2630 if (delta_search_threads < 0)
2631 die(_("invalid number of threads specified (%d)"),
2632 delta_search_threads);
2633 #ifdef NO_PTHREADS
2634 if (delta_search_threads != 1) {
2635 warning(_("no threads support, ignoring %s"), k);
2636 delta_search_threads = 0;
2637 }
2638 #endif
2639 return 0;
2640 }
2641 if (!strcmp(k, "pack.indexversion")) {
2642 pack_idx_opts.version = git_config_int(k, v);
2643 if (pack_idx_opts.version > 2)
2644 die(_("bad pack.indexversion=%"PRIu32),
2645 pack_idx_opts.version);
2646 return 0;
2647 }
2648 return git_default_config(k, v, cb);
2649 }
2650
2651 static void read_object_list_from_stdin(void)
2652 {
2653 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2654 struct object_id oid;
2655 const char *p;
2656
2657 for (;;) {
2658 if (!fgets(line, sizeof(line), stdin)) {
2659 if (feof(stdin))
2660 break;
2661 if (!ferror(stdin))
2662 die("BUG: fgets returned NULL, not EOF, not error!");
2663 if (errno != EINTR)
2664 die_errno("fgets");
2665 clearerr(stdin);
2666 continue;
2667 }
2668 if (line[0] == '-') {
2669 if (get_oid_hex(line+1, &oid))
2670 die(_("expected edge object ID, got garbage:\n %s"),
2671 line);
2672 add_preferred_base(&oid);
2673 continue;
2674 }
2675 if (parse_oid_hex(line, &oid, &p))
2676 die(_("expected object ID, got garbage:\n %s"), line);
2677
2678 add_preferred_base_object(p + 1);
2679 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2680 }
2681 }
2682
2683 /* Remember to update object flag allocation in object.h */
2684 #define OBJECT_ADDED (1u<<20)
2685
2686 static void show_commit(struct commit *commit, void *data)
2687 {
2688 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2689 commit->object.flags |= OBJECT_ADDED;
2690
2691 if (write_bitmap_index)
2692 index_commit_for_bitmap(commit);
2693 }
2694
2695 static void show_object(struct object *obj, const char *name, void *data)
2696 {
2697 add_preferred_base_object(name);
2698 add_object_entry(&obj->oid, obj->type, name, 0);
2699 obj->flags |= OBJECT_ADDED;
2700 }
2701
2702 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2703 {
2704 assert(arg_missing_action == MA_ALLOW_ANY);
2705
2706 /*
2707 * Quietly ignore ALL missing objects. This avoids problems with
2708 * staging them now and getting an odd error later.
2709 */
2710 if (!has_object_file(&obj->oid))
2711 return;
2712
2713 show_object(obj, name, data);
2714 }
2715
2716 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2717 {
2718 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2719
2720 /*
2721 * Quietly ignore EXPECTED missing objects. This avoids problems with
2722 * staging them now and getting an odd error later.
2723 */
2724 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2725 return;
2726
2727 show_object(obj, name, data);
2728 }
2729
2730 static int option_parse_missing_action(const struct option *opt,
2731 const char *arg, int unset)
2732 {
2733 assert(arg);
2734 assert(!unset);
2735
2736 if (!strcmp(arg, "error")) {
2737 arg_missing_action = MA_ERROR;
2738 fn_show_object = show_object;
2739 return 0;
2740 }
2741
2742 if (!strcmp(arg, "allow-any")) {
2743 arg_missing_action = MA_ALLOW_ANY;
2744 fetch_if_missing = 0;
2745 fn_show_object = show_object__ma_allow_any;
2746 return 0;
2747 }
2748
2749 if (!strcmp(arg, "allow-promisor")) {
2750 arg_missing_action = MA_ALLOW_PROMISOR;
2751 fetch_if_missing = 0;
2752 fn_show_object = show_object__ma_allow_promisor;
2753 return 0;
2754 }
2755
2756 die(_("invalid value for --missing"));
2757 return 0;
2758 }
2759
2760 static void show_edge(struct commit *commit)
2761 {
2762 add_preferred_base(&commit->object.oid);
2763 }
2764
2765 struct in_pack_object {
2766 off_t offset;
2767 struct object *object;
2768 };
2769
2770 struct in_pack {
2771 unsigned int alloc;
2772 unsigned int nr;
2773 struct in_pack_object *array;
2774 };
2775
2776 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2777 {
2778 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2779 in_pack->array[in_pack->nr].object = object;
2780 in_pack->nr++;
2781 }
2782
2783 /*
2784 * Compare the objects in the offset order, in order to emulate the
2785 * "git rev-list --objects" output that produced the pack originally.
