![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "progress.h"
18 #include "refs.h"
19 #include "thread-utils.h"
21 static const char pack_usage[] =
22 "git pack-objects [ -q | --progress | --all-progress ]\n"
23 " [--all-progress-implied]\n"
24 " [--max-pack-size=<n>] [--local] [--incremental]\n"
25 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
26 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
27 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
28 " [--reflog] [--stdout | base-name] [--include-tag]\n"
29 " [--keep-unreachable | --unpack-unreachable]\n"
30 " [< ref-list | < object-list]";
32 struct object_entry {
33 struct pack_idx_entry idx;
34 unsigned long size; /* uncompressed size */
35 struct packed_git *in_pack; /* already in pack */
36 off_t in_pack_offset;
37 struct object_entry *delta; /* delta base object */
38 struct object_entry *delta_child; /* deltified objects who bases me */
39 struct object_entry *delta_sibling; /* other deltified objects who
40 * uses the same base as me
41 */
42 void *delta_data; /* cached delta (uncompressed) */
43 unsigned long delta_size; /* delta data size (uncompressed) */
44 unsigned long z_delta_size; /* delta data size (compressed) */
45 unsigned int hash; /* name hint hash */
46 enum object_type type;
47 enum object_type in_pack_type; /* could be delta */
48 unsigned char in_pack_header_size;
49 unsigned char preferred_base; /* we do not pack this, but is available
50 * to be used as the base object to delta
51 * objects against.
52 */
53 unsigned char no_try_delta;
54 unsigned char tagged; /* near the very tip of refs */
55 unsigned char filled; /* assigned write-order */
56 };
58 /*
59 * Objects we are going to pack are collected in objects array (dynamically
60 * expanded). nr_objects & nr_alloc controls this array. They are stored
61 * in the order we see -- typically rev-list --objects order that gives us
62 * nice "minimum seek" order.
63 */
64 static struct object_entry *objects;
65 static struct pack_idx_entry **written_list;
66 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
68 static int non_empty;
69 static int reuse_delta = 1, reuse_object = 1;
70 static int keep_unreachable, unpack_unreachable, include_tag;
71 static int local;
72 static int incremental;
73 static int ignore_packed_keep;
74 static int allow_ofs_delta;
75 static struct pack_idx_option pack_idx_opts;
76 static const char *base_name;
77 static int progress = 1;
78 static int window = 10;
79 static unsigned long pack_size_limit, pack_size_limit_cfg;
80 static int depth = 50;
81 static int delta_search_threads;
82 static int pack_to_stdout;
83 static int num_preferred_base;
84 static struct progress *progress_state;
85 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
86 static int pack_compression_seen;
88 static unsigned long delta_cache_size = 0;
89 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
90 static unsigned long cache_max_small_delta_size = 1000;
92 static unsigned long window_memory_limit = 0;
94 /*
95 * The object names in objects array are hashed with this hashtable,
96 * to help looking up the entry by object name.
97 * This hashtable is built after all the objects are seen.
98 */
99 static int *object_ix;
100 static int object_ix_hashsz;
101 static struct object_entry *locate_object_entry(const unsigned char *sha1);
103 /*
104 * stats
105 */
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
110 static void *get_delta(struct object_entry *entry)
111 {
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
114 enum object_type type;
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
117 if (!buf)
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
120 if (!base_buf)
121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
122 delta_buf = diff_delta(base_buf, base_size,
123 buf, size, &delta_size, 0);
124 if (!delta_buf || delta_size != entry->delta_size)
125 die("delta size changed");
126 free(buf);
127 free(base_buf);
128 return delta_buf;
129 }
131 static unsigned long do_compress(void **pptr, unsigned long size)
132 {
133 git_zstream stream;
134 void *in, *out;
135 unsigned long maxsize;
137 memset(&stream, 0, sizeof(stream));
138 git_deflate_init(&stream, pack_compression_level);
139 maxsize = git_deflate_bound(&stream, size);
141 in = *pptr;
142 out = xmalloc(maxsize);
143 *pptr = out;
145 stream.next_in = in;
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (git_deflate(&stream, Z_FINISH) == Z_OK)
150 ; /* nothing */
151 git_deflate_end(&stream);
153 free(in);
154 return stream.total_out;
155 }
157 /*
158 * we are going to reuse the existing object data as is. make
159 * sure it is not corrupt.
160 */
161 static int check_pack_inflate(struct packed_git *p,
162 struct pack_window **w_curs,
163 off_t offset,
164 off_t len,
165 unsigned long expect)
166 {
167 git_zstream stream;
168 unsigned char fakebuf[4096], *in;
169 int st;
171 memset(&stream, 0, sizeof(stream));
172 git_inflate_init(&stream);
173 do {
174 in = use_pack(p, w_curs, offset, &stream.avail_in);
175 stream.next_in = in;
176 stream.next_out = fakebuf;
177 stream.avail_out = sizeof(fakebuf);
178 st = git_inflate(&stream, Z_FINISH);
179 offset += stream.next_in - in;
180 } while (st == Z_OK || st == Z_BUF_ERROR);
181 git_inflate_end(&stream);
182 return (st == Z_STREAM_END &&
183 stream.total_out == expect &&
184 stream.total_in == len) ? 0 : -1;
185 }
187 static void copy_pack_data(struct sha1file *f,
188 struct packed_git *p,
189 struct pack_window **w_curs,
190 off_t offset,
191 off_t len)
192 {
193 unsigned char *in;
194 unsigned long avail;
196 while (len) {
197 in = use_pack(p, w_curs, offset, &avail);
198 if (avail > len)
199 avail = (unsigned long)len;
200 sha1write(f, in, avail);
201 offset += avail;
202 len -= avail;
203 }
204 }
206 /* Return 0 if we will bust the pack-size limit */
207 static unsigned long write_object(struct sha1file *f,
208 struct object_entry *entry,
209 off_t write_offset)
210 {
211 unsigned long size, limit, datalen;
212 void *buf;
213 unsigned char header[10], dheader[10];
214 unsigned hdrlen;
215 enum object_type type;
216 int usable_delta, to_reuse;
218 if (!pack_to_stdout)
219 crc32_begin(f);
221 type = entry->type;
223 /* apply size limit if limited packsize and not first object */
224 if (!pack_size_limit || !nr_written)
225 limit = 0;
226 else if (pack_size_limit <= write_offset)
227 /*
228 * the earlier object did not fit the limit; avoid
229 * mistaking this with unlimited (i.e. limit = 0).
230 */
231 limit = 1;
232 else
233 limit = pack_size_limit - write_offset;
235 if (!entry->delta)
236 usable_delta = 0; /* no delta */
237 else if (!pack_size_limit)
238 usable_delta = 1; /* unlimited packfile */
239 else if (entry->delta->idx.offset == (off_t)-1)
240 usable_delta = 0; /* base was written to another pack */
241 else if (entry->delta->idx.offset)
242 usable_delta = 1; /* base already exists in this pack */
243 else
244 usable_delta = 0; /* base could end up in another pack */
246 if (!reuse_object)
247 to_reuse = 0; /* explicit */
248 else if (!entry->in_pack)
249 to_reuse = 0; /* can't reuse what we don't have */
250 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
251 /* check_object() decided it for us ... */
252 to_reuse = usable_delta;
253 /* ... but pack split may override that */
254 else if (type != entry->in_pack_type)
255 to_reuse = 0; /* pack has delta which is unusable */
256 else if (entry->delta)
257 to_reuse = 0; /* we want to pack afresh */
258 else
259 to_reuse = 1; /* we have it in-pack undeltified,
260 * and we do not need to deltify it.
261 */
263 if (!to_reuse) {
264 no_reuse:
265 if (!usable_delta) {
266 buf = read_sha1_file(entry->idx.sha1, &type, &size);
267 if (!buf)
268 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
269 /*
270 * make sure no cached delta data remains from a
271 * previous attempt before a pack split occurred.
272 */
273 free(entry->delta_data);
274 entry->delta_data = NULL;
275 entry->z_delta_size = 0;
276 } else if (entry->delta_data) {
277 size = entry->delta_size;
278 buf = entry->delta_data;
279 entry->delta_data = NULL;
280 type = (allow_ofs_delta && entry->delta->idx.offset) ?
281 OBJ_OFS_DELTA : OBJ_REF_DELTA;
282 } else {
283 buf = get_delta(entry);
284 size = entry->delta_size;
285 type = (allow_ofs_delta && entry->delta->idx.offset) ?
286 OBJ_OFS_DELTA : OBJ_REF_DELTA;
287 }
289 if (entry->z_delta_size)
290 datalen = entry->z_delta_size;
291 else
292 datalen = do_compress(&buf, size);
294 /*
295 * The object header is a byte of 'type' followed by zero or
296 * more bytes of length.
297 */
298 hdrlen = encode_in_pack_object_header(type, size, header);
300 if (type == OBJ_OFS_DELTA) {
301 /*
302 * Deltas with relative base contain an additional
303 * encoding of the relative offset for the delta
304 * base from this object's position in the pack.
305 */
306 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
307 unsigned pos = sizeof(dheader) - 1;
308 dheader[pos] = ofs & 127;
309 while (ofs >>= 7)
310 dheader[--pos] = 128 | (--ofs & 127);
311 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
312 free(buf);
313 return 0;
314 }
315 sha1write(f, header, hdrlen);
316 sha1write(f, dheader + pos, sizeof(dheader) - pos);
317 hdrlen += sizeof(dheader) - pos;
318 } else if (type == OBJ_REF_DELTA) {
319 /*
320 * Deltas with a base reference contain
321 * an additional 20 bytes for the base sha1.
