?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'maint-1.7.7' into maint
[git/jnareb-git.git] / builtin / pack-objects.c
blobef703dfeb611c0820a68f4c7cebae154689fa5f5
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"
20
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]";
31
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 };
57
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;
67
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;
87
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;
91
92 static unsigned long window_memory_limit = 0;
93
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);
102
103 /*
104 * stats
105 */
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
108
109
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;
115
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 }
130
131 static unsigned long do_compress(void **pptr, unsigned long size)
132 {
133 git_zstream stream;
134 void *in, *out;
135 unsigned long maxsize;
136
137 memset(&stream, 0, sizeof(stream));
138 git_deflate_init(&stream, pack_compression_level);
139 maxsize = git_deflate_bound(&stream, size);
140
141 in = *pptr;
142 out = xmalloc(maxsize);
143 *pptr = out;
144
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);
152
153 free(in);
154 return stream.total_out;
155 }
156
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;
170
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 }
186
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;
195
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 }
205
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;
217
218 if (!pack_to_stdout)
219 crc32_begin(f);
220
221 type = entry->type;
222
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;
234
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 */
245
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 */
262
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 }
288
289 if (entry->z_delta_size)
290 datalen = entry->z_delta_size;
291 else
292 datalen = do_compress(&buf, size);
293
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);
299
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;
345
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);
350
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 }
360
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 }
369
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 }
411
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 };
418
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;
425
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 }
440
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 }
456
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;
464
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 }
471
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);
477
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 }
487
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 }
497
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 }
540
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;
546
547 for (root = e; root->delta; root = root->delta)
548 ; /* nothing */
549 add_descendants_to_write_order(wo, endp, root);
550 }
551
552 static struct object_entry **compute_write_order(void)
553 {
554 unsigned int i, wo_end, last_untagged;
555
556 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));
557
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 }
564
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 }
578
579 /*
580 * Mark objects that are at the tip of tags.
581 */
582 for_each_tag_ref(mark_tagged, NULL);
583
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;
594
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 }
602
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 }
612
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 }
621
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 }
629
630 if (wo_end != nr_objects)
631 die("ordered %u objects, expected %"PRIu32, wo_end, nr_objects);
632
633 return wo;
634 }
635
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;
645
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();
650
651 do {
652 unsigned char sha1[20];
653 char *pack_tmp_name = NULL;
654
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 }
665
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 }
678
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 }
693
694 if (!pack_to_stdout) {
695 struct stat st;
696 const char *idx_tmp_name;
697 char tmpname[PATH_MAX];
698
699 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
700 &pack_idx_opts, sha1);
701
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");
709
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 }
730
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");
737
738 free((void *) idx_tmp_name);
739 free(pack_tmp_name);
740 puts(sha1_to_hex(sha1));
741 }
742
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);
749
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 }
757
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 }
772
773 static struct object_entry *locate_object_entry(const unsigned char *sha1)
774 {
775 int i;
776
777 if (!object_ix_hashsz)
778 return NULL;
779
780 i = locate_object_entry_hash(sha1);
781 if (0 <= i)
782 return &objects[object_ix[i]-1];
783 return NULL;
784 }
785
786 static void rehash_objects(void)
787 {
788 uint32_t i;
789 struct object_entry *oe;
790
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 }
804
805 static unsigned name_hash(const char *name)
806 {
807 unsigned c, hash = 0;
808
809 if (!name)
810 return 0;
811
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 }
824
825 static void setup_delta_attr_check(struct git_attr_check *check)
826 {
827 static struct git_attr *attr_delta;