2786 */
2787 static int ofscmp(const void *a_, const void *b_)
2788 {
2789 struct in_pack_object *a = (struct in_pack_object *)a_;
2790 struct in_pack_object *b = (struct in_pack_object *)b_;
2791
2792 if (a->offset < b->offset)
2793 return -1;
2794 else if (a->offset > b->offset)
2795 return 1;
2796 else
2797 return oidcmp(&a->object->oid, &b->object->oid);
2798 }
2799
2800 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2801 {
2802 struct packed_git *p;
2803 struct in_pack in_pack;
2804 uint32_t i;
2805
2806 memset(&in_pack, 0, sizeof(in_pack));
2807
2808 for (p = get_packed_git(the_repository); p; p = p->next) {
2809 struct object_id oid;
2810 struct object *o;
2811
2812 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2813 continue;
2814 if (open_pack_index(p))
2815 die(_("cannot open pack index"));
2816
2817 ALLOC_GROW(in_pack.array,
2818 in_pack.nr + p->num_objects,
2819 in_pack.alloc);
2820
2821 for (i = 0; i < p->num_objects; i++) {
2822 nth_packed_object_oid(&oid, p, i);
2823 o = lookup_unknown_object(oid.hash);
2824 if (!(o->flags & OBJECT_ADDED))
2825 mark_in_pack_object(o, p, &in_pack);
2826 o->flags |= OBJECT_ADDED;
2827 }
2828 }
2829
2830 if (in_pack.nr) {
2831 QSORT(in_pack.array, in_pack.nr, ofscmp);
2832 for (i = 0; i < in_pack.nr; i++) {
2833 struct object *o = in_pack.array[i].object;
2834 add_object_entry(&o->oid, o->type, "", 0);
2835 }
2836 }
2837 free(in_pack.array);
2838 }
2839
2840 static int add_loose_object(const struct object_id *oid, const char *path,
2841 void *data)
2842 {
2843 enum object_type type = oid_object_info(the_repository, oid, NULL);
2844
2845 if (type < 0) {
2846 warning(_("loose object at %s could not be examined"), path);
2847 return 0;
2848 }
2849
2850 add_object_entry(oid, type, "", 0);
2851 return 0;
2852 }
2853
2854 /*
2855 * We actually don't even have to worry about reachability here.
2856 * add_object_entry will weed out duplicates, so we just add every
2857 * loose object we find.
2858 */
2859 static void add_unreachable_loose_objects(void)
2860 {
2861 for_each_loose_file_in_objdir(get_object_directory(),
2862 add_loose_object,
2863 NULL, NULL, NULL);
2864 }
2865
2866 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2867 {
2868 static struct packed_git *last_found = (void *)1;
2869 struct packed_git *p;
2870
2871 p = (last_found != (void *)1) ? last_found :
2872 get_packed_git(the_repository);
2873
2874 while (p) {
2875 if ((!p->pack_local || p->pack_keep ||
2876 p->pack_keep_in_core) &&
2877 find_pack_entry_one(oid->hash, p)) {
2878 last_found = p;
2879 return 1;
2880 }
2881 if (p == last_found)
2882 p = get_packed_git(the_repository);
2883 else
2884 p = p->next;
2885 if (p == last_found)
2886 p = p->next;
2887 }
2888 return 0;
2889 }
2890
2891 /*
2892 * Store a list of sha1s that are should not be discarded
2893 * because they are either written too recently, or are
2894 * reachable from another object that was.
2895 *
2896 * This is filled by get_object_list.