322 */
323 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
324 free(buf);
325 return 0;
326 }
327 sha1write(f, header, hdrlen);
328 sha1write(f, entry->delta->idx.sha1, 20);
329 hdrlen += 20;
330 } else {
331 if (limit && hdrlen + datalen + 20 >= limit) {
332 free(buf);
333 return 0;
334 }
335 sha1write(f, header, hdrlen);
336 }
337 sha1write(f, buf, datalen);
338 free(buf);
339 }
340 else {
341 struct packed_git *p = entry->in_pack;
342 struct pack_window *w_curs = NULL;
343 struct revindex_entry *revidx;
344 off_t offset;
346 if (entry->delta)
347 type = (allow_ofs_delta && entry->delta->idx.offset) ?
348 OBJ_OFS_DELTA : OBJ_REF_DELTA;
349 hdrlen = encode_in_pack_object_header(type, entry->size, header);
351 offset = entry->in_pack_offset;
352 revidx = find_pack_revindex(p, offset);
353 datalen = revidx[1].offset - offset;
354 if (!pack_to_stdout && p->index_version > 1 &&
355 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
356 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
357 unuse_pack(&w_curs);
358 goto no_reuse;
359 }
361 offset += entry->in_pack_header_size;
362 datalen -= entry->in_pack_header_size;
363 if (!pack_to_stdout && p->index_version == 1 &&
364 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
365 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
366 unuse_pack(&w_curs);
367 goto no_reuse;
368 }
370 if (type == OBJ_OFS_DELTA) {
371 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
372 unsigned pos = sizeof(dheader) - 1;
373 dheader[pos] = ofs & 127;
374 while (ofs >>= 7)
375 dheader[--pos] = 128 | (--ofs & 127);
376 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
377 unuse_pack(&w_curs);
378 return 0;
379 }
380 sha1write(f, header, hdrlen);
381 sha1write(f, dheader + pos, sizeof(dheader) - pos);
382 hdrlen += sizeof(dheader) - pos;
383 reused_delta++;
384 } else if (type == OBJ_REF_DELTA) {
385 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
386 unuse_pack(&w_curs);
387 return 0;
388 }
389 sha1write(f, header, hdrlen);
390 sha1write(f, entry->delta->idx.sha1, 20);
391 hdrlen += 20;
392 reused_delta++;
393 } else {
394 if (limit && hdrlen + datalen + 20 >= limit) {
395 unuse_pack(&w_curs);
396 return 0;
397 }
398 sha1write(f, header, hdrlen);
399 }
400 copy_pack_data(f, p, &w_curs, offset, datalen);
401 unuse_pack(&w_curs);
402 reused++;
403 }
404 if (usable_delta)
405 written_delta++;
406 written++;
407 if (!pack_to_stdout)
408 entry->idx.crc32 = crc32_end(f);
409 return hdrlen + datalen;
410 }
412 enum write_one_status {
413 WRITE_ONE_SKIP = -1, /* already written */
414 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
415 WRITE_ONE_WRITTEN = 1, /* normal */
416 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
417 };
419 static enum write_one_status write_one(struct sha1file *f,
420 struct object_entry *e,
421 off_t *offset)
422 {
423 unsigned long size;
424 int recursing;
426 /*
427 * we set offset to 1 (which is an impossible value) to mark
428 * the fact that this object is involved in "write its base
429 * first before writing a deltified object" recursion.
430 */
431 recursing = (e->idx.offset == 1);
432 if (recursing) {
433 warning("recursive delta detected for object %s",
434 sha1_to_hex(e->idx.sha1));
435 return WRITE_ONE_RECURSIVE;
436 } else if (e->idx.offset || e->preferred_base) {
437 /* offset is non zero if object is written already. */
438 return WRITE_ONE_SKIP;
439 }
441 /* if we are deltified, write out base object first. */
442 if (e->delta) {
443 e->idx.offset = 1; /* now recurse */
444 switch (write_one(f, e->delta, offset)) {
445 case WRITE_ONE_RECURSIVE:
446 /* we cannot depend on this one */
447 e->delta = NULL;
448 break;
449 default:
450 break;
451 case WRITE_ONE_BREAK:
452 e->idx.offset = recursing;
453 return WRITE_ONE_BREAK;
454 }
455 }
457 e->idx.offset = *offset;
458 size = write_object(f, e, *offset);
459 if (!size) {
460 e->idx.offset = recursing;
461 return WRITE_ONE_BREAK;
462 }
463 written_list[nr_written++] = &e->idx;
465 /* make sure off_t is sufficiently large not to wrap */
466 if (signed_add_overflows(*offset, size))
467 die("pack too large for current definition of off_t");
468 *offset += size;
469 return WRITE_ONE_WRITTEN;
470 }
472 static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
473 void *cb_data)
474 {
475 unsigned char peeled[20];
476 struct object_entry *entry = locate_object_entry(sha1);
478 if (entry)
479 entry->tagged = 1;
480 if (!peel_ref(path, peeled)) {
481 entry = locate_object_entry(peeled);
482 if (entry)
483 entry->tagged = 1;
484 }
485 return 0;
486 }
488 static inline void add_to_write_order(struct object_entry **wo,
489 unsigned int *endp,
490 struct object_entry *e)
491 {
492 if (e->filled)
493 return;
494 wo[(*endp)++] = e;
495 e->filled = 1;
496 }
498 static void add_descendants_to_write_order(struct object_entry **wo,
499 unsigned int *endp,
500 struct object_entry *e)
501 {
502 int add_to_order = 1;
503 while (e) {
504 if (add_to_order) {
505 struct object_entry *s;
506 /* add this node... */
507 add_to_write_order(wo, endp, e);
508 /* all its siblings... */
509 for (s = e->delta_sibling; s; s = s->delta_sibling) {
510 add_to_write_order(wo, endp, s);
511 }
512 }
513 /* drop down a level to add left subtree nodes if possible */
514 if (e->delta_child) {
515 add_to_order = 1;
516 e = e->delta_child;
517 } else {
518 add_to_order = 0;
519 /* our sibling might have some children, it is next */
520 if (e->delta_sibling) {
521 e = e->delta_sibling;
522 continue;
523 }
524 /* go back to our parent node */
525 e = e->delta;
526 while (e && !e->delta_sibling) {
527 /* we're on the right side of a subtree, keep
528 * going up until we can go right again */
529 e = e->delta;
530 }
531 if (!e) {
532 /* done- we hit our original root node */
533 return;
534 }
535 /* pass it off to sibling at this level */
536 e = e->delta_sibling;
537 }
538 };
539 }
541 static void add_family_to_write_order(struct object_entry **wo,
542 unsigned int *endp,
543 struct object_entry *e)
544 {
545 struct object_entry *root;
547 for (root = e; root->delta; root = root->delta)
548 ; /* nothing */
549 add_descendants_to_write_order(wo, endp, root);
550 }
552 static struct object_entry **compute_write_order(void)
553 {
554 unsigned int i, wo_end, last_untagged;
556 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));
558 for (i = 0; i < nr_objects; i++) {
559 objects[i].tagged = 0;
560 objects[i].filled = 0;
561 objects[i].delta_child = NULL;
562 objects[i].delta_sibling = NULL;
563 }
565 /*
566 * Fully connect delta_child/delta_sibling network.
567 * Make sure delta_sibling is sorted in the original
568 * recency order.
569 */
570 for (i = nr_objects; i > 0;) {
571 struct object_entry *e = &objects[--i];
572 if (!e->delta)
573 continue;
574 /* Mark me as the first child */
575 e->delta_sibling = e->delta->delta_child;
576 e->delta->delta_child = e;
577 }
579 /*
580 * Mark objects that are at the tip of tags.
581 */
582 for_each_tag_ref(mark_tagged, NULL);
584 /*
585 * Give the objects in the original recency order until
586 * we see a tagged tip.
587 */
588 for (i = wo_end = 0; i < nr_objects; i++) {
589 if (objects[i].tagged)
590 break;
591 add_to_write_order(wo, &wo_end, &objects[i]);
592 }
593 last_untagged = i;
595 /*
596 * Then fill all the tagged tips.
597 */
598 for (; i < nr_objects; i++) {
599 if (objects[i].tagged)
600 add_to_write_order(wo, &wo_end, &objects[i]);
601 }
603 /*
604 * And then all remaining commits and tags.
605 */
606 for (i = last_untagged; i < nr_objects; i++) {
607 if (objects[i].type != OBJ_COMMIT &&
608 objects[i].type != OBJ_TAG)
609 continue;
610 add_to_write_order(wo, &wo_end, &objects[i]);
611 }
613 /*
614 * And then all the trees.
615 */
616 for (i = last_untagged; i < nr_objects; i++) {
617 if (objects[i].type != OBJ_TREE)
618 continue;
619 add_to_write_order(wo, &wo_end, &objects[i]);
620 }
622 /*
623 * Finally all the rest in really tight order
624 */
625 for (i = last_untagged; i < nr_objects; i++) {
626 if (!objects[i].filled)
627 add_family_to_write_order(wo, &wo_end, &objects[i]);
628 }
630 if (wo_end != nr_objects)
631 die("ordered %u objects, expected %"PRIu32, wo_end, nr_objects);
633 return wo;
634 }
636 static void write_pack_file(void)
637 {
638 uint32_t i = 0, j;
639 struct sha1file *f;
640 off_t offset;
641 struct pack_header hdr;
642 uint32_t nr_remaining = nr_result;
643 time_t last_mtime = 0;
644 struct object_entry **write_order;
646 if (progress > pack_to_stdout)
647 progress_state = start_progress("Writing objects", nr_result);
648 written_list = xmalloc(nr_objects * sizeof(*written_list));
649 write_order = compute_write_order();
651 do {
652 unsigned char sha1[20];
653 char *pack_tmp_name = NULL;
655 if (pack_to_stdout) {
656 f = sha1fd_throughput(1, "<stdout>", progress_state);
657 } else {
658 char tmpname[PATH_MAX];
659 int fd;
660 fd = odb_mkstemp(tmpname, sizeof(tmpname),
661 "pack/tmp_pack_XXXXXX");
662 pack_tmp_name = xstrdup(tmpname);
663 f = sha1fd(fd, pack_tmp_name);
664 }
666 hdr.hdr_signature = htonl(PACK_SIGNATURE);
667 hdr.hdr_version = htonl(PACK_VERSION);
668 hdr.hdr_entries = htonl(nr_remaining);
669 sha1write(f, &hdr, sizeof(hdr));
670 offset = sizeof(hdr);
671 nr_written = 0;
672 for (; i < nr_objects; i++) {
673 struct object_entry *e = write_order[i];
674 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
675 break;
676 display_progress(progress_state, written);
677 }
679 /*
680 * Did we write the wrong # entries in the header?
