![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
638d43da0e61c67a374ee124faf9391180171b21
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 static int write_one(struct sha1file *f,
413 struct object_entry *e,
414 off_t *offset)
415 {
416 unsigned long size;
418 /* offset is non zero if object is written already. */
419 if (e->idx.offset || e->preferred_base)
420 return -1;
422 /* if we are deltified, write out base object first. */
423 if (e->delta && !write_one(f, e->delta, offset))
424 return 0;
426 e->idx.offset = *offset;
427 size = write_object(f, e, *offset);
428 if (!size) {
429 e->idx.offset = 0;
430 return 0;
431 }
432 written_list[nr_written++] = &e->idx;
434 /* make sure off_t is sufficiently large not to wrap */
435 if (signed_add_overflows(*offset, size))
436 die("pack too large for current definition of off_t");
437 *offset += size;
438 return 1;
439 }
441 static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
442 void *cb_data)
443 {
444 unsigned char peeled[20];
445 struct object_entry *entry = locate_object_entry(sha1);
447 if (entry)
448 entry->tagged = 1;
449 if (!peel_ref(path, peeled)) {
450 entry = locate_object_entry(peeled);
451 if (entry)
452 entry->tagged = 1;
453 }
454 return 0;
455 }
457 static inline void add_to_write_order(struct object_entry **wo,
458 unsigned int *endp,
459 struct object_entry *e)
460 {
461 if (e->filled)
462 return;
463 wo[(*endp)++] = e;
464 e->filled = 1;
465 }
467 static void add_descendants_to_write_order(struct object_entry **wo,
468 unsigned int *endp,
469 struct object_entry *e)
470 {
471 int add_to_order = 1;
472 while (e) {
473 if (add_to_order) {
474 struct object_entry *s;
475 /* add this node... */
476 add_to_write_order(wo, endp, e);
477 /* all its siblings... */
478 for (s = e->delta_sibling; s; s = s->delta_sibling) {
479 add_to_write_order(wo, endp, s);
480 }
481 }
482 /* drop down a level to add left subtree nodes if possible */
483 if (e->delta_child) {
484 add_to_order = 1;
485 e = e->delta_child;
486 } else {
487 add_to_order = 0;
488 /* our sibling might have some children, it is next */
489 if (e->delta_sibling) {
490 e = e->delta_sibling;
491 continue;
492 }
493 /* go back to our parent node */
494 e = e->delta;
495 while (e && !e->delta_sibling) {
496 /* we're on the right side of a subtree, keep
497 * going up until we can go right again */
498 e = e->delta;
499 }
500 if (!e) {
501 /* done- we hit our original root node */
502 return;
503 }
504 /* pass it off to sibling at this level */
505 e = e->delta_sibling;
506 }
507 };
508 }
510 static void add_family_to_write_order(struct object_entry **wo,
511 unsigned int *endp,
512 struct object_entry *e)
513 {
514 struct object_entry *root;
516 for (root = e; root->delta; root = root->delta)
517 ; /* nothing */
518 add_descendants_to_write_order(wo, endp, root);
519 }
521 static struct object_entry **compute_write_order(void)
522 {
523 unsigned int i, wo_end, last_untagged;
525 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));
527 for (i = 0; i < nr_objects; i++) {
528 objects[i].tagged = 0;
529 objects[i].filled = 0;
530 objects[i].delta_child = NULL;
531 objects[i].delta_sibling = NULL;
532 }
534 /*
535 * Fully connect delta_child/delta_sibling network.
536 * Make sure delta_sibling is sorted in the original
537 * recency order.
538 */
539 for (i = nr_objects; i > 0;) {
540 struct object_entry *e = &objects[--i];
541 if (!e->delta)
542 continue;
543 /* Mark me as the first child */
544 e->delta_sibling = e->delta->delta_child;
545 e->delta->delta_child = e;
546 }
548 /*
549 * Mark objects that are at the tip of tags.
550 */
551 for_each_tag_ref(mark_tagged, NULL);
553 /*
554 * Give the objects in the original recency order until
555 * we see a tagged tip.
556 */
557 for (i = wo_end = 0; i < nr_objects; i++) {
558 if (objects[i].tagged)
559 break;
560 add_to_write_order(wo, &wo_end, &objects[i]);
561 }
562 last_untagged = i;
564 /*
565 * Then fill all the tagged tips.
566 */
567 for (; i < nr_objects; i++) {
568 if (objects[i].tagged)
569 add_to_write_order(wo, &wo_end, &objects[i]);
570 }
572 /*
573 * And then all remaining commits and tags.
574 */
575 for (i = last_untagged; i < nr_objects; i++) {
576 if (objects[i].type != OBJ_COMMIT &&
577 objects[i].type != OBJ_TAG)
578 continue;
579 add_to_write_order(wo, &wo_end, &objects[i]);
580 }
582 /*
583 * And then all the trees.
584 */
585 for (i = last_untagged; i < nr_objects; i++) {
586 if (objects[i].type != OBJ_TREE)
587 continue;
588 add_to_write_order(wo, &wo_end, &objects[i]);
589 }
591 /*
592 * Finally all the rest in really tight order
593 */
594 for (i = last_untagged; i < nr_objects; i++) {
595 if (!objects[i].filled)
596 add_family_to_write_order(wo, &wo_end, &objects[i]);
597 }
599 if (wo_end != nr_objects)
600 die("ordered %u objects, expected %"PRIu32, wo_end, nr_objects);
602 return wo;
603 }
605 static void write_pack_file(void)
606 {
607 uint32_t i = 0, j;
608 struct sha1file *f;
609 off_t offset;
610 struct pack_header hdr;
611 uint32_t nr_remaining = nr_result;
612 time_t last_mtime = 0;
613 struct object_entry **write_order;
615 if (progress > pack_to_stdout)
616 progress_state = start_progress("Writing objects", nr_result);
617 written_list = xmalloc(nr_objects * sizeof(*written_list));
618 write_order = compute_write_order();
620 do {
621 unsigned char sha1[20];
622 char *pack_tmp_name = NULL;
624 if (pack_to_stdout) {
625 f = sha1fd_throughput(1, "<stdout>", progress_state);
626 } else {
627 char tmpname[PATH_MAX];
628 int fd;
629 fd = odb_mkstemp(tmpname, sizeof(tmpname),
630 "pack/tmp_pack_XXXXXX");
631 pack_tmp_name = xstrdup(tmpname);
632 f = sha1fd(fd, pack_tmp_name);
633 }
635 hdr.hdr_signature = htonl(PACK_SIGNATURE);
636 hdr.hdr_version = htonl(PACK_VERSION);
637 hdr.hdr_entries = htonl(nr_remaining);
638 sha1write(f, &hdr, sizeof(hdr));
639 offset = sizeof(hdr);
640 nr_written = 0;
641 for (; i < nr_objects; i++) {
642 struct object_entry *e = write_order[i];
643 if (!write_one(f, e, &offset))
644 break;
645 display_progress(progress_state, written);
646 }
648 /*
649 * Did we write the wrong # entries in the header?
650 * If so, rewrite it like in fast-import
651 */
652 if (pack_to_stdout) {
653 sha1close(f, sha1, CSUM_CLOSE);
654 } else if (nr_written == nr_remaining) {
655 sha1close(f, sha1, CSUM_FSYNC);
656 } else {
657 int fd = sha1close(f, sha1, 0);
658 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
659 nr_written, sha1, offset);
660 close(fd);
661 }
663 if (!pack_to_stdout) {
664 struct stat st;
665 const char *idx_tmp_name;
666 char tmpname[PATH_MAX];
668 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
669 &pack_idx_opts, sha1);
671 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
672 base_name, sha1_to_hex(sha1));
673 free_pack_by_name(tmpname);
674 if (adjust_shared_perm(pack_tmp_name))
675 die_errno("unable to make temporary pack file readable");
676 if (rename(pack_tmp_name, tmpname))
677 die_errno("unable to rename temporary pack file");
679 /*
680 * Packs are runtime accessed in their mtime
681 * order since newer packs are more likely to contain
682 * younger objects. So if we are creating multiple
683 * packs then we should modify the mtime of later ones
684 * to preserve this property.