828
829 if (!attr_delta)
830 attr_delta = git_attr("delta");
831
832 check[0].attr = attr_delta;
833 }
834
835 static int no_try_delta(const char *path)
836 {
837 struct git_attr_check check[1];
838
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 }
846
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);
855
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 }
866
867 if (!exclude && local && has_loose_object_nonlocal(sha1))
868 return 0;
869
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 }
891
892 if (nr_objects >= nr_alloc) {
893 nr_alloc = (nr_alloc + 1024) * 3 / 2;
894 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
895 }
896
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 }
911
912 if (object_ix_hashsz * 3 <= nr_objects * 4)
913 rehash_objects();
914 else
915 object_ix[-1 - ix] = nr_objects;
916
917 display_progress(progress_state, nr_objects);
918
919 if (name && no_try_delta(name))
920 entry->no_try_delta = 1;
921
922 return 1;
923 }
924
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 };
932
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 }
942
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;
952
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;
962
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 }
981
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 }
992
993 /* We need to either cache or return a throwaway copy */
994
995 if (available_ix < 0)
996 ent = NULL;
997 else {
998 ent = pbase_tree_cache[available_ix];
999 my_ix = available_ix;
1000 }
1001
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 }
1019
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 }
1029
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 }
1037
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;
1045
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);
1066
1067 tree = pbase_tree_get(entry.sha1);
1068 if (!tree)
1069 return;
1070 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1071
1072 add_pbase_object(&sub, down, downlen, fullname);
1073 pbase_tree_put(tree);
1074 }
1075 }
1076 }
1077
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 }
1096
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 }
1117
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);
1123
1124 if (!num_preferred_base || check_pbase_path(hash))
1125 return;
1126
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 }
1139
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];
1146
1147 if (window <= num_preferred_base++)
1148 return;
1149
1150 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1151 if (!data)
1152 return;
1153
1154 for (it = pbase_tree; it; it = it->next) {
1155 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1156 free(data);
1157 return;
1158 }
1159 }
1160
1161 it = xcalloc(1, sizeof(*it));
1162 it->next = pbase_tree;
1163 pbase_tree = it;
1164
1165 hashcpy(it->pcache.sha1, tree_sha1);
1166 it->pcache.tree_data = data;
1167 it->pcache.tree_size = size;
1168 }
1169
1170 static void cleanup_preferred_base(void)
1171 {
1172 struct pbase_tree *it;
1173 unsigned i;
1174
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 }
1183
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 }
1191
1192 free(done_pbase_paths);
1193 done_pbase_paths = NULL;
1194 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1195 }
1196
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;
1208
1209 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1210
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;
1220
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 }
1273
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 }
1293
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 }
1307
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 }
1316
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 }
1325
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;
1330
1331 /* avoid filesystem trashing with loose objects */
1332 if (!a->in_pack && !b->in_pack)
1333 return hashcmp(a->idx.sha1, b->idx.sha1);
1334
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 }
1342
1343 static void get_object_details(void)
1344 {
1345 uint32_t i;
1346 struct object_entry **sorted_by_offset;
1347
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);
1352
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 }
1359
1360 free(sorted_by_offset);
1361 }
1362
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;
1376
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 }
1395
1396 struct unpacked {
1397 struct object_entry *entry;
1398 void *data;
1399 struct delta_index *index;
1400 unsigned depth;
1401 };
1402
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;
1408
1409 if (delta_size < cache_max_small_delta_size)
1410 return 1;
1411
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;
1415
1416 return 0;
1417 }
1418
1419 #ifndef NO_PTHREADS
1420
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)
1424
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)
1428
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)
1432
1433 #else
1434
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
1441
1442 #endif
1443
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;
1453
1454 /* Don't bother doing diffs between different types */
1455 if (trg_entry->type != src_entry->type)
1456 return -1;
1457
1458 /*
1459 * We do not bother to try a delta that we discarded on an
1460 * earlier try, but only when reusing delta data. Note that
1461 * src_entry that is marked as the preferred_base should always
1462 * be considered, as even if we produce a suboptimal delta against
1463 * it, we will still save the transfer cost, as we already know
1464 * the other side has it and we won't send src_entry at all.