681 * If so, rewrite it like in fast-import
682 */
683 if (pack_to_stdout) {
684 sha1close(f, sha1, CSUM_CLOSE);
685 } else if (nr_written == nr_remaining) {
686 sha1close(f, sha1, CSUM_FSYNC);
687 } else {
688 int fd = sha1close(f, sha1, 0);
689 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
690 nr_written, sha1, offset);
691 close(fd);
692 }
694 if (!pack_to_stdout) {
695 struct stat st;
696 const char *idx_tmp_name;
697 char tmpname[PATH_MAX];
699 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
700 &pack_idx_opts, sha1);
702 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
703 base_name, sha1_to_hex(sha1));
704 free_pack_by_name(tmpname);
705 if (adjust_shared_perm(pack_tmp_name))
706 die_errno("unable to make temporary pack file readable");
707 if (rename(pack_tmp_name, tmpname))
708 die_errno("unable to rename temporary pack file");
710 /*
711 * Packs are runtime accessed in their mtime
712 * order since newer packs are more likely to contain
713 * younger objects. So if we are creating multiple
714 * packs then we should modify the mtime of later ones
715 * to preserve this property.
716 */
717 if (stat(tmpname, &st) < 0) {
718 warning("failed to stat %s: %s",
719 tmpname, strerror(errno));
720 } else if (!last_mtime) {
721 last_mtime = st.st_mtime;
722 } else {
723 struct utimbuf utb;
724 utb.actime = st.st_atime;
725 utb.modtime = --last_mtime;
726 if (utime(tmpname, &utb) < 0)
727 warning("failed utime() on %s: %s",
728 tmpname, strerror(errno));
729 }
731 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
732 base_name, sha1_to_hex(sha1));
733 if (adjust_shared_perm(idx_tmp_name))
734 die_errno("unable to make temporary index file readable");
735 if (rename(idx_tmp_name, tmpname))
736 die_errno("unable to rename temporary index file");
738 free((void *) idx_tmp_name);
739 free(pack_tmp_name);
740 puts(sha1_to_hex(sha1));
741 }
743 /* mark written objects as written to previous pack */
744 for (j = 0; j < nr_written; j++) {
745 written_list[j]->offset = (off_t)-1;
746 }
747 nr_remaining -= nr_written;
748 } while (nr_remaining && i < nr_objects);
750 free(written_list);
751 free(write_order);
752 stop_progress(&progress_state);
753 if (written != nr_result)
754 die("wrote %"PRIu32" objects while expecting %"PRIu32,
755 written, nr_result);
756 }
758 static int locate_object_entry_hash(const unsigned char *sha1)
759 {
760 int i;
761 unsigned int ui;
762 memcpy(&ui, sha1, sizeof(unsigned int));
763 i = ui % object_ix_hashsz;
764 while (0 < object_ix[i]) {
765 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
766 return i;
767 if (++i == object_ix_hashsz)
768 i = 0;
769 }
770 return -1 - i;
771 }
773 static struct object_entry *locate_object_entry(const unsigned char *sha1)
774 {
775 int i;
777 if (!object_ix_hashsz)
778 return NULL;
780 i = locate_object_entry_hash(sha1);
781 if (0 <= i)
782 return &objects[object_ix[i]-1];
783 return NULL;
784 }
786 static void rehash_objects(void)
787 {
788 uint32_t i;
789 struct object_entry *oe;
791 object_ix_hashsz = nr_objects * 3;
792 if (object_ix_hashsz < 1024)
793 object_ix_hashsz = 1024;
794 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
795 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
796 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
797 int ix = locate_object_entry_hash(oe->idx.sha1);
798 if (0 <= ix)
799 continue;
800 ix = -1 - ix;
801 object_ix[ix] = i + 1;
802 }
803 }
805 static unsigned name_hash(const char *name)
806 {
807 unsigned c, hash = 0;
809 if (!name)
810 return 0;
812 /*
813 * This effectively just creates a sortable number from the
814 * last sixteen non-whitespace characters. Last characters
815 * count "most", so things that end in ".c" sort together.
816 */
817 while ((c = *name++) != 0) {
818 if (isspace(c))
819 continue;
820 hash = (hash >> 2) + (c << 24);
821 }
822 return hash;
823 }
825 static void setup_delta_attr_check(struct git_attr_check *check)
826 {
827 static struct git_attr *attr_delta;
829 if (!attr_delta)
830 attr_delta = git_attr("delta");
832 check[0].attr = attr_delta;
833 }
835 static int no_try_delta(const char *path)
836 {
837 struct git_attr_check check[1];
839 setup_delta_attr_check(check);
840 if (git_check_attr(path, ARRAY_SIZE(check), check))
841 return 0;
842 if (ATTR_FALSE(check->value))
843 return 1;
844 return 0;
845 }
847 static int add_object_entry(const unsigned char *sha1, enum object_type type,
848 const char *name, int exclude)
849 {
850 struct object_entry *entry;
851 struct packed_git *p, *found_pack = NULL;
852 off_t found_offset = 0;
853 int ix;
854 unsigned hash = name_hash(name);
856 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
857 if (ix >= 0) {
858 if (exclude) {
859 entry = objects + object_ix[ix] - 1;
860 if (!entry->preferred_base)
861 nr_result--;
862 entry->preferred_base = 1;
863 }
864 return 0;
865 }
867 if (!exclude && local && has_loose_object_nonlocal(sha1))
868 return 0;
870 for (p = packed_git; p; p = p->next) {
871 off_t offset = find_pack_entry_one(sha1, p);
872 if (offset) {
873 if (!found_pack) {
874 if (!is_pack_valid(p)) {
875 warning("packfile %s cannot be accessed", p->pack_name);
876 continue;
877 }
878 found_offset = offset;
879 found_pack = p;
880 }
881 if (exclude)
882 break;
883 if (incremental)
884 return 0;
885 if (local && !p->pack_local)
886 return 0;
887 if (ignore_packed_keep && p->pack_local && p->pack_keep)
888 return 0;
889 }
890 }
892 if (nr_objects >= nr_alloc) {
893 nr_alloc = (nr_alloc + 1024) * 3 / 2;
894 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
895 }
897 entry = objects + nr_objects++;
898 memset(entry, 0, sizeof(*entry));
899 hashcpy(entry->idx.sha1, sha1);
900 entry->hash = hash;
901 if (type)
902 entry->type = type;
903 if (exclude)
904 entry->preferred_base = 1;
905 else
906 nr_result++;
907 if (found_pack) {
908 entry->in_pack = found_pack;
909 entry->in_pack_offset = found_offset;
910 }
912 if (object_ix_hashsz * 3 <= nr_objects * 4)
913 rehash_objects();
914 else
915 object_ix[-1 - ix] = nr_objects;
917 display_progress(progress_state, nr_objects);
919 if (name && no_try_delta(name))
920 entry->no_try_delta = 1;
922 return 1;
923 }
925 struct pbase_tree_cache {
926 unsigned char sha1[20];
927 int ref;
928 int temporary;
929 void *tree_data;
930 unsigned long tree_size;
931 };
933 static struct pbase_tree_cache *(pbase_tree_cache[256]);
934 static int pbase_tree_cache_ix(const unsigned char *sha1)
935 {
936 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
937 }
938 static int pbase_tree_cache_ix_incr(int ix)
939 {
940 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
941 }
943 static struct pbase_tree {
944 struct pbase_tree *next;
945 /* This is a phony "cache" entry; we are not
946 * going to evict it nor find it through _get()
947 * mechanism -- this is for the toplevel node that
948 * would almost always change with any commit.
949 */
950 struct pbase_tree_cache pcache;
951 } *pbase_tree;
953 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
954 {
955 struct pbase_tree_cache *ent, *nent;
956 void *data;
957 unsigned long size;
958 enum object_type type;
959 int neigh;
960 int my_ix = pbase_tree_cache_ix(sha1);
961 int available_ix = -1;
963 /* pbase-tree-cache acts as a limited hashtable.
964 * your object will be found at your index or within a few
965 * slots after that slot if it is cached.
966 */
967 for (neigh = 0; neigh < 8; neigh++) {
968 ent = pbase_tree_cache[my_ix];
969 if (ent && !hashcmp(ent->sha1, sha1)) {
970 ent->ref++;
971 return ent;
972 }
973 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
974 ((0 <= available_ix) &&
975 (!ent && pbase_tree_cache[available_ix])))
976 available_ix = my_ix;
977 if (!ent)
978 break;
979 my_ix = pbase_tree_cache_ix_incr(my_ix);
980 }
982 /* Did not find one. Either we got a bogus request or
983 * we need to read and perhaps cache.