685 */
686 if (stat(tmpname, &st) < 0) {
687 warning("failed to stat %s: %s",
688 tmpname, strerror(errno));
689 } else if (!last_mtime) {
690 last_mtime = st.st_mtime;
691 } else {
692 struct utimbuf utb;
693 utb.actime = st.st_atime;
694 utb.modtime = --last_mtime;
695 if (utime(tmpname, &utb) < 0)
696 warning("failed utime() on %s: %s",
697 tmpname, strerror(errno));
698 }
700 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
701 base_name, sha1_to_hex(sha1));
702 if (adjust_shared_perm(idx_tmp_name))
703 die_errno("unable to make temporary index file readable");
704 if (rename(idx_tmp_name, tmpname))
705 die_errno("unable to rename temporary index file");
707 free((void *) idx_tmp_name);
708 free(pack_tmp_name);
709 puts(sha1_to_hex(sha1));
710 }
712 /* mark written objects as written to previous pack */
713 for (j = 0; j < nr_written; j++) {
714 written_list[j]->offset = (off_t)-1;
715 }
716 nr_remaining -= nr_written;
717 } while (nr_remaining && i < nr_objects);
719 free(written_list);
720 free(write_order);
721 stop_progress(&progress_state);
722 if (written != nr_result)
723 die("wrote %"PRIu32" objects while expecting %"PRIu32,
724 written, nr_result);
725 }
727 static int locate_object_entry_hash(const unsigned char *sha1)
728 {
729 int i;
730 unsigned int ui;
731 memcpy(&ui, sha1, sizeof(unsigned int));
732 i = ui % object_ix_hashsz;
733 while (0 < object_ix[i]) {
734 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
735 return i;
736 if (++i == object_ix_hashsz)
737 i = 0;
738 }
739 return -1 - i;
740 }
742 static struct object_entry *locate_object_entry(const unsigned char *sha1)
743 {
744 int i;
746 if (!object_ix_hashsz)
747 return NULL;
749 i = locate_object_entry_hash(sha1);
750 if (0 <= i)
751 return &objects[object_ix[i]-1];
752 return NULL;
753 }
755 static void rehash_objects(void)
756 {
757 uint32_t i;
758 struct object_entry *oe;
760 object_ix_hashsz = nr_objects * 3;
761 if (object_ix_hashsz < 1024)
762 object_ix_hashsz = 1024;
763 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
764 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
765 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
766 int ix = locate_object_entry_hash(oe->idx.sha1);
767 if (0 <= ix)
768 continue;
769 ix = -1 - ix;
770 object_ix[ix] = i + 1;
771 }
772 }
774 static unsigned name_hash(const char *name)
775 {
776 unsigned c, hash = 0;
778 if (!name)
779 return 0;
781 /*
782 * This effectively just creates a sortable number from the
783 * last sixteen non-whitespace characters. Last characters
784 * count "most", so things that end in ".c" sort together.
785 */
786 while ((c = *name++) != 0) {
787 if (isspace(c))
788 continue;
789 hash = (hash >> 2) + (c << 24);
790 }
791 return hash;
792 }
794 static void setup_delta_attr_check(struct git_attr_check *check)
795 {
796 static struct git_attr *attr_delta;
798 if (!attr_delta)
799 attr_delta = git_attr("delta");
801 check[0].attr = attr_delta;
802 }
804 static int no_try_delta(const char *path)
805 {
806 struct git_attr_check check[1];
808 setup_delta_attr_check(check);
809 if (git_check_attr(path, ARRAY_SIZE(check), check))
810 return 0;
811 if (ATTR_FALSE(check->value))
812 return 1;
813 return 0;
814 }
816 static int add_object_entry(const unsigned char *sha1, enum object_type type,
817 const char *name, int exclude)
818 {
819 struct object_entry *entry;
820 struct packed_git *p, *found_pack = NULL;
821 off_t found_offset = 0;
822 int ix;
823 unsigned hash = name_hash(name);
825 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
826 if (ix >= 0) {
827 if (exclude) {
828 entry = objects + object_ix[ix] - 1;
829 if (!entry->preferred_base)
830 nr_result--;
831 entry->preferred_base = 1;
832 }
833 return 0;
834 }
836 if (!exclude && local && has_loose_object_nonlocal(sha1))
837 return 0;
839 for (p = packed_git; p; p = p->next) {
840 off_t offset = find_pack_entry_one(sha1, p);
841 if (offset) {
842 if (!found_pack) {
843 found_offset = offset;
844 found_pack = p;
845 }
846 if (exclude)
847 break;
848 if (incremental)
849 return 0;
850 if (local && !p->pack_local)
851 return 0;
852 if (ignore_packed_keep && p->pack_local && p->pack_keep)
853 return 0;
854 }
855 }
857 if (nr_objects >= nr_alloc) {
858 nr_alloc = (nr_alloc + 1024) * 3 / 2;
859 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
860 }
862 entry = objects + nr_objects++;
863 memset(entry, 0, sizeof(*entry));
864 hashcpy(entry->idx.sha1, sha1);
865 entry->hash = hash;
866 if (type)
867 entry->type = type;
868 if (exclude)
869 entry->preferred_base = 1;
870 else
871 nr_result++;
872 if (found_pack) {
873 entry->in_pack = found_pack;
874 entry->in_pack_offset = found_offset;
875 }
877 if (object_ix_hashsz * 3 <= nr_objects * 4)
878 rehash_objects();
879 else
880 object_ix[-1 - ix] = nr_objects;
882 display_progress(progress_state, nr_objects);
884 if (name && no_try_delta(name))
885 entry->no_try_delta = 1;
887 return 1;
888 }
890 struct pbase_tree_cache {
891 unsigned char sha1[20];
892 int ref;
893 int temporary;
894 void *tree_data;
895 unsigned long tree_size;
896 };
898 static struct pbase_tree_cache *(pbase_tree_cache[256]);
899 static int pbase_tree_cache_ix(const unsigned char *sha1)
900 {
901 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
902 }
903 static int pbase_tree_cache_ix_incr(int ix)
904 {
905 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
906 }
908 static struct pbase_tree {
909 struct pbase_tree *next;
910 /* This is a phony "cache" entry; we are not
911 * going to evict it nor find it through _get()
912 * mechanism -- this is for the toplevel node that
913 * would almost always change with any commit.
914 */
915 struct pbase_tree_cache pcache;
916 } *pbase_tree;
918 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
919 {
920 struct pbase_tree_cache *ent, *nent;
921 void *data;
922 unsigned long size;
923 enum object_type type;
924 int neigh;
925 int my_ix = pbase_tree_cache_ix(sha1);
926 int available_ix = -1;
928 /* pbase-tree-cache acts as a limited hashtable.
929 * your object will be found at your index or within a few
930 * slots after that slot if it is cached.
931 */
932 for (neigh = 0; neigh < 8; neigh++) {
933 ent = pbase_tree_cache[my_ix];
934 if (ent && !hashcmp(ent->sha1, sha1)) {
935 ent->ref++;
936 return ent;
937 }
938 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
939 ((0 <= available_ix) &&
940 (!ent && pbase_tree_cache[available_ix])))
941 available_ix = my_ix;
942 if (!ent)
943 break;
944 my_ix = pbase_tree_cache_ix_incr(my_ix);
945 }
947 /* Did not find one. Either we got a bogus request or
948 * we need to read and perhaps cache.