1465 */
1466 if (reuse_delta && trg_entry->in_pack &&
1467 trg_entry->in_pack == src_entry->in_pack &&
1468 !src_entry->preferred_base &&
1469 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1470 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1471 return 0;
1472
1473 /* Let's not bust the allowed depth. */
1474 if (src->depth >= max_depth)
1475 return 0;
1476
1477 /* Now some size filtering heuristics. */
1478 trg_size = trg_entry->size;
1479 if (!trg_entry->delta) {
1480 max_size = trg_size/2 - 20;
1481 ref_depth = 1;
1482 } else {
1483 max_size = trg_entry->delta_size;
1484 ref_depth = trg->depth;
1485 }
1486 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1487 (max_depth - ref_depth + 1);
1488 if (max_size == 0)
1489 return 0;
1490 src_size = src_entry->size;
1491 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1492 if (sizediff >= max_size)
1493 return 0;
1494 if (trg_size < src_size / 32)
1495 return 0;
1496
1497 /* Load data if not already done */
1498 if (!trg->data) {
1499 read_lock();
1500 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1501 read_unlock();
1502 if (!trg->data)
1503 die("object %s cannot be read",
1504 sha1_to_hex(trg_entry->idx.sha1));
1505 if (sz != trg_size)
1506 die("object %s inconsistent object length (%lu vs %lu)",
1507 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1508 *mem_usage += sz;
1509 }
1510 if (!src->data) {
1511 read_lock();
1512 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1513 read_unlock();
1514 if (!src->data) {
1515 if (src_entry->preferred_base) {
1516 static int warned = 0;
1517 if (!warned++)
1518 warning("object %s cannot be read",
1519 sha1_to_hex(src_entry->idx.sha1));
1520 /*
1521 * Those objects are not included in the
1522 * resulting pack. Be resilient and ignore
1523 * them if they can't be read, in case the
1524 * pack could be created nevertheless.
1525 */
1526 return 0;
1527 }
1528 die("object %s cannot be read",
1529 sha1_to_hex(src_entry->idx.sha1));
1530 }
1531 if (sz != src_size)
1532 die("object %s inconsistent object length (%lu vs %lu)",
1533 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1534 *mem_usage += sz;
1535 }
1536 if (!src->index) {
1537 src->index = create_delta_index(src->data, src_size);
1538 if (!src->index) {
1539 static int warned = 0;
1540 if (!warned++)
1541 warning("suboptimal pack - out of memory");
1542 return 0;
1543 }
1544 *mem_usage += sizeof_delta_index(src->index);
1545 }
1546
1547 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1548 if (!delta_buf)
1549 return 0;
1550
1551 if (trg_entry->delta) {
1552 /* Prefer only shallower same-sized deltas. */
1553 if (delta_size == trg_entry->delta_size &&
1554 src->depth + 1 >= trg->depth) {
1555 free(delta_buf);
1556 return 0;
1557 }
1558 }
1559
1560 /*
1561 * Handle memory allocation outside of the cache
1562 * accounting lock. Compiler will optimize the strangeness
1563 * away when NO_PTHREADS is defined.
1564 */
1565 free(trg_entry->delta_data);
1566 cache_lock();
1567 if (trg_entry->delta_data) {
1568 delta_cache_size -= trg_entry->delta_size;
1569 trg_entry->delta_data = NULL;
1570 }
1571 if (delta_cacheable(src_size, trg_size, delta_size)) {
1572 delta_cache_size += delta_size;
1573 cache_unlock();
1574 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1575 } else {
1576 cache_unlock();
1577 free(delta_buf);
1578 }
1579
1580 trg_entry->delta = src_entry;
1581 trg_entry->delta_size = delta_size;
1582 trg->depth = src->depth + 1;
1583
1584 return 1;
1585 }
1586
1587 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1588 {
1589 struct object_entry *child = me->delta_child;
1590 unsigned int m = n;
1591 while (child) {