984 */
985 data = read_sha1_file(sha1, &type, &size);
986 if (!data)
987 return NULL;
988 if (type != OBJ_TREE) {
989 free(data);
990 return NULL;
991 }
993 /* We need to either cache or return a throwaway copy */
995 if (available_ix < 0)
996 ent = NULL;
997 else {
998 ent = pbase_tree_cache[available_ix];
999 my_ix = available_ix;
1000 }
1002 if (!ent) {
1003 nent = xmalloc(sizeof(*nent));
1004 nent->temporary = (available_ix < 0);
1005 }
1006 else {
1007 /* evict and reuse */
1008 free(ent->tree_data);
1009 nent = ent;
1010 }
1011 hashcpy(nent->sha1, sha1);
1012 nent->tree_data = data;
1013 nent->tree_size = size;
1014 nent->ref = 1;
1015 if (!nent->temporary)
1016 pbase_tree_cache[my_ix] = nent;
1017 return nent;
1018 }
1020 static void pbase_tree_put(struct pbase_tree_cache *cache)
1021 {
1022 if (!cache->temporary) {
1023 cache->ref--;
1024 return;
1025 }
1026 free(cache->tree_data);
1027 free(cache);
1028 }
1030 static int name_cmp_len(const char *name)
1031 {
1032 int i;
1033 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1034 ;
1035 return i;
1036 }
1038 static void add_pbase_object(struct tree_desc *tree,
1039 const char *name,
1040 int cmplen,
1041 const char *fullname)
1042 {
1043 struct name_entry entry;
1044 int cmp;
1046 while (tree_entry(tree,&entry)) {
1047 if (S_ISGITLINK(entry.mode))
1048 continue;
1049 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1050 memcmp(name, entry.path, cmplen);
1051 if (cmp > 0)
1052 continue;
1053 if (cmp < 0)
1054 return;
1055 if (name[cmplen] != '/') {
1056 add_object_entry(entry.sha1,
1057 object_type(entry.mode),
1058 fullname, 1);
1059 return;
1060 }
1061 if (S_ISDIR(entry.mode)) {
1062 struct tree_desc sub;
1063 struct pbase_tree_cache *tree;
1064 const char *down = name+cmplen+1;
1065 int downlen = name_cmp_len(down);
1067 tree = pbase_tree_get(entry.sha1);
1068 if (!tree)
1069 return;
1070 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1072 add_pbase_object(&sub, down, downlen, fullname);
1073 pbase_tree_put(tree);
1074 }
1075 }
1076 }
1078 static unsigned *done_pbase_paths;
1079 static int done_pbase_paths_num;
1080 static int done_pbase_paths_alloc;
1081 static int done_pbase_path_pos(unsigned hash)
1082 {
1083 int lo = 0;
1084 int hi = done_pbase_paths_num;
1085 while (lo < hi) {
1086 int mi = (hi + lo) / 2;
1087 if (done_pbase_paths[mi] == hash)
1088 return mi;
1089 if (done_pbase_paths[mi] < hash)
1090 hi = mi;
1091 else
1092 lo = mi + 1;
1093 }
1094 return -lo-1;
1095 }
1097 static int check_pbase_path(unsigned hash)
1098 {
1099 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1100 if (0 <= pos)
1101 return 1;
1102 pos = -pos - 1;
1103 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1104 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1105 done_pbase_paths = xrealloc(done_pbase_paths,
1106 done_pbase_paths_alloc *
1107 sizeof(unsigned));
1108 }
1109 done_pbase_paths_num++;
1110 if (pos < done_pbase_paths_num)
1111 memmove(done_pbase_paths + pos + 1,
1112 done_pbase_paths + pos,
1113 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1114 done_pbase_paths[pos] = hash;
1115 return 0;
1116 }
1118 static void add_preferred_base_object(const char *name)
1119 {
1120 struct pbase_tree *it;
1121 int cmplen;
1122 unsigned hash = name_hash(name);
1124 if (!num_preferred_base || check_pbase_path(hash))
1125 return;
1127 cmplen = name_cmp_len(name);
1128 for (it = pbase_tree; it; it = it->next) {
1129 if (cmplen == 0) {
1130 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1131 }
1132 else {
1133 struct tree_desc tree;
1134 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1135 add_pbase_object(&tree, name, cmplen, name);
1136 }
1137 }
1138 }
1140 static void add_preferred_base(unsigned char *sha1)
1141 {
1142 struct pbase_tree *it;
1143 void *data;
1144 unsigned long size;
1145 unsigned char tree_sha1[20];
1147 if (window <= num_preferred_base++)
1148 return;
1150 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1151 if (!data)
1152 return;
1154 for (it = pbase_tree; it; it = it->next) {
1155 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1156 free(data);
1157 return;
1158 }
1159 }
1161 it = xcalloc(1, sizeof(*it));
1162 it->next = pbase_tree;
1163 pbase_tree = it;
1165 hashcpy(it->pcache.sha1, tree_sha1);
1166 it->pcache.tree_data = data;
1167 it->pcache.tree_size = size;
1168 }
1170 static void cleanup_preferred_base(void)
1171 {
1172 struct pbase_tree *it;
1173 unsigned i;
1175 it = pbase_tree;
1176 pbase_tree = NULL;
1177 while (it) {
1178 struct pbase_tree *this = it;
1179 it = this->next;
1180 free(this->pcache.tree_data);
1181 free(this);
1182 }
1184 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1185 if (!pbase_tree_cache[i])
1186 continue;
1187 free(pbase_tree_cache[i]->tree_data);
1188 free(pbase_tree_cache[i]);
1189 pbase_tree_cache[i] = NULL;
1190 }
1192 free(done_pbase_paths);
1193 done_pbase_paths = NULL;
1194 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1195 }
1197 static void check_object(struct object_entry *entry)
1198 {
1199 if (entry->in_pack) {
1200 struct packed_git *p = entry->in_pack;
1201 struct pack_window *w_curs = NULL;
1202 const unsigned char *base_ref = NULL;
1203 struct object_entry *base_entry;
1204 unsigned long used, used_0;
1205 unsigned long avail;
1206 off_t ofs;
1207 unsigned char *buf, c;
1209 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1211 /*
1212 * We want in_pack_type even if we do not reuse delta
1213 * since non-delta representations could still be reused.
1214 */
1215 used = unpack_object_header_buffer(buf, avail,
1216 &entry->in_pack_type,
1217 &entry->size);
1218 if (used == 0)
1219 goto give_up;
1221 /*
1222 * Determine if this is a delta and if so whether we can
1223 * reuse it or not. Otherwise let's find out as cheaply as
1224 * possible what the actual type and size for this object is.
1225 */
1226 switch (entry->in_pack_type) {
1227 default:
1228 /* Not a delta hence we've already got all we need. */
1229 entry->type = entry->in_pack_type;
1230 entry->in_pack_header_size = used;
1231 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1232 goto give_up;
1233 unuse_pack(&w_curs);
1234 return;
1235 case OBJ_REF_DELTA:
1236 if (reuse_delta && !entry->preferred_base)
1237 base_ref = use_pack(p, &w_curs,
1238 entry->in_pack_offset + used, NULL);
1239 entry->in_pack_header_size = used + 20;
1240 break;
1241 case OBJ_OFS_DELTA:
1242 buf = use_pack(p, &w_curs,
1243 entry->in_pack_offset + used, NULL);
1244 used_0 = 0;
1245 c = buf[used_0++];
1246 ofs = c & 127;
1247 while (c & 128) {
1248 ofs += 1;
1249 if (!ofs || MSB(ofs, 7)) {
1250 error("delta base offset overflow in pack for %s",
1251 sha1_to_hex(entry->idx.sha1));
1252 goto give_up;
1253 }
1254 c = buf[used_0++];
1255 ofs = (ofs << 7) + (c & 127);
1256 }
1257 ofs = entry->in_pack_offset - ofs;
1258 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1259 error("delta base offset out of bound for %s",
1260 sha1_to_hex(entry->idx.sha1));
1261 goto give_up;
1262 }
1263 if (reuse_delta && !entry->preferred_base) {
1264 struct revindex_entry *revidx;
1265 revidx = find_pack_revindex(p, ofs);
1266 if (!revidx)
1267 goto give_up;
1268 base_ref = nth_packed_object_sha1(p, revidx->nr);
1269 }
1270 entry->in_pack_header_size = used + used_0;
1271 break;
1272 }
1274 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1275 /*
1276 * If base_ref was set above that means we wish to
1277 * reuse delta data, and we even found that base
1278 * in the list of objects we want to pack. Goodie!
1279 *
1280 * Depth value does not matter - find_deltas() will
1281 * never consider reused delta as the base object to
1282 * deltify other objects against, in order to avoid
1283 * circular deltas.
1284 */
1285 entry->type = entry->in_pack_type;
1286 entry->delta = base_entry;
1287 entry->delta_size = entry->size;
1288 entry->delta_sibling = base_entry->delta_child;
1289 base_entry->delta_child = entry;
1290 unuse_pack(&w_curs);
1291 return;
1292 }
1294 if (entry->type) {
1295 /*
1296 * This must be a delta and we already know what the
1297 * final object type is. Let's extract the actual
1298 * object size from the delta header.
1299 */
1300 entry->size = get_size_from_delta(p, &w_curs,
1301 entry->in_pack_offset + entry->in_pack_header_size);
1302 if (entry->size == 0)
1303 goto give_up;
1304 unuse_pack(&w_curs);
1305 return;
1306 }
1308 /*
1309 * No choice but to fall back to the recursive delta walk
1310 * with sha1_object_info() to find about the object type
1311 * at this point...