949 */
950 data = read_sha1_file(sha1, &type, &size);
951 if (!data)
952 return NULL;
953 if (type != OBJ_TREE) {
954 free(data);
955 return NULL;
956 }
958 /* We need to either cache or return a throwaway copy */
960 if (available_ix < 0)
961 ent = NULL;
962 else {
963 ent = pbase_tree_cache[available_ix];
964 my_ix = available_ix;
965 }
967 if (!ent) {
968 nent = xmalloc(sizeof(*nent));
969 nent->temporary = (available_ix < 0);
970 }
971 else {
972 /* evict and reuse */
973 free(ent->tree_data);
974 nent = ent;
975 }
976 hashcpy(nent->sha1, sha1);
977 nent->tree_data = data;
978 nent->tree_size = size;
979 nent->ref = 1;
980 if (!nent->temporary)
981 pbase_tree_cache[my_ix] = nent;
982 return nent;
983 }
985 static void pbase_tree_put(struct pbase_tree_cache *cache)
986 {
987 if (!cache->temporary) {
988 cache->ref--;
989 return;
990 }
991 free(cache->tree_data);
992 free(cache);
993 }
995 static int name_cmp_len(const char *name)
996 {
997 int i;
998 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
999 ;
1000 return i;
1001 }
1003 static void add_pbase_object(struct tree_desc *tree,
1004 const char *name,
1005 int cmplen,
1006 const char *fullname)
1007 {
1008 struct name_entry entry;
1009 int cmp;
1011 while (tree_entry(tree,&entry)) {
1012 if (S_ISGITLINK(entry.mode))
1013 continue;
1014 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1015 memcmp(name, entry.path, cmplen);
1016 if (cmp > 0)
1017 continue;
1018 if (cmp < 0)
1019 return;
1020 if (name[cmplen] != '/') {
1021 add_object_entry(entry.sha1,
1022 object_type(entry.mode),
1023 fullname, 1);
1024 return;
1025 }
1026 if (S_ISDIR(entry.mode)) {
1027 struct tree_desc sub;
1028 struct pbase_tree_cache *tree;
1029 const char *down = name+cmplen+1;
1030 int downlen = name_cmp_len(down);
1032 tree = pbase_tree_get(entry.sha1);
1033 if (!tree)
1034 return;
1035 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1037 add_pbase_object(&sub, down, downlen, fullname);
1038 pbase_tree_put(tree);
1039 }
1040 }
1041 }
1043 static unsigned *done_pbase_paths;
1044 static int done_pbase_paths_num;
1045 static int done_pbase_paths_alloc;
1046 static int done_pbase_path_pos(unsigned hash)
1047 {
1048 int lo = 0;
1049 int hi = done_pbase_paths_num;
1050 while (lo < hi) {
1051 int mi = (hi + lo) / 2;
1052 if (done_pbase_paths[mi] == hash)
1053 return mi;
1054 if (done_pbase_paths[mi] < hash)
1055 hi = mi;
1056 else
1057 lo = mi + 1;
1058 }
1059 return -lo-1;
1060 }
1062 static int check_pbase_path(unsigned hash)
1063 {
1064 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1065 if (0 <= pos)
1066 return 1;
1067 pos = -pos - 1;
1068 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1069 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1070 done_pbase_paths = xrealloc(done_pbase_paths,
1071 done_pbase_paths_alloc *
1072 sizeof(unsigned));
1073 }
1074 done_pbase_paths_num++;
1075 if (pos < done_pbase_paths_num)
1076 memmove(done_pbase_paths + pos + 1,
1077 done_pbase_paths + pos,
1078 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1079 done_pbase_paths[pos] = hash;
1080 return 0;
1081 }
1083 static void add_preferred_base_object(const char *name)
1084 {
1085 struct pbase_tree *it;
1086 int cmplen;
1087 unsigned hash = name_hash(name);
1089 if (!num_preferred_base || check_pbase_path(hash))
1090 return;
1092 cmplen = name_cmp_len(name);
1093 for (it = pbase_tree; it; it = it->next) {
1094 if (cmplen == 0) {
1095 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1096 }
1097 else {
1098 struct tree_desc tree;
1099 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1100 add_pbase_object(&tree, name, cmplen, name);
1101 }
1102 }
1103 }
1105 static void add_preferred_base(unsigned char *sha1)
1106 {
1107 struct pbase_tree *it;
1108 void *data;
1109 unsigned long size;
1110 unsigned char tree_sha1[20];
1112 if (window <= num_preferred_base++)
1113 return;
1115 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1116 if (!data)
1117 return;
1119 for (it = pbase_tree; it; it = it->next) {
1120 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1121 free(data);
1122 return;
1123 }
1124 }
1126 it = xcalloc(1, sizeof(*it));
1127 it->next = pbase_tree;
1128 pbase_tree = it;
1130 hashcpy(it->pcache.sha1, tree_sha1);
1131 it->pcache.tree_data = data;
1132 it->pcache.tree_size = size;
1133 }
1135 static void cleanup_preferred_base(void)
1136 {
1137 struct pbase_tree *it;
1138 unsigned i;
1140 it = pbase_tree;
1141 pbase_tree = NULL;
1142 while (it) {
1143 struct pbase_tree *this = it;
1144 it = this->next;
1145 free(this->pcache.tree_data);
1146 free(this);
1147 }
1149 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1150 if (!pbase_tree_cache[i])
1151 continue;
1152 free(pbase_tree_cache[i]->tree_data);
1153 free(pbase_tree_cache[i]);
1154 pbase_tree_cache[i] = NULL;
1155 }
1157 free(done_pbase_paths);
1158 done_pbase_paths = NULL;
1159 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1160 }
1162 static void check_object(struct object_entry *entry)
1163 {
1164 if (entry->in_pack) {
1165 struct packed_git *p = entry->in_pack;
1166 struct pack_window *w_curs = NULL;
1167 const unsigned char *base_ref = NULL;
1168 struct object_entry *base_entry;
1169 unsigned long used, used_0;
1170 unsigned long avail;
1171 off_t ofs;
1172 unsigned char *buf, c;
1174 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1176 /*
1177 * We want in_pack_type even if we do not reuse delta
1178 * since non-delta representations could still be reused.
1179 */
1180 used = unpack_object_header_buffer(buf, avail,
1181 &entry->in_pack_type,
1182 &entry->size);
1183 if (used == 0)
1184 goto give_up;
1186 /*
1187 * Determine if this is a delta and if so whether we can
1188 * reuse it or not. Otherwise let's find out as cheaply as
1189 * possible what the actual type and size for this object is.
1190 */
1191 switch (entry->in_pack_type) {
1192 default:
1193 /* Not a delta hence we've already got all we need. */
1194 entry->type = entry->in_pack_type;
1195 entry->in_pack_header_size = used;
1196 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1197 goto give_up;
1198 unuse_pack(&w_curs);
1199 return;
1200 case OBJ_REF_DELTA:
1201 if (reuse_delta && !entry->preferred_base)
1202 base_ref = use_pack(p, &w_curs,
1203 entry->in_pack_offset + used, NULL);
1204 entry->in_pack_header_size = used + 20;
1205 break;
1206 case OBJ_OFS_DELTA:
1207 buf = use_pack(p, &w_curs,
1208 entry->in_pack_offset + used, NULL);
1209 used_0 = 0;
1210 c = buf[used_0++];
1211 ofs = c & 127;
1212 while (c & 128) {
1213 ofs += 1;
1214 if (!ofs || MSB(ofs, 7)) {
1215 error("delta base offset overflow in pack for %s",
1216 sha1_to_hex(entry->idx.sha1));
1217 goto give_up;
1218 }
1219 c = buf[used_0++];
1220 ofs = (ofs << 7) + (c & 127);
1221 }
1222 ofs = entry->in_pack_offset - ofs;
1223 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1224 error("delta base offset out of bound for %s",
1225 sha1_to_hex(entry->idx.sha1));
1226 goto give_up;
1227 }
1228 if (reuse_delta && !entry->preferred_base) {
1229 struct revindex_entry *revidx;
1230 revidx = find_pack_revindex(p, ofs);
1231 if (!revidx)
1232 goto give_up;
1233 base_ref = nth_packed_object_sha1(p, revidx->nr);
1234 }
1235 entry->in_pack_header_size = used + used_0;
1236 break;
1237 }
1239 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1240 /*
1241 * If base_ref was set above that means we wish to
1242 * reuse delta data, and we even found that base
1243 * in the list of objects we want to pack. Goodie!
1244 *
1245 * Depth value does not matter - find_deltas() will
1246 * never consider reused delta as the base object to
1247 * deltify other objects against, in order to avoid
1248 * circular deltas.