1592 unsigned int c = check_delta_limit(child, n + 1);
1593 if (m < c)
1594 m = c;
1595 child = child->delta_sibling;
1596 }
1597 return m;
1598 }
1599
1600 static unsigned long free_unpacked(struct unpacked *n)
1601 {
1602 unsigned long freed_mem = sizeof_delta_index(n->index);
1603 free_delta_index(n->index);
1604 n->index = NULL;
1605 if (n->data) {
1606 freed_mem += n->entry->size;
1607 free(n->data);
1608 n->data = NULL;
1609 }
1610 n->entry = NULL;
1611 n->depth = 0;
1612 return freed_mem;
1613 }
1614
1615 static void find_deltas(struct object_entry **list, unsigned *list_size,
1616 int window, int depth, unsigned *processed)
1617 {
1618 uint32_t i, idx = 0, count = 0;
1619 struct unpacked *array;
1620 unsigned long mem_usage = 0;
1621
1622 array = xcalloc(window, sizeof(struct unpacked));
1623
1624 for (;;) {
1625 struct object_entry *entry;
1626 struct unpacked *n = array + idx;
1627 int j, max_depth, best_base = -1;
1628
1629 progress_lock();
1630 if (!*list_size) {
1631 progress_unlock();
1632 break;
1633 }
1634 entry = *list++;
1635 (*list_size)--;
1636 if (!entry->preferred_base) {
1637 (*processed)++;
1638 display_progress(progress_state, *processed);
1639 }
1640 progress_unlock();
1641
1642 mem_usage -= free_unpacked(n);
1643 n->entry = entry;
1644
1645 while (window_memory_limit &&
1646 mem_usage > window_memory_limit &&
1647 count > 1) {
1648 uint32_t tail = (idx + window - count) % window;
1649 mem_usage -= free_unpacked(array + tail);
1650 count--;
1651 }
1652
1653 /* We do not compute delta to *create* objects we are not
1654 * going to pack.
1655 */
1656 if (entry->preferred_base)
1657 goto next;
1658
1659 /*
1660 * If the current object is at pack edge, take the depth the
1661 * objects that depend on the current object into account
1662 * otherwise they would become too deep.
1663 */
1664 max_depth = depth;
1665 if (entry->delta_child) {
1666 max_depth -= check_delta_limit(entry, 0);
1667 if (max_depth <= 0)
1668 goto next;
1669 }
1670
1671 j = window;
1672 while (--j > 0) {
1673 int ret;
1674 uint32_t other_idx = idx + j;
1675 struct unpacked *m;
1676 if (other_idx >= window)
1677 other_idx -= window;
1678 m = array + other_idx;
1679 if (!m->entry)
1680 break;
1681 ret = try_delta(n, m, max_depth, &mem_usage);
1682 if (ret < 0)
1683 break;
1684 else if (ret > 0)
1685 best_base = other_idx;
1686 }
1687
1688 /*
1689 * If we decided to cache the delta data, then it is best
1690 * to compress it right away. First because we have to do
1691 * it anyway, and doing it here while we're threaded will
1692 * save a lot of time in the non threaded write phase,
1693 * as well as allow for caching more deltas within
1694 * the same cache size limit.
1695 * ...
1696 * But only if not writing to stdout, since in that case
1697 * the network is most likely throttling writes anyway,
1698 * and therefore it is best to go to the write phase ASAP
1699 * instead, as we can afford spending more time compressing
1700 * between writes at that moment.
1701 */
1702 if (entry->delta_data && !pack_to_stdout) {
1703 entry->z_delta_size = do_compress(&entry->delta_data,
1704 entry->delta_size);
1705 cache_lock();
1706 delta_cache_size -= entry->delta_size;
1707 delta_cache_size += entry->z_delta_size;
1708 cache_unlock();
1709 }
1710
1711 /* if we made n a delta, and if n is already at max
1712 * depth, leaving it in the window is pointless. we
1713 * should evict it first.
1714 */
1715 if (entry->delta && max_depth <= n->depth)
1716 continue;
1717
1718 /*
1719 * Move the best delta base up in the window, after the
1720 * currently deltified object, to keep it longer. It will
1721 * be the first base object to be attempted next.