1312 */
1313 give_up:
1314 unuse_pack(&w_curs);
1315 }
1317 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1318 /*
1319 * The error condition is checked in prepare_pack(). This is
1320 * to permit a missing preferred base object to be ignored
1321 * as a preferred base. Doing so can result in a larger
1322 * pack file, but the transfer will still take place.
1323 */
1324 }
1326 static int pack_offset_sort(const void *_a, const void *_b)
1327 {
1328 const struct object_entry *a = *(struct object_entry **)_a;
1329 const struct object_entry *b = *(struct object_entry **)_b;
1331 /* avoid filesystem trashing with loose objects */
1332 if (!a->in_pack && !b->in_pack)
1333 return hashcmp(a->idx.sha1, b->idx.sha1);
1335 if (a->in_pack < b->in_pack)
1336 return -1;
1337 if (a->in_pack > b->in_pack)
1338 return 1;
1339 return a->in_pack_offset < b->in_pack_offset ? -1 :
1340 (a->in_pack_offset > b->in_pack_offset);
1341 }
1343 static void get_object_details(void)
1344 {
1345 uint32_t i;
1346 struct object_entry **sorted_by_offset;
1348 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1349 for (i = 0; i < nr_objects; i++)
1350 sorted_by_offset[i] = objects + i;
1351 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1353 for (i = 0; i < nr_objects; i++) {
1354 struct object_entry *entry = sorted_by_offset[i];
1355 check_object(entry);
1356 if (big_file_threshold <= entry->size)
1357 entry->no_try_delta = 1;
1358 }
1360 free(sorted_by_offset);
1361 }
1363 /*
1364 * We search for deltas in a list sorted by type, by filename hash, and then
1365 * by size, so that we see progressively smaller and smaller files.
1366 * That's because we prefer deltas to be from the bigger file
1367 * to the smaller -- deletes are potentially cheaper, but perhaps
1368 * more importantly, the bigger file is likely the more recent
1369 * one. The deepest deltas are therefore the oldest objects which are
1370 * less susceptible to be accessed often.
1371 */
1372 static int type_size_sort(const void *_a, const void *_b)
1373 {
1374 const struct object_entry *a = *(struct object_entry **)_a;
1375 const struct object_entry *b = *(struct object_entry **)_b;
1377 if (a->type > b->type)
1378 return -1;
1379 if (a->type < b->type)
1380 return 1;
1381 if (a->hash > b->hash)
1382 return -1;
1383 if (a->hash < b->hash)
1384 return 1;
1385 if (a->preferred_base > b->preferred_base)
1386 return -1;
1387 if (a->preferred_base < b->preferred_base)
1388 return 1;
1389 if (a->size > b->size)
1390 return -1;
1391 if (a->size < b->size)
1392 return 1;
1393 return a < b ? -1 : (a > b); /* newest first */
1394 }
1396 struct unpacked {
1397 struct object_entry *entry;
1398 void *data;
1399 struct delta_index *index;
1400 unsigned depth;
1401 };
1403 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1404 unsigned long delta_size)
1405 {
1406 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1407 return 0;
1409 if (delta_size < cache_max_small_delta_size)
1410 return 1;
1412 /* cache delta, if objects are large enough compared to delta size */
1413 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1414 return 1;
1416 return 0;
1417 }
1419 #ifndef NO_PTHREADS
1421 static pthread_mutex_t read_mutex;
1422 #define read_lock() pthread_mutex_lock(&read_mutex)
1423 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1425 static pthread_mutex_t cache_mutex;
1426 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1427 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1429 static pthread_mutex_t progress_mutex;
1430 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1431 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1433 #else
1435 #define read_lock() (void)0
1436 #define read_unlock() (void)0
1437 #define cache_lock() (void)0
1438 #define cache_unlock() (void)0
1439 #define progress_lock() (void)0
1440 #define progress_unlock() (void)0
1442 #endif
1444 static int try_delta(struct unpacked *trg, struct unpacked *src,
1445 unsigned max_depth, unsigned long *mem_usage)
1446 {
1447 struct object_entry *trg_entry = trg->entry;
1448 struct object_entry *src_entry = src->entry;
1449 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1450 unsigned ref_depth;
1451 enum object_type type;
1452 void *delta_buf;
1454 /* Don't bother doing diffs between different types */
1455 if (trg_entry->type != src_entry->type)
1456 return -1;
1458 /*
1459 * We do not bother to try a delta that we discarded
1460 * on an earlier try, but only when reusing delta data.
1461 */
1462 if (reuse_delta && trg_entry->in_pack &&
1463 trg_entry->in_pack == src_entry->in_pack &&
1464 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1465 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1466 return 0;
1468 /* Let's not bust the allowed depth. */
1469 if (src->depth >= max_depth)
1470 return 0;
1472 /* Now some size filtering heuristics. */
1473 trg_size = trg_entry->size;
1474 if (!trg_entry->delta) {
1475 max_size = trg_size/2 - 20;
1476 ref_depth = 1;
1477 } else {
1478 max_size = trg_entry->delta_size;
1479 ref_depth = trg->depth;
1480 }
1481 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1482 (max_depth - ref_depth + 1);
1483 if (max_size == 0)
1484 return 0;
1485 src_size = src_entry->size;
1486 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1487 if (sizediff >= max_size)
1488 return 0;
1489 if (trg_size < src_size / 32)
1490 return 0;
1492 /* Load data if not already done */
1493 if (!trg->data) {
1494 read_lock();
1495 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1496 read_unlock();
1497 if (!trg->data)
1498 die("object %s cannot be read",
1499 sha1_to_hex(trg_entry->idx.sha1));
1500 if (sz != trg_size)
1501 die("object %s inconsistent object length (%lu vs %lu)",
1502 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1503 *mem_usage += sz;
1504 }
1505 if (!src->data) {
1506 read_lock();
1507 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1508 read_unlock();
1509 if (!src->data) {
1510 if (src_entry->preferred_base) {
1511 static int warned = 0;
1512 if (!warned++)
1513 warning("object %s cannot be read",
1514 sha1_to_hex(src_entry->idx.sha1));
1515 /*
1516 * Those objects are not included in the
1517 * resulting pack. Be resilient and ignore
1518 * them if they can't be read, in case the
1519 * pack could be created nevertheless.
1520 */
1521 return 0;
1522 }
1523 die("object %s cannot be read",
1524 sha1_to_hex(src_entry->idx.sha1));
1525 }
1526 if (sz != src_size)
1527 die("object %s inconsistent object length (%lu vs %lu)",
1528 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1529 *mem_usage += sz;
1530 }
1531 if (!src->index) {
1532 src->index = create_delta_index(src->data, src_size);
1533 if (!src->index) {
1534 static int warned = 0;
1535 if (!warned++)
1536 warning("suboptimal pack - out of memory");
1537 return 0;
1538 }
1539 *mem_usage += sizeof_delta_index(src->index);
1540 }
1542 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1543 if (!delta_buf)
1544 return 0;
1546 if (trg_entry->delta) {
1547 /* Prefer only shallower same-sized deltas. */
1548 if (delta_size == trg_entry->delta_size &&
1549 src->depth + 1 >= trg->depth) {
1550 free(delta_buf);
1551 return 0;
1552 }
1553 }
1555 /*
1556 * Handle memory allocation outside of the cache
1557 * accounting lock. Compiler will optimize the strangeness
1558 * away when NO_PTHREADS is defined.
1559 */
1560 free(trg_entry->delta_data);
1561 cache_lock();
1562 if (trg_entry->delta_data) {
1563 delta_cache_size -= trg_entry->delta_size;
1564 trg_entry->delta_data = NULL;
1565 }
1566 if (delta_cacheable(src_size, trg_size, delta_size)) {
1567 delta_cache_size += delta_size;
1568 cache_unlock();
1569 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1570 } else {
1571 cache_unlock();
1572 free(delta_buf);
1573 }
1575 trg_entry->delta = src_entry;
1576 trg_entry->delta_size = delta_size;
1577 trg->depth = src->depth + 1;
1579 return 1;
1580 }
1582 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1583 {
1584 struct object_entry *child = me->delta_child;
1585 unsigned int m = n;
1586 while (child) {
1587 unsigned int c = check_delta_limit(child, n + 1);
1588 if (m < c)
1589 m = c;
1590 child = child->delta_sibling;
1591 }
1592 return m;
1593 }
1595 static unsigned long free_unpacked(struct unpacked *n)
1596 {
1597 unsigned long freed_mem = sizeof_delta_index(n->index);
1598 free_delta_index(n->index);
1599 n->index = NULL;
1600 if (n->data) {
1601 freed_mem += n->entry->size;
1602 free(n->data);
1603 n->data = NULL;
1604 }
1605 n->entry = NULL;
1606 n->depth = 0;
1607 return freed_mem;
1608 }
1610 static void find_deltas(struct object_entry **list, unsigned *list_size,
1611 int window, int depth, unsigned *processed)
1612 {
1613 uint32_t i, idx = 0, count = 0;
1614 struct unpacked *array;
1615 unsigned long mem_usage = 0;
1617 array = xcalloc(window, sizeof(struct unpacked));
1619 for (;;) {
1620 struct object_entry *entry;
1621 struct unpacked *n = array + idx;
1622 int j, max_depth, best_base = -1;
1624 progress_lock();
1625 if (!*list_size) {
1626 progress_unlock();
1627 break;
1628 }
1629 entry = *list++;
1630 (*list_size)--;
1631 if (!entry->preferred_base) {
1632 (*processed)++;
1633 display_progress(progress_state, *processed);
1634 }
1635 progress_unlock();
1637 mem_usage -= free_unpacked(n);
1638 n->entry = entry;
1640 while (window_memory_limit &&
1641 mem_usage > window_memory_limit &&
1642 count > 1) {
1643 uint32_t tail = (idx + window - count) % window;
1644 mem_usage -= free_unpacked(array + tail);
1645 count--;
1646 }
1648 /* We do not compute delta to *create* objects we are not
1649 * going to pack.