1249 */
1250 entry->type = entry->in_pack_type;
1251 entry->delta = base_entry;
1252 entry->delta_size = entry->size;
1253 entry->delta_sibling = base_entry->delta_child;
1254 base_entry->delta_child = entry;
1255 unuse_pack(&w_curs);
1256 return;
1257 }
1259 if (entry->type) {
1260 /*
1261 * This must be a delta and we already know what the
1262 * final object type is. Let's extract the actual
1263 * object size from the delta header.
1264 */
1265 entry->size = get_size_from_delta(p, &w_curs,
1266 entry->in_pack_offset + entry->in_pack_header_size);
1267 if (entry->size == 0)
1268 goto give_up;
1269 unuse_pack(&w_curs);
1270 return;
1271 }
1273 /*
1274 * No choice but to fall back to the recursive delta walk
1275 * with sha1_object_info() to find about the object type
1276 * at this point...
1277 */
1278 give_up:
1279 unuse_pack(&w_curs);
1280 }
1282 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1283 /*
1284 * The error condition is checked in prepare_pack(). This is
1285 * to permit a missing preferred base object to be ignored
1286 * as a preferred base. Doing so can result in a larger
1287 * pack file, but the transfer will still take place.
1288 */
1289 }
1291 static int pack_offset_sort(const void *_a, const void *_b)
1292 {
1293 const struct object_entry *a = *(struct object_entry **)_a;
1294 const struct object_entry *b = *(struct object_entry **)_b;
1296 /* avoid filesystem trashing with loose objects */
1297 if (!a->in_pack && !b->in_pack)
1298 return hashcmp(a->idx.sha1, b->idx.sha1);
1300 if (a->in_pack < b->in_pack)
1301 return -1;
1302 if (a->in_pack > b->in_pack)
1303 return 1;
1304 return a->in_pack_offset < b->in_pack_offset ? -1 :
1305 (a->in_pack_offset > b->in_pack_offset);
1306 }
1308 static void get_object_details(void)
1309 {
1310 uint32_t i;
1311 struct object_entry **sorted_by_offset;
1313 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1314 for (i = 0; i < nr_objects; i++)
1315 sorted_by_offset[i] = objects + i;
1316 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1318 for (i = 0; i < nr_objects; i++) {
1319 struct object_entry *entry = sorted_by_offset[i];
1320 check_object(entry);
1321 if (big_file_threshold <= entry->size)
1322 entry->no_try_delta = 1;
1323 }
1325 free(sorted_by_offset);
1326 }
1328 /*
1329 * We search for deltas in a list sorted by type, by filename hash, and then
1330 * by size, so that we see progressively smaller and smaller files.
1331 * That's because we prefer deltas to be from the bigger file
1332 * to the smaller -- deletes are potentially cheaper, but perhaps
1333 * more importantly, the bigger file is likely the more recent
1334 * one. The deepest deltas are therefore the oldest objects which are
1335 * less susceptible to be accessed often.
1336 */
1337 static int type_size_sort(const void *_a, const void *_b)
1338 {
1339 const struct object_entry *a = *(struct object_entry **)_a;
1340 const struct object_entry *b = *(struct object_entry **)_b;
1342 if (a->type > b->type)
1343 return -1;
1344 if (a->type < b->type)
1345 return 1;
1346 if (a->hash > b->hash)
1347 return -1;
1348 if (a->hash < b->hash)
1349 return 1;
1350 if (a->preferred_base > b->preferred_base)
1351 return -1;
1352 if (a->preferred_base < b->preferred_base)
1353 return 1;
1354 if (a->size > b->size)
1355 return -1;
1356 if (a->size < b->size)
1357 return 1;
1358 return a < b ? -1 : (a > b); /* newest first */
1359 }
1361 struct unpacked {
1362 struct object_entry *entry;
1363 void *data;
1364 struct delta_index *index;
1365 unsigned depth;
1366 };
1368 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1369 unsigned long delta_size)
1370 {
1371 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1372 return 0;
1374 if (delta_size < cache_max_small_delta_size)
1375 return 1;
1377 /* cache delta, if objects are large enough compared to delta size */
1378 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1379 return 1;
1381 return 0;
1382 }
1384 #ifndef NO_PTHREADS
1386 static pthread_mutex_t read_mutex;
1387 #define read_lock() pthread_mutex_lock(&read_mutex)
1388 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1390 static pthread_mutex_t cache_mutex;
1391 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1392 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1394 static pthread_mutex_t progress_mutex;
1395 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1396 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1398 #else
1400 #define read_lock() (void)0
1401 #define read_unlock() (void)0
1402 #define cache_lock() (void)0
1403 #define cache_unlock() (void)0
1404 #define progress_lock() (void)0
1405 #define progress_unlock() (void)0
1407 #endif
1409 static int try_delta(struct unpacked *trg, struct unpacked *src,
1410 unsigned max_depth, unsigned long *mem_usage)
1411 {
1412 struct object_entry *trg_entry = trg->entry;
1413 struct object_entry *src_entry = src->entry;
1414 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1415 unsigned ref_depth;
1416 enum object_type type;
1417 void *delta_buf;
1419 /* Don't bother doing diffs between different types */
1420 if (trg_entry->type != src_entry->type)
1421 return -1;
1423 /*
1424 * We do not bother to try a delta that we discarded on an
1425 * earlier try, but only when reusing delta data. Note that
1426 * src_entry that is marked as the preferred_base should always
1427 * be considered, as even if we produce a suboptimal delta against
1428 * it, we will still save the transfer cost, as we already know
1429 * the other side has it and we won't send src_entry at all.
1430 */
1431 if (reuse_delta && trg_entry->in_pack &&
1432 trg_entry->in_pack == src_entry->in_pack &&
1433 !src_entry->preferred_base &&
1434 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1435 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1436 return 0;
1438 /* Let's not bust the allowed depth. */
1439 if (src->depth >= max_depth)
1440 return 0;
1442 /* Now some size filtering heuristics. */
1443 trg_size = trg_entry->size;
1444 if (!trg_entry->delta) {
1445 max_size = trg_size/2 - 20;
1446 ref_depth = 1;
1447 } else {
1448 max_size = trg_entry->delta_size;
1449 ref_depth = trg->depth;
1450 }
1451 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1452 (max_depth - ref_depth + 1);
1453 if (max_size == 0)
1454 return 0;
1455 src_size = src_entry->size;
1456 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1457 if (sizediff >= max_size)
1458 return 0;
1459 if (trg_size < src_size / 32)
1460 return 0;
1462 /* Load data if not already done */
1463 if (!trg->data) {
1464 read_lock();
1465 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1466 read_unlock();
1467 if (!trg->data)
1468 die("object %s cannot be read",
1469 sha1_to_hex(trg_entry->idx.sha1));
1470 if (sz != trg_size)
1471 die("object %s inconsistent object length (%lu vs %lu)",
1472 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1473 *mem_usage += sz;
1474 }
1475 if (!src->data) {
1476 read_lock();
1477 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1478 read_unlock();
1479 if (!src->data) {
1480 if (src_entry->preferred_base) {
1481 static int warned = 0;
1482 if (!warned++)
1483 warning("object %s cannot be read",
1484 sha1_to_hex(src_entry->idx.sha1));
1485 /*
1486 * Those objects are not included in the
1487 * resulting pack. Be resilient and ignore
1488 * them if they can't be read, in case the
1489 * pack could be created nevertheless.