1722 */
1723 if (entry->delta) {
1724 struct unpacked swap = array[best_base];
1725 int dist = (window + idx - best_base) % window;
1726 int dst = best_base;
1727 while (dist--) {
1728 int src = (dst + 1) % window;
1729 array[dst] = array[src];
1730 dst = src;
1731 }
1732 array[dst] = swap;
1733 }
1734
1735 next:
1736 idx++;
1737 if (count + 1 < window)
1738 count++;
1739 if (idx >= window)
1740 idx = 0;
1741 }
1742
1743 for (i = 0; i < window; ++i) {
1744 free_delta_index(array[i].index);
1745 free(array[i].data);
1746 }
1747 free(array);
1748 }
1749
1750 #ifndef NO_PTHREADS
1751
1752 static void try_to_free_from_threads(size_t size)
1753 {
1754 read_lock();
1755 release_pack_memory(size, -1);
1756 read_unlock();
1757 }
1758
1759 static try_to_free_t old_try_to_free_routine;
1760
1761 /*
1762 * The main thread waits on the condition that (at least) one of the workers
1763 * has stopped working (which is indicated in the .working member of
1764 * struct thread_params).
1765 * When a work thread has completed its work, it sets .working to 0 and
1766 * signals the main thread and waits on the condition that .data_ready
1767 * becomes 1.
1768 */
1769
1770 struct thread_params {
1771 pthread_t thread;
1772 struct object_entry **list;
1773 unsigned list_size;
1774 unsigned remaining;
1775 int window;
1776 int depth;
1777 int working;
1778 int data_ready;
1779 pthread_mutex_t mutex;
1780 pthread_cond_t cond;
1781 unsigned *processed;
1782 };
1783
1784 static pthread_cond_t progress_cond;
1785
1786 /*
1787 * Mutex and conditional variable can't be statically-initialized on Windows.
1788 */
1789 static void init_threaded_search(void)
1790 {
1791 init_recursive_mutex(&read_mutex);
1792 pthread_mutex_init(&cache_mutex, NULL);
1793 pthread_mutex_init(&progress_mutex, NULL);
1794 pthread_cond_init(&progress_cond, NULL);
1795 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1796 }
1797
1798 static void cleanup_threaded_search(void)
1799 {
1800 set_try_to_free_routine(old_try_to_free_routine);
1801 pthread_cond_destroy(&progress_cond);
1802 pthread_mutex_destroy(&read_mutex);
1803 pthread_mutex_destroy(&cache_mutex);
1804 pthread_mutex_destroy(&progress_mutex);
1805 }
1806
1807 static void *threaded_find_deltas(void *arg)
1808 {
1809 struct thread_params *me = arg;
1810
1811 while (me->remaining) {
1812 find_deltas(me->list, &me->remaining,
1813 me->window, me->depth, me->processed);
1814
1815 progress_lock();
1816 me->working = 0;
1817 pthread_cond_signal(&progress_cond);
1818 progress_unlock();
1819
1820 /*
1821 * We must not set ->data_ready before we wait on the
1822 * condition because the main thread may have set it to 1
1823 * before we get here. In order to be sure that new
1824 * work is available if we see 1 in ->data_ready, it
1825 * was initialized to 0 before this thread was spawned
1826 * and we reset it to 0 right away.