1650 */
1651 if (entry->preferred_base)
1652 goto next;
1654 /*
1655 * If the current object is at pack edge, take the depth the
1656 * objects that depend on the current object into account
1657 * otherwise they would become too deep.
1658 */
1659 max_depth = depth;
1660 if (entry->delta_child) {
1661 max_depth -= check_delta_limit(entry, 0);
1662 if (max_depth <= 0)
1663 goto next;
1664 }
1666 j = window;
1667 while (--j > 0) {
1668 int ret;
1669 uint32_t other_idx = idx + j;
1670 struct unpacked *m;
1671 if (other_idx >= window)
1672 other_idx -= window;
1673 m = array + other_idx;
1674 if (!m->entry)
1675 break;
1676 ret = try_delta(n, m, max_depth, &mem_usage);
1677 if (ret < 0)
1678 break;
1679 else if (ret > 0)
1680 best_base = other_idx;
1681 }
1683 /*
1684 * If we decided to cache the delta data, then it is best
1685 * to compress it right away. First because we have to do
1686 * it anyway, and doing it here while we're threaded will
1687 * save a lot of time in the non threaded write phase,
1688 * as well as allow for caching more deltas within
1689 * the same cache size limit.
1690 * ...
1691 * But only if not writing to stdout, since in that case
1692 * the network is most likely throttling writes anyway,
1693 * and therefore it is best to go to the write phase ASAP
1694 * instead, as we can afford spending more time compressing
1695 * between writes at that moment.
1696 */
1697 if (entry->delta_data && !pack_to_stdout) {
1698 entry->z_delta_size = do_compress(&entry->delta_data,
1699 entry->delta_size);
1700 cache_lock();
1701 delta_cache_size -= entry->delta_size;
1702 delta_cache_size += entry->z_delta_size;
1703 cache_unlock();
1704 }
1706 /* if we made n a delta, and if n is already at max
1707 * depth, leaving it in the window is pointless. we
1708 * should evict it first.
1709 */
1710 if (entry->delta && max_depth <= n->depth)
1711 continue;
1713 /*
1714 * Move the best delta base up in the window, after the
1715 * currently deltified object, to keep it longer. It will
1716 * be the first base object to be attempted next.
1717 */
1718 if (entry->delta) {
1719 struct unpacked swap = array[best_base];
1720 int dist = (window + idx - best_base) % window;
1721 int dst = best_base;
1722 while (dist--) {
1723 int src = (dst + 1) % window;
1724 array[dst] = array[src];
1725 dst = src;
1726 }
1727 array[dst] = swap;
1728 }
1730 next:
1731 idx++;
1732 if (count + 1 < window)
1733 count++;
1734 if (idx >= window)
1735 idx = 0;
1736 }
1738 for (i = 0; i < window; ++i) {
1739 free_delta_index(array[i].index);
1740 free(array[i].data);
1741 }
1742 free(array);
1743 }
1745 #ifndef NO_PTHREADS
1747 static void try_to_free_from_threads(size_t size)
1748 {
1749 read_lock();
1750 release_pack_memory(size, -1);
1751 read_unlock();
1752 }
1754 static try_to_free_t old_try_to_free_routine;
1756 /*
1757 * The main thread waits on the condition that (at least) one of the workers
1758 * has stopped working (which is indicated in the .working member of
1759 * struct thread_params).
1760 * When a work thread has completed its work, it sets .working to 0 and
1761 * signals the main thread and waits on the condition that .data_ready
1762 * becomes 1.
1763 */
1765 struct thread_params {
1766 pthread_t thread;
1767 struct object_entry **list;
1768 unsigned list_size;
1769 unsigned remaining;
1770 int window;
1771 int depth;
1772 int working;
1773 int data_ready;
1774 pthread_mutex_t mutex;
1775 pthread_cond_t cond;
1776 unsigned *processed;
1777 };
1779 static pthread_cond_t progress_cond;
1781 /*
1782 * Mutex and conditional variable can't be statically-initialized on Windows.
1783 */
1784 static void init_threaded_search(void)
1785 {
1786 init_recursive_mutex(&read_mutex);
1787 pthread_mutex_init(&cache_mutex, NULL);
1788 pthread_mutex_init(&progress_mutex, NULL);
1789 pthread_cond_init(&progress_cond, NULL);
1790 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1791 }
1793 static void cleanup_threaded_search(void)
1794 {
1795 set_try_to_free_routine(old_try_to_free_routine);
1796 pthread_cond_destroy(&progress_cond);
1797 pthread_mutex_destroy(&read_mutex);
1798 pthread_mutex_destroy(&cache_mutex);
1799 pthread_mutex_destroy(&progress_mutex);
1800 }
1802 static void *threaded_find_deltas(void *arg)
1803 {
1804 struct thread_params *me = arg;
1806 while (me->remaining) {
1807 find_deltas(me->list, &me->remaining,
1808 me->window, me->depth, me->processed);
1810 progress_lock();
1811 me->working = 0;
1812 pthread_cond_signal(&progress_cond);
1813 progress_unlock();
1815 /*
1816 * We must not set ->data_ready before we wait on the
1817 * condition because the main thread may have set it to 1
1818 * before we get here. In order to be sure that new
1819 * work is available if we see 1 in ->data_ready, it
1820 * was initialized to 0 before this thread was spawned
1821 * and we reset it to 0 right away.
1822 */
1823 pthread_mutex_lock(&me->mutex);
1824 while (!me->data_ready)
1825 pthread_cond_wait(&me->cond, &me->mutex);
1826 me->data_ready = 0;
1827 pthread_mutex_unlock(&me->mutex);
1828 }
1829 /* leave ->working 1 so that this doesn't get more work assigned */
1830 return NULL;
1831 }
1833 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1834 int window, int depth, unsigned *processed)
1835 {
1836 struct thread_params *p;
1837 int i, ret, active_threads = 0;
1839 init_threaded_search();
1841 if (!delta_search_threads) /* --threads=0 means autodetect */
1842 delta_search_threads = online_cpus();
1843 if (delta_search_threads <= 1) {
1844 find_deltas(list, &list_size, window, depth, processed);
1845 cleanup_threaded_search();
1846 return;
1847 }
1848 if (progress > pack_to_stdout)
1849 fprintf(stderr, "Delta compression using up to %d threads.\n",
1850 delta_search_threads);
1851 p = xcalloc(delta_search_threads, sizeof(*p));
1853 /* Partition the work amongst work threads. */
1854 for (i = 0; i < delta_search_threads; i++) {
1855 unsigned sub_size = list_size / (delta_search_threads - i);
1857 /* don't use too small segments or no deltas will be found */
1858 if (sub_size < 2*window && i+1 < delta_search_threads)
1859 sub_size = 0;
1861 p[i].window = window;
1862 p[i].depth = depth;
1863 p[i].processed = processed;
1864 p[i].working = 1;
1865 p[i].data_ready = 0;
1867 /* try to split chunks on "path" boundaries */
1868 while (sub_size && sub_size < list_size &&
1869 list[sub_size]->hash &&
1870 list[sub_size]->hash == list[sub_size-1]->hash)
1871 sub_size++;
1873 p[i].list = list;
1874 p[i].list_size = sub_size;
1875 p[i].remaining = sub_size;
1877 list += sub_size;
1878 list_size -= sub_size;
1879 }
1881 /* Start work threads. */
1882 for (i = 0; i < delta_search_threads; i++) {
1883 if (!p[i].list_size)
1884 continue;
1885 pthread_mutex_init(&p[i].mutex, NULL);
1886 pthread_cond_init(&p[i].cond, NULL);
1887 ret = pthread_create(&p[i].thread, NULL,
1888 threaded_find_deltas, &p[i]);
1889 if (ret)
1890 die("unable to create thread: %s", strerror(ret));
1891 active_threads++;
1892 }
1894 /*
1895 * Now let's wait for work completion. Each time a thread is done
1896 * with its work, we steal half of the remaining work from the
1897 * thread with the largest number of unprocessed objects and give
1898 * it to that newly idle thread. This ensure good load balancing
1899 * until the remaining object list segments are simply too short
1900 * to be worth splitting anymore.
1901 */
1902 while (active_threads) {
1903 struct thread_params *target = NULL;
1904 struct thread_params *victim = NULL;
1905 unsigned sub_size = 0;
1907 progress_lock();
1908 for (;;) {
1909 for (i = 0; !target && i < delta_search_threads; i++)
1910 if (!p[i].working)
1911 target = &p[i];
1912 if (target)
1913 break;
1914 pthread_cond_wait(&progress_cond, &progress_mutex);
1915 }
1917 for (i = 0; i < delta_search_threads; i++)
1918 if (p[i].remaining > 2*window &&
1919 (!victim || victim->remaining < p[i].remaining))
1920 victim = &p[i];
1921 if (victim) {
1922 sub_size = victim->remaining / 2;
1923 list = victim->list + victim->list_size - sub_size;
1924 while (sub_size && list[0]->hash &&
1925 list[0]->hash == list[-1]->hash) {
1926 list++;
1927 sub_size--;
1928 }
1929 if (!sub_size) {
1930 /*
1931 * It is possible for some "paths" to have
1932 * so many objects that no hash boundary
1933 * might be found. Let's just steal the
1934 * exact half in that case.