2897 */
2898 static struct oid_array recent_objects;
2899
2900 static int loosened_object_can_be_discarded(const struct object_id *oid,
2901 timestamp_t mtime)
2902 {
2903 if (!unpack_unreachable_expiration)
2904 return 0;
2905 if (mtime > unpack_unreachable_expiration)
2906 return 0;
2907 if (oid_array_lookup(&recent_objects, oid) >= 0)
2908 return 0;
2909 return 1;
2910 }
2911
2912 static void loosen_unused_packed_objects(struct rev_info *revs)
2913 {
2914 struct packed_git *p;
2915 uint32_t i;
2916 struct object_id oid;
2917
2918 for (p = get_packed_git(the_repository); p; p = p->next) {
2919 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2920 continue;
2921
2922 if (open_pack_index(p))
2923 die(_("cannot open pack index"));
2924
2925 for (i = 0; i < p->num_objects; i++) {
2926 nth_packed_object_oid(&oid, p, i);
2927 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2928 !has_sha1_pack_kept_or_nonlocal(&oid) &&
2929 !loosened_object_can_be_discarded(&oid, p->mtime))
2930 if (force_object_loose(&oid, p->mtime))
2931 die(_("unable to force loose object"));
2932 }
2933 }
2934 }
2935
2936 /*
2937 * This tracks any options which pack-reuse code expects to be on, or which a
2938 * reader of the pack might not understand, and which would therefore prevent
2939 * blind reuse of what we have on disk.
2940 */
2941 static int pack_options_allow_reuse(void)
2942 {
2943 return pack_to_stdout &&
2944 allow_ofs_delta &&
2945 !ignore_packed_keep_on_disk &&
2946 !ignore_packed_keep_in_core &&
2947 (!local || !have_non_local_packs) &&
2948 !incremental;
2949 }
2950
2951 static int get_object_list_from_bitmap(struct rev_info *revs)
2952 {
2953 struct bitmap_index *bitmap_git;
2954 if (!(bitmap_git = prepare_bitmap_walk(revs)))
2955 return -1;
2956
2957 if (pack_options_allow_reuse() &&
2958 !reuse_partial_packfile_from_bitmap(
2959 bitmap_git,
2960 &reuse_packfile,
2961 &reuse_packfile_objects,
2962 &reuse_packfile_offset)) {
2963 assert(reuse_packfile_objects);
2964 nr_result += reuse_packfile_objects;
2965 display_progress(progress_state, nr_result);
2966 }
2967
2968 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
2969 free_bitmap_index(bitmap_git);
2970 return 0;
2971 }
2972
2973 static void record_recent_object(struct object *obj,
2974 const char *name,
2975 void *data)
2976 {
2977 oid_array_append(&recent_objects, &obj->oid);
2978 }
2979
2980 static void record_recent_commit(struct commit *commit, void *data)
2981 {
2982 oid_array_append(&recent_objects, &commit->object.oid);
2983 }
2984
2985 static void get_object_list(int ac, const char **av)
2986 {
2987 struct rev_info revs;
2988 char line[1000];
2989 int flags = 0;
2990
2991 init_revisions(&revs, NULL);
2992 save_commit_buffer = 0;
2993 setup_revisions(ac, av, &revs, NULL);
2994
2995 /* make sure shallows are read */
2996 is_repository_shallow(the_repository);
2997
2998 while (fgets(line, sizeof(line), stdin) != NULL) {
2999 int len = strlen(line);
3000 if (len && line[len - 1] == '\n')
3001 line[--len] = 0;
3002 if (!len)
3003 break;
3004 if (*line == '-') {
3005 if (!strcmp(line, "--not")) {
3006 flags ^= UNINTERESTING;
3007 write_bitmap_index = 0;
3008 continue;
3009 }
3010 if (starts_with(line, "--shallow ")) {
3011 struct object_id oid;
3012 if (get_oid_hex(line + 10, &oid))
3013 die("not an SHA-1 '%s'", line + 10);
3014 register_shallow(the_repository, &oid);
3015 use_bitmap_index = 0;
3016 continue;
3017 }
3018 die(_("not a rev '%s'"), line);
3019 }
3020 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3021 die(_("bad revision '%s'"), line);
3022 }
3023
3024 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3025 return;
3026
3027 if (prepare_revision_walk(&revs))
3028 die(_("revision walk setup failed"));
3029 mark_edges_uninteresting(&revs, show_edge);
3030
3031 if (!fn_show_object)
3032 fn_show_object = show_object;
3033 traverse_commit_list_filtered(&filter_options, &revs,
3034 show_commit, fn_show_object, NULL,
3035 NULL);
3036
3037 if (unpack_unreachable_expiration) {