1827 */
1828 pthread_mutex_lock(&me->mutex);
1829 while (!me->data_ready)
1830 pthread_cond_wait(&me->cond, &me->mutex);
1831 me->data_ready = 0;
1832 pthread_mutex_unlock(&me->mutex);
1833 }
1834 /* leave ->working 1 so that this doesn't get more work assigned */
1835 return NULL;
1836 }
1837
1838 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1839 int window, int depth, unsigned *processed)
1840 {
1841 struct thread_params *p;
1842 int i, ret, active_threads = 0;
1843
1844 init_threaded_search();
1845
1846 if (!delta_search_threads) /* --threads=0 means autodetect */
1847 delta_search_threads = online_cpus();
1848 if (delta_search_threads <= 1) {
1849 find_deltas(list, &list_size, window, depth, processed);
1850 cleanup_threaded_search();
1851 return;
1852 }
1853 if (progress > pack_to_stdout)
1854 fprintf(stderr, "Delta compression using up to %d threads.\n",
1855 delta_search_threads);
1856 p = xcalloc(delta_search_threads, sizeof(*p));
1857
1858 /* Partition the work amongst work threads. */
1859 for (i = 0; i < delta_search_threads; i++) {
1860 unsigned sub_size = list_size / (delta_search_threads - i);
1861
1862 /* don't use too small segments or no deltas will be found */
1863 if (sub_size < 2*window && i+1 < delta_search_threads)
1864 sub_size = 0;
1865
1866 p[i].window = window;
1867 p[i].depth = depth;
1868 p[i].processed = processed;
1869 p[i].working = 1;
1870 p[i].data_ready = 0;
1871
1872 /* try to split chunks on "path" boundaries */
1873 while (sub_size && sub_size < list_size &&
1874 list[sub_size]->hash &&
1875 list[sub_size]->hash == list[sub_size-1]->hash)
1876 sub_size++;
1877
1878 p[i].list = list;
1879 p[i].list_size = sub_size;
1880 p[i].remaining = sub_size;
1881
1882 list += sub_size;
1883 list_size -= sub_size;
1884 }
1885
1886 /* Start work threads. */
1887 for (i = 0; i < delta_search_threads; i++) {
1888 if (!p[i].list_size)
1889 continue;
1890 pthread_mutex_init(&p[i].mutex, NULL);
1891 pthread_cond_init(&p[i].cond, NULL);
1892 ret = pthread_create(&p[i].thread, NULL,
1893 threaded_find_deltas, &p[i]);
1894 if (ret)
1895 die("unable to create thread: %s", strerror(ret));
1896 active_threads++;
1897 }
1898
1899 /*
1900 * Now let's wait for work completion. Each time a thread is done
1901 * with its work, we steal half of the remaining work from the
1902 * thread with the largest number of unprocessed objects and give
1903 * it to that newly idle thread. This ensure good load balancing
1904 * until the remaining object list segments are simply too short
1905 * to be worth splitting anymore.
1906 */
1907 while (active_threads) {
1908 struct thread_params *target = NULL;
1909 struct thread_params *victim = NULL;
1910 unsigned sub_size = 0;
1911
1912 progress_lock();
1913 for (;;) {
1914 for (i = 0; !target && i < delta_search_threads; i++)
1915 if (!p[i].working)
1916 target = &p[i];
1917 if (target)
1918 break;
1919 pthread_cond_wait(&progress_cond, &progress_mutex);
1920 }
1921
1922 for (i = 0; i < delta_search_threads; i++)
1923 if (p[i].remaining > 2*window &&
1924 (!victim || victim->remaining < p[i].remaining))
1925 victim = &p[i];
1926 if (victim) {
1927 sub_size = victim->remaining / 2;
1928 list = victim->list + victim->list_size - sub_size;
1929 while (sub_size && list[0]->hash &&
1930 list[0]->hash == list[-1]->hash) {
1931 list++;
1932 sub_size--;
1933 }
1934 if (!sub_size) {
1935 /*
1936 * It is possible for some "paths" to have
1937 * so many objects that no hash boundary
1938 * might be found. Let's just steal the
1939 * exact half in that case.
1940 */
1941 sub_size = victim->remaining / 2;
1942 list -= sub_size;
1943 }
1944 target->list = list;
1945 victim->list_size -= sub_size;
1946 victim->remaining -= sub_size;
1947 }
1948 target->list_size = sub_size;
1949 target->remaining = sub_size;
1950 target->working = 1;
1951 progress_unlock();
1952
1953 pthread_mutex_lock(&target->mutex);
1954 target->data_ready = 1;
1955 pthread_cond_signal(&target->cond);
1956 pthread_mutex_unlock(&target->mutex);
1957
1958 if (!sub_size) {
1959 pthread_join(target->thread, NULL);
1960 pthread_cond_destroy(&target->cond);
1961 pthread_mutex_destroy(&target->mutex);
1962 active_threads--;
1963 }
1964 }
1965 cleanup_threaded_search();
1966 free(p);
1967 }
1968
1969 #else
1970 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1971 #endif
1972
1973 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1974 {
1975 unsigned char peeled[20];
1976
1977 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1978 !peel_ref(path, peeled) && /* peelable? */
1979 !is_null_sha1(peeled) && /* annotated tag? */
1980 locate_object_entry(peeled)) /* object packed? */
1981 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1982 return 0;
1983 }
1984
1985 static void prepare_pack(int window, int depth)
1986 {
1987 struct object_entry **delta_list;
1988 uint32_t i, nr_deltas;
1989 unsigned n;
1990
1991 get_object_details();
1992
1993 /*
1994 * If we're locally repacking then we need to be doubly careful
1995 * from now on in order to make sure no stealth corruption gets
1996 * propagated to the new pack. Clients receiving streamed packs
1997 * should validate everything they get anyway so no need to incur
1998 * the additional cost here in that case.