1490 */
1491 return 0;
1492 }
1493 die("object %s cannot be read",
1494 sha1_to_hex(src_entry->idx.sha1));
1495 }
1496 if (sz != src_size)
1497 die("object %s inconsistent object length (%lu vs %lu)",
1498 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1499 *mem_usage += sz;
1500 }
1501 if (!src->index) {
1502 src->index = create_delta_index(src->data, src_size);
1503 if (!src->index) {
1504 static int warned = 0;
1505 if (!warned++)
1506 warning("suboptimal pack - out of memory");
1507 return 0;
1508 }
1509 *mem_usage += sizeof_delta_index(src->index);
1510 }
1512 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1513 if (!delta_buf)
1514 return 0;
1516 if (trg_entry->delta) {
1517 /* Prefer only shallower same-sized deltas. */
1518 if (delta_size == trg_entry->delta_size &&
1519 src->depth + 1 >= trg->depth) {
1520 free(delta_buf);
1521 return 0;
1522 }
1523 }
1525 /*
1526 * Handle memory allocation outside of the cache
1527 * accounting lock. Compiler will optimize the strangeness
1528 * away when NO_PTHREADS is defined.
1529 */
1530 free(trg_entry->delta_data);
1531 cache_lock();
1532 if (trg_entry->delta_data) {
1533 delta_cache_size -= trg_entry->delta_size;
1534 trg_entry->delta_data = NULL;
1535 }
1536 if (delta_cacheable(src_size, trg_size, delta_size)) {
1537 delta_cache_size += delta_size;
1538 cache_unlock();
1539 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1540 } else {
1541 cache_unlock();
1542 free(delta_buf);
1543 }
1545 trg_entry->delta = src_entry;
1546 trg_entry->delta_size = delta_size;
1547 trg->depth = src->depth + 1;
1549 return 1;
1550 }
1552 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1553 {
1554 struct object_entry *child = me->delta_child;
1555 unsigned int m = n;
1556 while (child) {
1557 unsigned int c = check_delta_limit(child, n + 1);
1558 if (m < c)
1559 m = c;
1560 child = child->delta_sibling;
1561 }
1562 return m;
1563 }
1565 static unsigned long free_unpacked(struct unpacked *n)
1566 {
1567 unsigned long freed_mem = sizeof_delta_index(n->index);
1568 free_delta_index(n->index);
1569 n->index = NULL;
1570 if (n->data) {
1571 freed_mem += n->entry->size;
1572 free(n->data);
1573 n->data = NULL;
1574 }
1575 n->entry = NULL;
1576 n->depth = 0;
1577 return freed_mem;
1578 }
1580 static void find_deltas(struct object_entry **list, unsigned *list_size,
1581 int window, int depth, unsigned *processed)
1582 {
1583 uint32_t i, idx = 0, count = 0;
1584 struct unpacked *array;
1585 unsigned long mem_usage = 0;
1587 array = xcalloc(window, sizeof(struct unpacked));
1589 for (;;) {
1590 struct object_entry *entry;
1591 struct unpacked *n = array + idx;
1592 int j, max_depth, best_base = -1;
1594 progress_lock();
1595 if (!*list_size) {
1596 progress_unlock();
1597 break;
1598 }
1599 entry = *list++;
1600 (*list_size)--;
1601 if (!entry->preferred_base) {
1602 (*processed)++;
1603 display_progress(progress_state, *processed);
1604 }
1605 progress_unlock();
1607 mem_usage -= free_unpacked(n);
1608 n->entry = entry;
1610 while (window_memory_limit &&
1611 mem_usage > window_memory_limit &&
1612 count > 1) {
1613 uint32_t tail = (idx + window - count) % window;
1614 mem_usage -= free_unpacked(array + tail);
1615 count--;
1616 }
1618 /* We do not compute delta to *create* objects we are not
1619 * going to pack.
1620 */
1621 if (entry->preferred_base)
1622 goto next;
1624 /*
1625 * If the current object is at pack edge, take the depth the
1626 * objects that depend on the current object into account
1627 * otherwise they would become too deep.
1628 */
1629 max_depth = depth;
1630 if (entry->delta_child) {
1631 max_depth -= check_delta_limit(entry, 0);
1632 if (max_depth <= 0)
1633 goto next;
1634 }
1636 j = window;
1637 while (--j > 0) {
1638 int ret;
1639 uint32_t other_idx = idx + j;
1640 struct unpacked *m;
1641 if (other_idx >= window)
1642 other_idx -= window;
1643 m = array + other_idx;
1644 if (!m->entry)
1645 break;
1646 ret = try_delta(n, m, max_depth, &mem_usage);
1647 if (ret < 0)
1648 break;
1649 else if (ret > 0)
1650 best_base = other_idx;
1651 }
1653 /*
1654 * If we decided to cache the delta data, then it is best
1655 * to compress it right away. First because we have to do
1656 * it anyway, and doing it here while we're threaded will
1657 * save a lot of time in the non threaded write phase,
1658 * as well as allow for caching more deltas within
1659 * the same cache size limit.
1660 * ...
1661 * But only if not writing to stdout, since in that case
1662 * the network is most likely throttling writes anyway,
1663 * and therefore it is best to go to the write phase ASAP
1664 * instead, as we can afford spending more time compressing
1665 * between writes at that moment.
1666 */
1667 if (entry->delta_data && !pack_to_stdout) {
1668 entry->z_delta_size = do_compress(&entry->delta_data,
1669 entry->delta_size);
1670 cache_lock();
1671 delta_cache_size -= entry->delta_size;
1672 delta_cache_size += entry->z_delta_size;
1673 cache_unlock();
1674 }
1676 /* if we made n a delta, and if n is already at max
1677 * depth, leaving it in the window is pointless. we
1678 * should evict it first.
1679 */
1680 if (entry->delta && max_depth <= n->depth)
1681 continue;
1683 /*
1684 * Move the best delta base up in the window, after the
1685 * currently deltified object, to keep it longer. It will
1686 * be the first base object to be attempted next.
1687 */
1688 if (entry->delta) {
1689 struct unpacked swap = array[best_base];
1690 int dist = (window + idx - best_base) % window;
1691 int dst = best_base;
1692 while (dist--) {
1693 int src = (dst + 1) % window;
1694 array[dst] = array[src];
1695 dst = src;
1696 }
1697 array[dst] = swap;
1698 }
1700 next:
1701 idx++;
1702 if (count + 1 < window)
1703 count++;
1704 if (idx >= window)
1705 idx = 0;
1706 }
1708 for (i = 0; i < window; ++i) {
1709 free_delta_index(array[i].index);
1710 free(array[i].data);
1711 }
1712 free(array);
1713 }
1715 #ifndef NO_PTHREADS
1717 static void try_to_free_from_threads(size_t size)
1718 {
1719 read_lock();
1720 release_pack_memory(size, -1);
1721 read_unlock();
1722 }
1724 static try_to_free_t old_try_to_free_routine;
1726 /*
1727 * The main thread waits on the condition that (at least) one of the workers
1728 * has stopped working (which is indicated in the .working member of
1729 * struct thread_params).
1730 * When a work thread has completed its work, it sets .working to 0 and
1731 * signals the main thread and waits on the condition that .data_ready
1732 * becomes 1.
1733 */
1735 struct thread_params {
1736 pthread_t thread;
1737 struct object_entry **list;
1738 unsigned list_size;
1739 unsigned remaining;
1740 int window;
1741 int depth;
1742 int working;
1743 int data_ready;
1744 pthread_mutex_t mutex;
1745 pthread_cond_t cond;
1746 unsigned *processed;
1747 };
1749 static pthread_cond_t progress_cond;
1751 /*
1752 * Mutex and conditional variable can't be statically-initialized on Windows.
1753 */
1754 static void init_threaded_search(void)
1755 {
1756 init_recursive_mutex(&read_mutex);
1757 pthread_mutex_init(&cache_mutex, NULL);
1758 pthread_mutex_init(&progress_mutex, NULL);
1759 pthread_cond_init(&progress_cond, NULL);
1760 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1761 }
1763 static void cleanup_threaded_search(void)
1764 {
1765 set_try_to_free_routine(old_try_to_free_routine);
1766 pthread_cond_destroy(&progress_cond);
1767 pthread_mutex_destroy(&read_mutex);
1768 pthread_mutex_destroy(&cache_mutex);
1769 pthread_mutex_destroy(&progress_mutex);
1770 }
1772 static void *threaded_find_deltas(void *arg)
1773 {
1774 struct thread_params *me = arg;
1776 while (me->remaining) {
1777 find_deltas(me->list, &me->remaining,
1778 me->window, me->depth, me->processed);
1780 progress_lock();
1781 me->working = 0;
1782 pthread_cond_signal(&progress_cond);
1783 progress_unlock();
1785 /*
1786 * We must not set ->data_ready before we wait on the
1787 * condition because the main thread may have set it to 1
1788 * before we get here. In order to be sure that new
1789 * work is available if we see 1 in ->data_ready, it
1790 * was initialized to 0 before this thread was spawned
1791 * and we reset it to 0 right away.