3038 revs.ignore_missing_links = 1;
3039 if (add_unseen_recent_objects_to_traversal(&revs,
3040 unpack_unreachable_expiration))
3041 die(_("unable to add recent objects"));
3042 if (prepare_revision_walk(&revs))
3043 die(_("revision walk setup failed"));
3044 traverse_commit_list(&revs, record_recent_commit,
3045 record_recent_object, NULL);
3046 }
3047
3048 if (keep_unreachable)
3049 add_objects_in_unpacked_packs(&revs);
3050 if (pack_loose_unreachable)
3051 add_unreachable_loose_objects();
3052 if (unpack_unreachable)
3053 loosen_unused_packed_objects(&revs);
3054
3055 oid_array_clear(&recent_objects);
3056 }
3057
3058 static void add_extra_kept_packs(const struct string_list *names)
3059 {
3060 struct packed_git *p;
3061
3062 if (!names->nr)
3063 return;
3064
3065 for (p = get_packed_git(the_repository); p; p = p->next) {
3066 const char *name = basename(p->pack_name);
3067 int i;
3068
3069 if (!p->pack_local)
3070 continue;
3071
3072 for (i = 0; i < names->nr; i++)
3073 if (!fspathcmp(name, names->items[i].string))
3074 break;
3075
3076 if (i < names->nr) {
3077 p->pack_keep_in_core = 1;
3078 ignore_packed_keep_in_core = 1;
3079 continue;
3080 }
3081 }
3082 }
3083
3084 static int option_parse_index_version(const struct option *opt,
3085 const char *arg, int unset)
3086 {
3087 char *c;
3088 const char *val = arg;
3089 pack_idx_opts.version = strtoul(val, &c, 10);
3090 if (pack_idx_opts.version > 2)
3091 die(_("unsupported index version %s"), val);
3092 if (*c == ',' && c[1])
3093 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3094 if (*c || pack_idx_opts.off32_limit & 0x80000000)
3095 die(_("bad index version '%s'"), val);
3096 return 0;
3097 }
3098
3099 static int option_parse_unpack_unreachable(const struct option *opt,
3100 const char *arg, int unset)
3101 {
3102 if (unset) {
3103 unpack_unreachable = 0;
3104 unpack_unreachable_expiration = 0;
3105 }
3106 else {
3107 unpack_unreachable = 1;
3108 if (arg)
3109 unpack_unreachable_expiration = approxidate(arg);
3110 }
3111 return 0;
3112 }
3113
3114 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3115 {
3116 int use_internal_rev_list = 0;
3117 int thin = 0;
3118 int shallow = 0;
3119 int all_progress_implied = 0;
3120 struct argv_array rp = ARGV_ARRAY_INIT;
3121 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3122 int rev_list_index = 0;
3123 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3124 struct option pack_objects_options[] = {
3125 OPT_SET_INT('q', "quiet", &progress,
3126 N_("do not show progress meter"), 0),
3127 OPT_SET_INT(0, "progress", &progress,
3128 N_("show progress meter"), 1),
3129 OPT_SET_INT(0, "all-progress", &progress,
3130 N_("show progress meter during object writing phase"), 2),
3131 OPT_BOOL(0, "all-progress-implied",
3132 &all_progress_implied,
3133 N_("similar to --all-progress when progress meter is shown")),
3134 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),
3135 N_("write the pack index file in the specified idx format version"),
3136 0, option_parse_index_version },
3137 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3138 N_("maximum size of each output pack file")),
3139 OPT_BOOL(0, "local", &local,
3140 N_("ignore borrowed objects from alternate object store")),
3141 OPT_BOOL(0, "incremental", &incremental,
3142 N_("ignore packed objects")),
3143 OPT_INTEGER(0, "window", &window,
3144 N_("limit pack window by objects")),
3145 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3146 N_("limit pack window by memory in addition to object limit")),
3147 OPT_INTEGER(0, "depth", &depth,
3148 N_("maximum length of delta chain allowed in the resulting pack")),
3149 OPT_BOOL(0, "reuse-delta", &reuse_delta,
3150 N_("reuse existing deltas")),
3151 OPT_BOOL(0, "reuse-object", &reuse_object,
3152 N_("reuse existing objects")),
3153 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3154 N_("use OFS_DELTA objects")),
3155 OPT_INTEGER(0, "threads", &delta_search_threads,