1935 */
1936 sub_size = victim->remaining / 2;
1937 list -= sub_size;
1938 }
1939 target->list = list;
1940 victim->list_size -= sub_size;
1941 victim->remaining -= sub_size;
1942 }
1943 target->list_size = sub_size;
1944 target->remaining = sub_size;
1945 target->working = 1;
1946 progress_unlock();
1948 pthread_mutex_lock(&target->mutex);
1949 target->data_ready = 1;
1950 pthread_cond_signal(&target->cond);
1951 pthread_mutex_unlock(&target->mutex);
1953 if (!sub_size) {
1954 pthread_join(target->thread, NULL);
1955 pthread_cond_destroy(&target->cond);
1956 pthread_mutex_destroy(&target->mutex);
1957 active_threads--;
1958 }
1959 }
1960 cleanup_threaded_search();
1961 free(p);
1962 }
1964 #else
1965 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1966 #endif
1968 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1969 {
1970 unsigned char peeled[20];
1972 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1973 !peel_ref(path, peeled) && /* peelable? */
1974 !is_null_sha1(peeled) && /* annotated tag? */
1975 locate_object_entry(peeled)) /* object packed? */
1976 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1977 return 0;
1978 }
1980 static void prepare_pack(int window, int depth)
1981 {
1982 struct object_entry **delta_list;
1983 uint32_t i, nr_deltas;
1984 unsigned n;
1986 get_object_details();
1988 /*
1989 * If we're locally repacking then we need to be doubly careful
1990 * from now on in order to make sure no stealth corruption gets
1991 * propagated to the new pack. Clients receiving streamed packs
1992 * should validate everything they get anyway so no need to incur
1993 * the additional cost here in that case.
1994 */
1995 if (!pack_to_stdout)
1996 do_check_packed_object_crc = 1;
1998 if (!nr_objects || !window || !depth)
1999 return;
2001 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
2002 nr_deltas = n = 0;
2004 for (i = 0; i < nr_objects; i++) {
2005 struct object_entry *entry = objects + i;
2007 if (entry->delta)
2008 /* This happens if we decided to reuse existing
2009 * delta from a pack. "reuse_delta &&" is implied.
2010 */
2011 continue;
2013 if (entry->size < 50)
2014 continue;
2016 if (entry->no_try_delta)
2017 continue;
2019 if (!entry->preferred_base) {
2020 nr_deltas++;
2021 if (entry->type < 0)
2022 die("unable to get type of object %s",
2023 sha1_to_hex(entry->idx.sha1));
2024 } else {
2025 if (entry->type < 0) {
2026 /*
2027 * This object is not found, but we
2028 * don't have to include it anyway.
2029 */
2030 continue;
2031 }
2032 }
2034 delta_list[n++] = entry;
2035 }
2037 if (nr_deltas && n > 1) {
2038 unsigned nr_done = 0;
2039 if (progress)
2040 progress_state = start_progress("Compressing objects",
2041 nr_deltas);
2042 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2043 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2044 stop_progress(&progress_state);
2045 if (nr_done != nr_deltas)
2046 die("inconsistency with delta count");
2047 }
2048 free(delta_list);
2049 }
2051 static int git_pack_config(const char *k, const char *v, void *cb)
2052 {
2053 if (!strcmp(k, "pack.window")) {
2054 window = git_config_int(k, v);
2055 return 0;
2056 }
2057 if (!strcmp(k, "pack.windowmemory")) {
2058 window_memory_limit = git_config_ulong(k, v);
2059 return 0;
2060 }
2061 if (!strcmp(k, "pack.depth")) {
2062 depth = git_config_int(k, v);
2063 return 0;
2064 }
2065 if (!strcmp(k, "pack.compression")) {
2066 int level = git_config_int(k, v);
2067 if (level == -1)
2068 level = Z_DEFAULT_COMPRESSION;
2069 else if (level < 0 || level > Z_BEST_COMPRESSION)
2070 die("bad pack compression level %d", level);
2071 pack_compression_level = level;
2072 pack_compression_seen = 1;
2073 return 0;
2074 }
2075 if (!strcmp(k, "pack.deltacachesize")) {
2076 max_delta_cache_size = git_config_int(k, v);
2077 return 0;
2078 }
2079 if (!strcmp(k, "pack.deltacachelimit")) {
2080 cache_max_small_delta_size = git_config_int(k, v);
2081 return 0;
2082 }
2083 if (!strcmp(k, "pack.threads")) {
2084 delta_search_threads = git_config_int(k, v);
2085 if (delta_search_threads < 0)
2086 die("invalid number of threads specified (%d)",
2087 delta_search_threads);
2088 #ifdef NO_PTHREADS
2089 if (delta_search_threads != 1)
2090 warning("no threads support, ignoring %s", k);
2091 #endif
2092 return 0;
2093 }
2094 if (!strcmp(k, "pack.indexversion")) {
2095 pack_idx_opts.version = git_config_int(k, v);
2096 if (pack_idx_opts.version > 2)
2097 die("bad pack.indexversion=%"PRIu32,
2098 pack_idx_opts.version);
2099 return 0;
2100 }
2101 if (!strcmp(k, "pack.packsizelimit")) {
2102 pack_size_limit_cfg = git_config_ulong(k, v);
2103 return 0;
2104 }
2105 return git_default_config(k, v, cb);
2106 }
2108 static void read_object_list_from_stdin(void)
2109 {
2110 char line[40 + 1 + PATH_MAX + 2];
2111 unsigned char sha1[20];
2113 for (;;) {
2114 if (!fgets(line, sizeof(line), stdin)) {
2115 if (feof(stdin))
2116 break;
2117 if (!ferror(stdin))
2118 die("fgets returned NULL, not EOF, not error!");
2119 if (errno != EINTR)
2120 die_errno("fgets");
2121 clearerr(stdin);
2122 continue;
2123 }
2124 if (line[0] == '-') {
2125 if (get_sha1_hex(line+1, sha1))
2126 die("expected edge sha1, got garbage:\n %s",
2127 line);
2128 add_preferred_base(sha1);
2129 continue;
2130 }
2131 if (get_sha1_hex(line, sha1))
2132 die("expected sha1, got garbage:\n %s", line);
2134 add_preferred_base_object(line+41);
2135 add_object_entry(sha1, 0, line+41, 0);
2136 }
2137 }
2139 #define OBJECT_ADDED (1u<<20)
2141 static void show_commit(struct commit *commit, void *data)
2142 {
2143 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2144 commit->object.flags |= OBJECT_ADDED;
2145 }
2147 static void show_object(struct object *obj,
2148 const struct name_path *path, const char *last,
2149 void *data)
2150 {
2151 char *name = path_name(path, last);
2153 add_preferred_base_object(name);
2154 add_object_entry(obj->sha1, obj->type, name, 0);
2155 obj->flags |= OBJECT_ADDED;
2157 /*
2158 * We will have generated the hash from the name,
2159 * but not saved a pointer to it - we can free it
2160 */
2161 free((char *)name);
2162 }
2164 static void show_edge(struct commit *commit)
2165 {
2166 add_preferred_base(commit->object.sha1);
2167 }
2169 struct in_pack_object {
2170 off_t offset;
2171 struct object *object;
2172 };
2174 struct in_pack {
2175 int alloc;
2176 int nr;
2177 struct in_pack_object *array;
2178 };
2180 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2181 {
2182 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2183 in_pack->array[in_pack->nr].object = object;
2184 in_pack->nr++;
2185 }
2187 /*
2188 * Compare the objects in the offset order, in order to emulate the
2189 * "git rev-list --objects" output that produced the pack originally.