1792 */
1793 pthread_mutex_lock(&me->mutex);
1794 while (!me->data_ready)
1795 pthread_cond_wait(&me->cond, &me->mutex);
1796 me->data_ready = 0;
1797 pthread_mutex_unlock(&me->mutex);
1798 }
1799 /* leave ->working 1 so that this doesn't get more work assigned */
1800 return NULL;
1801 }
1803 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1804 int window, int depth, unsigned *processed)
1805 {
1806 struct thread_params *p;
1807 int i, ret, active_threads = 0;
1809 init_threaded_search();
1811 if (!delta_search_threads) /* --threads=0 means autodetect */
1812 delta_search_threads = online_cpus();
1813 if (delta_search_threads <= 1) {
1814 find_deltas(list, &list_size, window, depth, processed);
1815 cleanup_threaded_search();
1816 return;
1817 }
1818 if (progress > pack_to_stdout)
1819 fprintf(stderr, "Delta compression using up to %d threads.\n",
1820 delta_search_threads);
1821 p = xcalloc(delta_search_threads, sizeof(*p));
1823 /* Partition the work amongst work threads. */
1824 for (i = 0; i < delta_search_threads; i++) {
1825 unsigned sub_size = list_size / (delta_search_threads - i);
1827 /* don't use too small segments or no deltas will be found */
1828 if (sub_size < 2*window && i+1 < delta_search_threads)
1829 sub_size = 0;
1831 p[i].window = window;
1832 p[i].depth = depth;
1833 p[i].processed = processed;
1834 p[i].working = 1;
1835 p[i].data_ready = 0;
1837 /* try to split chunks on "path" boundaries */
1838 while (sub_size && sub_size < list_size &&
1839 list[sub_size]->hash &&
1840 list[sub_size]->hash == list[sub_size-1]->hash)
1841 sub_size++;
1843 p[i].list = list;
1844 p[i].list_size = sub_size;
1845 p[i].remaining = sub_size;
1847 list += sub_size;
1848 list_size -= sub_size;
1849 }
1851 /* Start work threads. */
1852 for (i = 0; i < delta_search_threads; i++) {
1853 if (!p[i].list_size)
1854 continue;
1855 pthread_mutex_init(&p[i].mutex, NULL);
1856 pthread_cond_init(&p[i].cond, NULL);
1857 ret = pthread_create(&p[i].thread, NULL,
1858 threaded_find_deltas, &p[i]);
1859 if (ret)
1860 die("unable to create thread: %s", strerror(ret));
1861 active_threads++;
1862 }
1864 /*
1865 * Now let's wait for work completion. Each time a thread is done
1866 * with its work, we steal half of the remaining work from the
1867 * thread with the largest number of unprocessed objects and give
1868 * it to that newly idle thread. This ensure good load balancing
1869 * until the remaining object list segments are simply too short
1870 * to be worth splitting anymore.
1871 */
1872 while (active_threads) {
1873 struct thread_params *target = NULL;
1874 struct thread_params *victim = NULL;
1875 unsigned sub_size = 0;
1877 progress_lock();
1878 for (;;) {
1879 for (i = 0; !target && i < delta_search_threads; i++)
1880 if (!p[i].working)
1881 target = &p[i];
1882 if (target)
1883 break;
1884 pthread_cond_wait(&progress_cond, &progress_mutex);
1885 }
1887 for (i = 0; i < delta_search_threads; i++)
1888 if (p[i].remaining > 2*window &&
1889 (!victim || victim->remaining < p[i].remaining))
1890 victim = &p[i];
1891 if (victim) {
1892 sub_size = victim->remaining / 2;
1893 list = victim->list + victim->list_size - sub_size;
1894 while (sub_size && list[0]->hash &&
1895 list[0]->hash == list[-1]->hash) {
1896 list++;
1897 sub_size--;
1898 }
1899 if (!sub_size) {
1900 /*
1901 * It is possible for some "paths" to have
1902 * so many objects that no hash boundary
1903 * might be found. Let's just steal the
1904 * exact half in that case.
1905 */
1906 sub_size = victim->remaining / 2;
1907 list -= sub_size;
1908 }
1909 target->list = list;
1910 victim->list_size -= sub_size;
1911 victim->remaining -= sub_size;
1912 }
1913 target->list_size = sub_size;
1914 target->remaining = sub_size;
1915 target->working = 1;
1916 progress_unlock();
1918 pthread_mutex_lock(&target->mutex);
1919 target->data_ready = 1;
1920 pthread_cond_signal(&target->cond);
1921 pthread_mutex_unlock(&target->mutex);
1923 if (!sub_size) {
1924 pthread_join(target->thread, NULL);
1925 pthread_cond_destroy(&target->cond);
1926 pthread_mutex_destroy(&target->mutex);
1927 active_threads--;
1928 }
1929 }
1930 cleanup_threaded_search();
1931 free(p);
1932 }
1934 #else
1935 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1936 #endif
1938 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1939 {
1940 unsigned char peeled[20];
1942 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1943 !peel_ref(path, peeled) && /* peelable? */
1944 !is_null_sha1(peeled) && /* annotated tag? */
1945 locate_object_entry(peeled)) /* object packed? */
1946 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1947 return 0;
1948 }
1950 static void prepare_pack(int window, int depth)
1951 {
1952 struct object_entry **delta_list;
1953 uint32_t i, nr_deltas;
1954 unsigned n;
1956 get_object_details();
1958 /*
1959 * If we're locally repacking then we need to be doubly careful
1960 * from now on in order to make sure no stealth corruption gets
1961 * propagated to the new pack. Clients receiving streamed packs
1962 * should validate everything they get anyway so no need to incur
1963 * the additional cost here in that case.
1964 */
1965 if (!pack_to_stdout)
1966 do_check_packed_object_crc = 1;
1968 if (!nr_objects || !window || !depth)
1969 return;
1971 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1972 nr_deltas = n = 0;
1974 for (i = 0; i < nr_objects; i++) {
1975 struct object_entry *entry = objects + i;
1977 if (entry->delta)
1978 /* This happens if we decided to reuse existing
1979 * delta from a pack. "reuse_delta &&" is implied.
1980 */
1981 continue;
1983 if (entry->size < 50)
1984 continue;
1986 if (entry->no_try_delta)
1987 continue;
1989 if (!entry->preferred_base) {
1990 nr_deltas++;
1991 if (entry->type < 0)
1992 die("unable to get type of object %s",
1993 sha1_to_hex(entry->idx.sha1));
1994 } else {
1995 if (entry->type < 0) {
1996 /*
1997 * This object is not found, but we
1998 * don't have to include it anyway.