3156 N_("use threads when searching for best delta matches")),
3157 OPT_BOOL(0, "non-empty", &non_empty,
3158 N_("do not create an empty pack output")),
3159 OPT_BOOL(0, "revs", &use_internal_rev_list,
3160 N_("read revision arguments from standard input")),
3161 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3162 N_("limit the objects to those that are not yet packed"),
3163 1, PARSE_OPT_NONEG),
3164 OPT_SET_INT_F(0, "all", &rev_list_all,
3165 N_("include objects reachable from any reference"),
3166 1, PARSE_OPT_NONEG),
3167 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3168 N_("include objects referred by reflog entries"),
3169 1, PARSE_OPT_NONEG),
3170 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3171 N_("include objects referred to by the index"),
3172 1, PARSE_OPT_NONEG),
3173 OPT_BOOL(0, "stdout", &pack_to_stdout,
3174 N_("output pack to stdout")),
3175 OPT_BOOL(0, "include-tag", &include_tag,
3176 N_("include tag objects that refer to objects to be packed")),
3177 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3178 N_("keep unreachable objects")),
3179 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3180 N_("pack loose unreachable objects")),
3181 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3182 N_("unpack unreachable objects newer than <time>"),
3183 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3184 OPT_BOOL(0, "thin", &thin,
3185 N_("create thin packs")),
3186 OPT_BOOL(0, "shallow", &shallow,
3187 N_("create packs suitable for shallow fetches")),
3188 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3189 N_("ignore packs that have companion .keep file")),
3190 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3191 N_("ignore this pack")),
3192 OPT_INTEGER(0, "compression", &pack_compression_level,
3193 N_("pack compression level")),
3194 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3195 N_("do not hide commits by grafts"), 0),
3196 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3197 N_("use a bitmap index if available to speed up counting objects")),
3198 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3199 N_("write a bitmap index together with the pack index")),
3200 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3201 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3202 N_("handling for missing objects"), PARSE_OPT_NONEG,
3203 option_parse_missing_action },
3204 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3205 N_("do not pack objects in promisor packfiles")),
3206 OPT_END(),
3207 };
3208
3209 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3210 BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3211
3212 read_replace_refs = 0;
3213
3214 reset_pack_idx_option(&pack_idx_opts);
3215 git_config(git_pack_config, NULL);
3216
3217 progress = isatty(2);
3218 argc = parse_options(argc, argv, prefix, pack_objects_options,
3219 pack_usage, 0);
3220
3221 if (argc) {
3222 base_name = argv[0];
3223 argc--;
3224 }
3225 if (pack_to_stdout != !base_name || argc)
3226 usage_with_options(pack_usage, pack_objects_options);
3227
3228 if (depth >= (1 << OE_DEPTH_BITS)) {
3229 warning(_("delta chain depth %d is too deep, forcing %d"),
3230 depth, (1 << OE_DEPTH_BITS) - 1);
3231 depth = (1 << OE_DEPTH_BITS) - 1;
3232 }
3233 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3234 warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3235 (1U << OE_Z_DELTA_BITS) - 1);
3236 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3237 }
3238
3239 argv_array_push(&rp, "pack-objects");
3240 if (thin) {
3241 use_internal_rev_list = 1;
3242 argv_array_push(&rp, shallow
3243 ? "--objects-edge-aggressive"
3244 : "--objects-edge");
3245 } else
3246 argv_array_push(&rp, "--objects");
3247
3248 if (rev_list_all) {
3249 use_internal_rev_list = 1;
3250 argv_array_push(&rp, "--all");
3251 }
3252 if (rev_list_reflog) {
3253 use_internal_rev_list = 1;
3254 argv_array_push(&rp, "--reflog");
3255 }
3256 if (rev_list_index) {
3257 use_internal_rev_list = 1;
3258 argv_array_push(&rp, "--indexed-objects");