2190 */
2191 static int ofscmp(const void *a_, const void *b_)
2192 {
2193 struct in_pack_object *a = (struct in_pack_object *)a_;
2194 struct in_pack_object *b = (struct in_pack_object *)b_;
2196 if (a->offset < b->offset)
2197 return -1;
2198 else if (a->offset > b->offset)
2199 return 1;
2200 else
2201 return hashcmp(a->object->sha1, b->object->sha1);
2202 }
2204 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2205 {
2206 struct packed_git *p;
2207 struct in_pack in_pack;
2208 uint32_t i;
2210 memset(&in_pack, 0, sizeof(in_pack));
2212 for (p = packed_git; p; p = p->next) {
2213 const unsigned char *sha1;
2214 struct object *o;
2216 if (!p->pack_local || p->pack_keep)
2217 continue;
2218 if (open_pack_index(p))
2219 die("cannot open pack index");
2221 ALLOC_GROW(in_pack.array,
2222 in_pack.nr + p->num_objects,
2223 in_pack.alloc);
2225 for (i = 0; i < p->num_objects; i++) {
2226 sha1 = nth_packed_object_sha1(p, i);
2227 o = lookup_unknown_object(sha1);
2228 if (!(o->flags & OBJECT_ADDED))
2229 mark_in_pack_object(o, p, &in_pack);
2230 o->flags |= OBJECT_ADDED;
2231 }
2232 }
2234 if (in_pack.nr) {
2235 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2236 ofscmp);
2237 for (i = 0; i < in_pack.nr; i++) {
2238 struct object *o = in_pack.array[i].object;
2239 add_object_entry(o->sha1, o->type, "", 0);
2240 }
2241 }
2242 free(in_pack.array);
2243 }
2245 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2246 {
2247 static struct packed_git *last_found = (void *)1;
2248 struct packed_git *p;
2250 p = (last_found != (void *)1) ? last_found : packed_git;
2252 while (p) {
2253 if ((!p->pack_local || p->pack_keep) &&
2254 find_pack_entry_one(sha1, p)) {
2255 last_found = p;
2256 return 1;
2257 }
2258 if (p == last_found)
2259 p = packed_git;
2260 else
2261 p = p->next;
2262 if (p == last_found)
2263 p = p->next;
2264 }
2265 return 0;
2266 }
2268 static void loosen_unused_packed_objects(struct rev_info *revs)
2269 {
2270 struct packed_git *p;
2271 uint32_t i;
2272 const unsigned char *sha1;
2274 for (p = packed_git; p; p = p->next) {
2275 if (!p->pack_local || p->pack_keep)
2276 continue;
2278 if (open_pack_index(p))
2279 die("cannot open pack index");
2281 for (i = 0; i < p->num_objects; i++) {
2282 sha1 = nth_packed_object_sha1(p, i);
2283 if (!locate_object_entry(sha1) &&
2284 !has_sha1_pack_kept_or_nonlocal(sha1))
2285 if (force_object_loose(sha1, p->mtime))
2286 die("unable to force loose object");
2287 }
2288 }
2289 }
2291 static void get_object_list(int ac, const char **av)
2292 {
2293 struct rev_info revs;
2294 char line[1000];
2295 int flags = 0;
2297 init_revisions(&revs, NULL);
2298 save_commit_buffer = 0;
2299 setup_revisions(ac, av, &revs, NULL);
2301 while (fgets(line, sizeof(line), stdin) != NULL) {
2302 int len = strlen(line);
2303 if (len && line[len - 1] == '\n')
2304 line[--len] = 0;
2305 if (!len)
2306 break;
2307 if (*line == '-') {
2308 if (!strcmp(line, "--not")) {
2309 flags ^= UNINTERESTING;
2310 continue;
2311 }
2312 die("not a rev '%s'", line);
2313 }
2314 if (handle_revision_arg(line, &revs, flags, 1))
2315 die("bad revision '%s'", line);
2316 }
2318 if (prepare_revision_walk(&revs))
2319 die("revision walk setup failed");
2320 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2321 traverse_commit_list(&revs, show_commit, show_object, NULL);
2323 if (keep_unreachable)
2324 add_objects_in_unpacked_packs(&revs);
2325 if (unpack_unreachable)
2326 loosen_unused_packed_objects(&revs);
2327 }
2329 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2330 {
2331 int use_internal_rev_list = 0;
2332 int thin = 0;
2333 int all_progress_implied = 0;
2334 uint32_t i;
2335 const char **rp_av;
2336 int rp_ac_alloc = 64;
2337 int rp_ac;
2339 read_replace_refs = 0;
2341 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2343 rp_av[0] = "pack-objects";
2344 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2345 rp_ac = 2;
2347 reset_pack_idx_option(&pack_idx_opts);
2348 git_config(git_pack_config, NULL);
2349 if (!pack_compression_seen && core_compression_seen)
2350 pack_compression_level = core_compression_level;
2352 progress = isatty(2);
2353 for (i = 1; i < argc; i++) {
2354 const char *arg = argv[i];
2356 if (*arg != '-')
2357 break;
2359 if (!strcmp("--non-empty", arg)) {
2360 non_empty = 1;
2361 continue;
2362 }
2363 if (!strcmp("--local", arg)) {
2364 local = 1;
2365 continue;
2366 }
2367 if (!strcmp("--incremental", arg)) {
2368 incremental = 1;
2369 continue;
2370 }
2371 if (!strcmp("--honor-pack-keep", arg)) {
2372 ignore_packed_keep = 1;
2373 continue;
2374 }
2375 if (!prefixcmp(arg, "--compression=")) {
2376 char *end;
2377 int level = strtoul(arg+14, &end, 0);
2378 if (!arg[14] || *end)
2379 usage(pack_usage);
2380 if (level == -1)
2381 level = Z_DEFAULT_COMPRESSION;
2382 else if (level < 0 || level > Z_BEST_COMPRESSION)
2383 die("bad pack compression level %d", level);
2384 pack_compression_level = level;
2385 continue;
2386 }
2387 if (!prefixcmp(arg, "--max-pack-size=")) {
2388 pack_size_limit_cfg = 0;
2389 if (!git_parse_ulong(arg+16, &pack_size_limit))
2390 usage(pack_usage);
2391 continue;
2392 }
2393 if (!prefixcmp(arg, "--window=")) {
2394 char *end;
2395 window = strtoul(arg+9, &end, 0);
2396 if (!arg[9] || *end)
2397 usage(pack_usage);
2398 continue;
2399 }
2400 if (!prefixcmp(arg, "--window-memory=")) {
2401 if (!git_parse_ulong(arg+16, &window_memory_limit))
2402 usage(pack_usage);
2403 continue;
2404 }
2405 if (!prefixcmp(arg, "--threads=")) {
2406 char *end;
2407 delta_search_threads = strtoul(arg+10, &end, 0);
2408 if (!arg[10] || *end || delta_search_threads < 0)
2409 usage(pack_usage);
2410 #ifdef NO_PTHREADS
2411 if (delta_search_threads != 1)
2412 warning("no threads support, "
2413 "ignoring %s", arg);
2414 #endif
2415 continue;
2416 }
2417 if (!prefixcmp(arg, "--depth=")) {
2418 char *end;
2419 depth = strtoul(arg+8, &end, 0);
2420 if (!arg[8] || *end)
2421 usage(pack_usage);
2422 continue;
2423 }
2424 if (!strcmp("--progress", arg)) {
2425 progress = 1;
2426 continue;
2427 }
2428 if (!strcmp("--all-progress", arg)) {
2429 progress = 2;
2430 continue;
2431 }
2432 if (!strcmp("--all-progress-implied", arg)) {
2433 all_progress_implied = 1;
2434 continue;
2435 }
2436 if (!strcmp("-q", arg)) {
2437 progress = 0;
2438 continue;
2439 }
2440 if (!strcmp("--no-reuse-delta", arg)) {
2441 reuse_delta = 0;
2442 continue;
2443 }
2444 if (!strcmp("--no-reuse-object", arg)) {
2445 reuse_object = reuse_delta = 0;
2446 continue;
2447 }
2448 if (!strcmp("--delta-base-offset", arg)) {
2449 allow_ofs_delta = 1;
2450 continue;
2451 }
2452 if (!strcmp("--stdout", arg)) {
2453 pack_to_stdout = 1;
2454 continue;
2455 }
2456 if (!strcmp("--revs", arg)) {
2457 use_internal_rev_list = 1;
2458 continue;
2459 }
2460 if (!strcmp("--keep-unreachable", arg)) {
2461 keep_unreachable = 1;
2462 continue;
2463 }
2464 if (!strcmp("--unpack-unreachable", arg)) {
2465 unpack_unreachable = 1;
2466 continue;
2467 }
2468 if (!strcmp("--include-tag", arg)) {
2469 include_tag = 1;
2470 continue;
2471 }
2472 if (!strcmp("--unpacked", arg) ||
2473 !strcmp("--reflog", arg) ||
2474 !strcmp("--all", arg)) {
2475 use_internal_rev_list = 1;
2476 if (rp_ac >= rp_ac_alloc - 1) {
2477 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2478 rp_av = xrealloc(rp_av,
2479 rp_ac_alloc * sizeof(*rp_av));
2480 }
2481 rp_av[rp_ac++] = arg;
2482 continue;
2483 }
2484 if (!strcmp("--thin", arg)) {
2485 use_internal_rev_list = 1;
2486 thin = 1;
2487 rp_av[1] = "--objects-edge";
2488 continue;
2489 }
2490 if (!prefixcmp(arg, "--index-version=")) {
2491 char *c;
2492 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2493 if (pack_idx_opts.version > 2)
2494 die("bad %s", arg);
2495 if (*c == ',')
2496 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2497 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2498 die("bad %s", arg);
2499 continue;
2500 }
2501 if (!strcmp(arg, "--keep-true-parents")) {
2502 grafts_replace_parents = 0;
2503 continue;
2504 }
2505 usage(pack_usage);
2506 }
2508 /* Traditionally "pack-objects [options] base extra" failed;
2509 * we would however want to take refs parameter that would
2510 * have been given to upstream rev-list ourselves, which means
2511 * we somehow want to say what the base name is. So the
2512 * syntax would be:
2513 *
2514 * pack-objects [options] base <refs...>
2515 *
2516 * in other words, we would treat the first non-option as the
2517 * base_name and send everything else to the internal revision
2518 * walker.
2519 */
2521 if (!pack_to_stdout)
2522 base_name = argv[i++];
2524 if (pack_to_stdout != !base_name)
2525 usage(pack_usage);
2527 if (!pack_to_stdout && !pack_size_limit)
2528 pack_size_limit = pack_size_limit_cfg;
2529 if (pack_to_stdout && pack_size_limit)
2530 die("--max-pack-size cannot be used to build a pack for transfer.");
2531 if (pack_size_limit && pack_size_limit < 1024*1024) {
2532 warning("minimum pack size limit is 1 MiB");
2533 pack_size_limit = 1024*1024;
2534 }
2536 if (!pack_to_stdout && thin)
2537 die("--thin cannot be used to build an indexable pack.");
2539 if (keep_unreachable && unpack_unreachable)
2540 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2542 if (progress && all_progress_implied)
2543 progress = 2;
2545 prepare_packed_git();
2547 if (progress)
2548 progress_state = start_progress("Counting objects", 0);
2549 if (!use_internal_rev_list)
2550 read_object_list_from_stdin();
2551 else {
2552 rp_av[rp_ac] = NULL;
2553 get_object_list(rp_ac, rp_av);
2554 }
2555 cleanup_preferred_base();
2556 if (include_tag && nr_result)
2557 for_each_ref(add_ref_tag, NULL);
2558 stop_progress(&progress_state);
2560 if (non_empty && !nr_result)
2561 return 0;
2562 if (nr_result)
2563 prepare_pack(window, depth);
2564 write_pack_file();
2565 if (progress)
2566 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2567 " reused %"PRIu32" (delta %"PRIu32")\n",
2568 written, written_delta, reused, reused_delta);
2569 return 0;
2570 }