1999 */
2000 continue;
2001 }
2002 }
2004 delta_list[n++] = entry;
2005 }
2007 if (nr_deltas && n > 1) {
2008 unsigned nr_done = 0;
2009 if (progress)
2010 progress_state = start_progress("Compressing objects",
2011 nr_deltas);
2012 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2013 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2014 stop_progress(&progress_state);
2015 if (nr_done != nr_deltas)
2016 die("inconsistency with delta count");
2017 }
2018 free(delta_list);
2019 }
2021 static int git_pack_config(const char *k, const char *v, void *cb)
2022 {
2023 if (!strcmp(k, "pack.window")) {
2024 window = git_config_int(k, v);
2025 return 0;
2026 }
2027 if (!strcmp(k, "pack.windowmemory")) {
2028 window_memory_limit = git_config_ulong(k, v);
2029 return 0;
2030 }
2031 if (!strcmp(k, "pack.depth")) {
2032 depth = git_config_int(k, v);
2033 return 0;
2034 }
2035 if (!strcmp(k, "pack.compression")) {
2036 int level = git_config_int(k, v);
2037 if (level == -1)
2038 level = Z_DEFAULT_COMPRESSION;
2039 else if (level < 0 || level > Z_BEST_COMPRESSION)
2040 die("bad pack compression level %d", level);
2041 pack_compression_level = level;
2042 pack_compression_seen = 1;
2043 return 0;
2044 }
2045 if (!strcmp(k, "pack.deltacachesize")) {
2046 max_delta_cache_size = git_config_int(k, v);
2047 return 0;
2048 }
2049 if (!strcmp(k, "pack.deltacachelimit")) {
2050 cache_max_small_delta_size = git_config_int(k, v);
2051 return 0;
2052 }
2053 if (!strcmp(k, "pack.threads")) {
2054 delta_search_threads = git_config_int(k, v);
2055 if (delta_search_threads < 0)
2056 die("invalid number of threads specified (%d)",
2057 delta_search_threads);
2058 #ifdef NO_PTHREADS
2059 if (delta_search_threads != 1)
2060 warning("no threads support, ignoring %s", k);
2061 #endif
2062 return 0;
2063 }
2064 if (!strcmp(k, "pack.indexversion")) {
2065 pack_idx_opts.version = git_config_int(k, v);
2066 if (pack_idx_opts.version > 2)
2067 die("bad pack.indexversion=%"PRIu32,
2068 pack_idx_opts.version);
2069 return 0;
2070 }
2071 if (!strcmp(k, "pack.packsizelimit")) {
2072 pack_size_limit_cfg = git_config_ulong(k, v);
2073 return 0;
2074 }
2075 return git_default_config(k, v, cb);
2076 }
2078 static void read_object_list_from_stdin(void)
2079 {
2080 char line[40 + 1 + PATH_MAX + 2];
2081 unsigned char sha1[20];
2083 for (;;) {
2084 if (!fgets(line, sizeof(line), stdin)) {
2085 if (feof(stdin))
2086 break;
2087 if (!ferror(stdin))
2088 die("fgets returned NULL, not EOF, not error!");
2089 if (errno != EINTR)
2090 die_errno("fgets");
2091 clearerr(stdin);
2092 continue;
2093 }
2094 if (line[0] == '-') {
2095 if (get_sha1_hex(line+1, sha1))
2096 die("expected edge sha1, got garbage:\n %s",
2097 line);
2098 add_preferred_base(sha1);
2099 continue;
2100 }
2101 if (get_sha1_hex(line, sha1))
2102 die("expected sha1, got garbage:\n %s", line);
2104 add_preferred_base_object(line+41);
2105 add_object_entry(sha1, 0, line+41, 0);
2106 }
2107 }
2109 #define OBJECT_ADDED (1u<<20)
2111 static void show_commit(struct commit *commit, void *data)
2112 {
2113 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2114 commit->object.flags |= OBJECT_ADDED;
2115 }
2117 static void show_object(struct object *obj, const struct name_path *path, const char *last)
2118 {
2119 char *name = path_name(path, last);
2121 add_preferred_base_object(name);
2122 add_object_entry(obj->sha1, obj->type, name, 0);
2123 obj->flags |= OBJECT_ADDED;
2125 /*
2126 * We will have generated the hash from the name,
2127 * but not saved a pointer to it - we can free it
2128 */
2129 free((char *)name);
2130 }
2132 static void show_edge(struct commit *commit)
2133 {
2134 add_preferred_base(commit->object.sha1);
2135 }
2137 struct in_pack_object {
2138 off_t offset;
2139 struct object *object;
2140 };
2142 struct in_pack {
2143 int alloc;
2144 int nr;
2145 struct in_pack_object *array;
2146 };
2148 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2149 {
2150 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2151 in_pack->array[in_pack->nr].object = object;
2152 in_pack->nr++;
2153 }
2155 /*
2156 * Compare the objects in the offset order, in order to emulate the
2157 * "git rev-list --objects" output that produced the pack originally.
2158 */
2159 static int ofscmp(const void *a_, const void *b_)
2160 {
2161 struct in_pack_object *a = (struct in_pack_object *)a_;
2162 struct in_pack_object *b = (struct in_pack_object *)b_;
2164 if (a->offset < b->offset)
2165 return -1;
2166 else if (a->offset > b->offset)
2167 return 1;
2168 else
2169 return hashcmp(a->object->sha1, b->object->sha1);
2170 }
2172 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2173 {
2174 struct packed_git *p;
2175 struct in_pack in_pack;
2176 uint32_t i;
2178 memset(&in_pack, 0, sizeof(in_pack));
2180 for (p = packed_git; p; p = p->next) {
2181 const unsigned char *sha1;
2182 struct object *o;
2184 if (!p->pack_local || p->pack_keep)
2185 continue;
2186 if (open_pack_index(p))
2187 die("cannot open pack index");
2189 ALLOC_GROW(in_pack.array,
2190 in_pack.nr + p->num_objects,
2191 in_pack.alloc);
2193 for (i = 0; i < p->num_objects; i++) {
2194 sha1 = nth_packed_object_sha1(p, i);
2195 o = lookup_unknown_object(sha1);
2196 if (!(o->flags & OBJECT_ADDED))
2197 mark_in_pack_object(o, p, &in_pack);
2198 o->flags |= OBJECT_ADDED;
2199 }
2200 }
2202 if (in_pack.nr) {
2203 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2204 ofscmp);
2205 for (i = 0; i < in_pack.nr; i++) {
2206 struct object *o = in_pack.array[i].object;
2207 add_object_entry(o->sha1, o->type, "", 0);
2208 }
2209 }
2210 free(in_pack.array);
2211 }
2213 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2214 {
2215 static struct packed_git *last_found = (void *)1;
2216 struct packed_git *p;
2218 p = (last_found != (void *)1) ? last_found : packed_git;
2220 while (p) {
2221 if ((!p->pack_local || p->pack_keep) &&
2222 find_pack_entry_one(sha1, p)) {
2223 last_found = p;
2224 return 1;
2225 }
2226 if (p == last_found)
2227 p = packed_git;
2228 else
2229 p = p->next;
2230 if (p == last_found)
2231 p = p->next;
2232 }
2233 return 0;
2234 }
2236 static void loosen_unused_packed_objects(struct rev_info *revs)
2237 {
2238 struct packed_git *p;
2239 uint32_t i;
2240 const unsigned char *sha1;
2242 for (p = packed_git; p; p = p->next) {
2243 if (!p->pack_local || p->pack_keep)
2244 continue;
2246 if (open_pack_index(p))
2247 die("cannot open pack index");
2249 for (i = 0; i < p->num_objects; i++) {
2250 sha1 = nth_packed_object_sha1(p, i);
2251 if (!locate_object_entry(sha1) &&
2252 !has_sha1_pack_kept_or_nonlocal(sha1))
2253 if (force_object_loose(sha1, p->mtime))
2254 die("unable to force loose object");
2255 }
2256 }
2257 }
2259 static void get_object_list(int ac, const char **av)
2260 {
2261 struct rev_info revs;
2262 char line[1000];
2263 int flags = 0;
2265 init_revisions(&revs, NULL);
2266 save_commit_buffer = 0;
2267 setup_revisions(ac, av, &revs, NULL);
2269 while (fgets(line, sizeof(line), stdin) != NULL) {
2270 int len = strlen(line);
2271 if (len && line[len - 1] == '\n')
2272 line[--len] = 0;
2273 if (!len)