3259 }
3260 if (rev_list_unpacked) {
3261 use_internal_rev_list = 1;
3262 argv_array_push(&rp, "--unpacked");
3263 }
3264
3265 if (exclude_promisor_objects) {
3266 use_internal_rev_list = 1;
3267 fetch_if_missing = 0;
3268 argv_array_push(&rp, "--exclude-promisor-objects");
3269 }
3270 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3271 use_internal_rev_list = 1;
3272
3273 if (!reuse_object)
3274 reuse_delta = 0;
3275 if (pack_compression_level == -1)
3276 pack_compression_level = Z_DEFAULT_COMPRESSION;
3277 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3278 die(_("bad pack compression level %d"), pack_compression_level);
3279
3280 if (!delta_search_threads) /* --threads=0 means autodetect */
3281 delta_search_threads = online_cpus();
3282
3283 #ifdef NO_PTHREADS
3284 if (delta_search_threads != 1)
3285 warning(_("no threads support, ignoring --threads"));
3286 #endif
3287 if (!pack_to_stdout && !pack_size_limit)
3288 pack_size_limit = pack_size_limit_cfg;
3289 if (pack_to_stdout && pack_size_limit)
3290 die(_("--max-pack-size cannot be used to build a pack for transfer"));
3291 if (pack_size_limit && pack_size_limit < 1024*1024) {
3292 warning(_("minimum pack size limit is 1 MiB"));
3293 pack_size_limit = 1024*1024;
3294 }
3295
3296 if (!pack_to_stdout && thin)
3297 die(_("--thin cannot be used to build an indexable pack"));
3298
3299 if (keep_unreachable && unpack_unreachable)
3300 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));
3301 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3302 unpack_unreachable_expiration = 0;
3303
3304 if (filter_options.choice) {
3305 if (!pack_to_stdout)
3306 die(_("cannot use --filter without --stdout"));
3307 use_bitmap_index = 0;
3308 }
3309
3310 /*
3311 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3312 *
3313 * - to produce good pack (with bitmap index not-yet-packed objects are
3314 * packed in suboptimal order).
3315 *
3316 * - to use more robust pack-generation codepath (avoiding possible
3317 * bugs in bitmap code and possible bitmap index corruption).
3318 */
3319 if (!pack_to_stdout)
3320 use_bitmap_index_default = 0;
3321
3322 if (use_bitmap_index < 0)
3323 use_bitmap_index = use_bitmap_index_default;
3324
3325 /* "hard" reasons not to use bitmaps; these just won't work at all */
3326 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))
3327 use_bitmap_index = 0;
3328
3329 if (pack_to_stdout || !rev_list_all)
3330 write_bitmap_index = 0;
3331
3332 if (progress && all_progress_implied)
3333 progress = 2;
3334
3335 add_extra_kept_packs(&keep_pack_list);
3336 if (ignore_packed_keep_on_disk) {
3337 struct packed_git *p;
3338 for (p = get_packed_git(the_repository); p; p = p->next)
3339 if (p->pack_local && p->pack_keep)
3340 break;
3341 if (!p) /* no keep-able packs found */
3342 ignore_packed_keep_on_disk = 0;
3343 }
3344 if (local) {
3345 /*
3346 * unlike ignore_packed_keep_on_disk above, we do not
3347 * want to unset "local" based on looking at packs, as
3348 * it also covers non-local objects
3349 */
3350 struct packed_git *p;
3351 for (p = get_packed_git(the_repository); p; p = p->next) {
3352 if (!p->pack_local) {
3353 have_non_local_packs = 1;
3354 break;
3355 }
3356 }
3357 }
3358
3359 prepare_packing_data(&to_pack);
3360
3361 if (progress)
3362 progress_state = start_progress(_("Enumerating objects"), 0);
3363 if (!use_internal_rev_list)
3364 read_object_list_from_stdin();
3365 else {
3366 get_object_list(rp.argc, rp.argv);
3367 argv_array_clear(&rp);
3368 }
3369 cleanup_preferred_base();
3370 if (include_tag && nr_result)
3371 for_each_ref(add_ref_tag, NULL);
3372 stop_progress(&progress_state);
3373
3374 if (non_empty && !nr_result)
3375 return 0;
3376 if (nr_result)
3377 prepare_pack(window, depth);
3378 write_pack_file();
3379 if (progress)
3380 fprintf_ln(stderr,
3381 _("Total %"PRIu32" (delta %"PRIu32"),"
3382 " reused %"PRIu32" (delta %"PRIu32")"),
3383 written, written_delta, reused, reused_delta);
3384 return 0;
3385 }