2274 break;
2275 if (*line == '-') {
2276 if (!strcmp(line, "--not")) {
2277 flags ^= UNINTERESTING;
2278 continue;
2279 }
2280 die("not a rev '%s'", line);
2281 }
2282 if (handle_revision_arg(line, &revs, flags, 1))
2283 die("bad revision '%s'", line);
2284 }
2286 if (prepare_revision_walk(&revs))
2287 die("revision walk setup failed");
2288 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2289 traverse_commit_list(&revs, show_commit, show_object, NULL);
2291 if (keep_unreachable)
2292 add_objects_in_unpacked_packs(&revs);
2293 if (unpack_unreachable)
2294 loosen_unused_packed_objects(&revs);
2295 }
2297 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2298 {
2299 int use_internal_rev_list = 0;
2300 int thin = 0;
2301 int all_progress_implied = 0;
2302 uint32_t i;
2303 const char **rp_av;
2304 int rp_ac_alloc = 64;
2305 int rp_ac;
2307 read_replace_refs = 0;
2309 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2311 rp_av[0] = "pack-objects";
2312 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2313 rp_ac = 2;
2315 reset_pack_idx_option(&pack_idx_opts);
2316 git_config(git_pack_config, NULL);
2317 if (!pack_compression_seen && core_compression_seen)
2318 pack_compression_level = core_compression_level;
2320 progress = isatty(2);
2321 for (i = 1; i < argc; i++) {
2322 const char *arg = argv[i];
2324 if (*arg != '-')
2325 break;
2327 if (!strcmp("--non-empty", arg)) {
2328 non_empty = 1;
2329 continue;
2330 }
2331 if (!strcmp("--local", arg)) {
2332 local = 1;
2333 continue;
2334 }
2335 if (!strcmp("--incremental", arg)) {
2336 incremental = 1;
2337 continue;
2338 }
2339 if (!strcmp("--honor-pack-keep", arg)) {
2340 ignore_packed_keep = 1;
2341 continue;
2342 }
2343 if (!prefixcmp(arg, "--compression=")) {
2344 char *end;
2345 int level = strtoul(arg+14, &end, 0);
2346 if (!arg[14] || *end)
2347 usage(pack_usage);
2348 if (level == -1)
2349 level = Z_DEFAULT_COMPRESSION;
2350 else if (level < 0 || level > Z_BEST_COMPRESSION)
2351 die("bad pack compression level %d", level);
2352 pack_compression_level = level;
2353 continue;
2354 }
2355 if (!prefixcmp(arg, "--max-pack-size=")) {
2356 pack_size_limit_cfg = 0;
2357 if (!git_parse_ulong(arg+16, &pack_size_limit))
2358 usage(pack_usage);
2359 continue;
2360 }
2361 if (!prefixcmp(arg, "--window=")) {
2362 char *end;
2363 window = strtoul(arg+9, &end, 0);
2364 if (!arg[9] || *end)
2365 usage(pack_usage);
2366 continue;
2367 }
2368 if (!prefixcmp(arg, "--window-memory=")) {
2369 if (!git_parse_ulong(arg+16, &window_memory_limit))
2370 usage(pack_usage);
2371 continue;
2372 }
2373 if (!prefixcmp(arg, "--threads=")) {
2374 char *end;
2375 delta_search_threads = strtoul(arg+10, &end, 0);
2376 if (!arg[10] || *end || delta_search_threads < 0)
2377 usage(pack_usage);
2378 #ifdef NO_PTHREADS
2379 if (delta_search_threads != 1)
2380 warning("no threads support, "
2381 "ignoring %s", arg);
2382 #endif
2383 continue;
2384 }
2385 if (!prefixcmp(arg, "--depth=")) {
2386 char *end;
2387 depth = strtoul(arg+8, &end, 0);
2388 if (!arg[8] || *end)
2389 usage(pack_usage);
2390 continue;
2391 }
2392 if (!strcmp("--progress", arg)) {
2393 progress = 1;
2394 continue;
2395 }
2396 if (!strcmp("--all-progress", arg)) {
2397 progress = 2;
2398 continue;
2399 }
2400 if (!strcmp("--all-progress-implied", arg)) {
2401 all_progress_implied = 1;
2402 continue;
2403 }
2404 if (!strcmp("-q", arg)) {
2405 progress = 0;
2406 continue;
2407 }
2408 if (!strcmp("--no-reuse-delta", arg)) {
2409 reuse_delta = 0;
2410 continue;
2411 }
2412 if (!strcmp("--no-reuse-object", arg)) {
2413 reuse_object = reuse_delta = 0;
2414 continue;
2415 }
2416 if (!strcmp("--delta-base-offset", arg)) {
2417 allow_ofs_delta = 1;
2418 continue;
2419 }
2420 if (!strcmp("--stdout", arg)) {
2421 pack_to_stdout = 1;
2422 continue;
2423 }
2424 if (!strcmp("--revs", arg)) {
2425 use_internal_rev_list = 1;
2426 continue;
2427 }
2428 if (!strcmp("--keep-unreachable", arg)) {
2429 keep_unreachable = 1;
2430 continue;
2431 }
2432 if (!strcmp("--unpack-unreachable", arg)) {
2433 unpack_unreachable = 1;
2434 continue;
2435 }
2436 if (!strcmp("--include-tag", arg)) {
2437 include_tag = 1;
2438 continue;
2439 }
2440 if (!strcmp("--unpacked", arg) ||
2441 !strcmp("--reflog", arg) ||
2442 !strcmp("--all", arg)) {
2443 use_internal_rev_list = 1;
2444 if (rp_ac >= rp_ac_alloc - 1) {
2445 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2446 rp_av = xrealloc(rp_av,
2447 rp_ac_alloc * sizeof(*rp_av));
2448 }
2449 rp_av[rp_ac++] = arg;
2450 continue;
2451 }
2452 if (!strcmp("--thin", arg)) {
2453 use_internal_rev_list = 1;
2454 thin = 1;
2455 rp_av[1] = "--objects-edge";
2456 continue;
2457 }
2458 if (!prefixcmp(arg, "--index-version=")) {
2459 char *c;
2460 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2461 if (pack_idx_opts.version > 2)
2462 die("bad %s", arg);
2463 if (*c == ',')
2464 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2465 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2466 die("bad %s", arg);
2467 continue;
2468 }
2469 if (!strcmp(arg, "--keep-true-parents")) {
2470 grafts_replace_parents = 0;
2471 continue;
2472 }
2473 usage(pack_usage);
2474 }
2476 /* Traditionally "pack-objects [options] base extra" failed;
2477 * we would however want to take refs parameter that would
2478 * have been given to upstream rev-list ourselves, which means
2479 * we somehow want to say what the base name is. So the
2480 * syntax would be:
2481 *
2482 * pack-objects [options] base <refs...>
2483 *
2484 * in other words, we would treat the first non-option as the
2485 * base_name and send everything else to the internal revision
2486 * walker.
2487 */
2489 if (!pack_to_stdout)
2490 base_name = argv[i++];
2492 if (pack_to_stdout != !base_name)
2493 usage(pack_usage);
2495 if (!pack_to_stdout && !pack_size_limit)
2496 pack_size_limit = pack_size_limit_cfg;
2497 if (pack_to_stdout && pack_size_limit)
2498 die("--max-pack-size cannot be used to build a pack for transfer.");
2499 if (pack_size_limit && pack_size_limit < 1024*1024) {
2500 warning("minimum pack size limit is 1 MiB");
2501 pack_size_limit = 1024*1024;
2502 }
2504 if (!pack_to_stdout && thin)
2505 die("--thin cannot be used to build an indexable pack.");
2507 if (keep_unreachable && unpack_unreachable)
2508 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2510 if (progress && all_progress_implied)
2511 progress = 2;
2513 prepare_packed_git();
2515 if (progress)
2516 progress_state = start_progress("Counting objects", 0);
2517 if (!use_internal_rev_list)
2518 read_object_list_from_stdin();
2519 else {
2520 rp_av[rp_ac] = NULL;
2521 get_object_list(rp_ac, rp_av);
2522 }
2523 cleanup_preferred_base();
2524 if (include_tag && nr_result)
2525 for_each_ref(add_ref_tag, NULL);
2526 stop_progress(&progress_state);
2528 if (non_empty && !nr_result)
2529 return 0;
2530 if (nr_result)
2531 prepare_pack(window, depth);
2532 write_pack_file();
2533 if (progress)
2534 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2535 " reused %"PRIu32" (delta %"PRIu32")\n",
2536 written, written_delta, reused, reused_delta);
2537 return 0;
2538 }