1999 */
2000 if (!pack_to_stdout)
2001 do_check_packed_object_crc = 1;
2002
2003 if (!nr_objects || !window || !depth)
2004 return;
2005
2006 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
2007 nr_deltas = n = 0;
2008
2009 for (i = 0; i < nr_objects; i++) {
2010 struct object_entry *entry = objects + i;
2011
2012 if (entry->delta)
2013 /* This happens if we decided to reuse existing
2014 * delta from a pack. "reuse_delta &&" is implied.
2015 */
2016 continue;
2017
2018 if (entry->size < 50)
2019 continue;
2020
2021 if (entry->no_try_delta)
2022 continue;
2023
2024 if (!entry->preferred_base) {
2025 nr_deltas++;
2026 if (entry->type < 0)
2027 die("unable to get type of object %s",
2028 sha1_to_hex(entry->idx.sha1));
2029 } else {
2030 if (entry->type < 0) {
2031 /*
2032 * This object is not found, but we
2033 * don't have to include it anyway.
2034 */
2035 continue;
2036 }
2037 }
2038
2039 delta_list[n++] = entry;
2040 }
2041
2042 if (nr_deltas && n > 1) {
2043 unsigned nr_done = 0;
2044 if (progress)
2045 progress_state = start_progress("Compressing objects",
2046 nr_deltas);
2047 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2048 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2049 stop_progress(&progress_state);
2050 if (nr_done != nr_deltas)
2051 die("inconsistency with delta count");
2052 }
2053 free(delta_list);
2054 }
2055
2056 static int git_pack_config(const char *k, const char *v, void *cb)
2057 {
2058 if (!strcmp(k, "pack.window")) {
2059 window = git_config_int(k, v);
2060 return 0;
2061 }
2062 if (!strcmp(k, "pack.windowmemory")) {
2063 window_memory_limit = git_config_ulong(k, v);
2064 return 0;
2065 }
2066 if (!strcmp(k, "pack.depth")) {
2067 depth = git_config_int(k, v);
2068 return 0;
2069 }
2070 if (!strcmp(k, "pack.compression")) {
2071 int level = git_config_int(k, v);
2072 if (level == -1)
2073 level = Z_DEFAULT_COMPRESSION;
2074 else if (level < 0 || level > Z_BEST_COMPRESSION)
2075 die("bad pack compression level %d", level);
2076 pack_compression_level = level;
2077 pack_compression_seen = 1;
2078 return 0;
2079 }
2080 if (!strcmp(k, "pack.deltacachesize")) {
2081 max_delta_cache_size = git_config_int(k, v);
2082 return 0;
2083 }
2084 if (!strcmp(k, "pack.deltacachelimit")) {
2085 cache_max_small_delta_size = git_config_int(k, v);
2086 return 0;
2087 }
2088 if (!strcmp(k, "pack.threads")) {
2089 delta_search_threads = git_config_int(k, v);
2090 if (delta_search_threads < 0)
2091 die("invalid number of threads specified (%d)",
2092 delta_search_threads);
2093 #ifdef NO_PTHREADS
2094 if (delta_search_threads != 1)
2095 warning("no threads support, ignoring %s", k);
2096 #endif
2097 return 0;
2098 }
2099 if (!strcmp(k, "pack.indexversion")) {
2100 pack_idx_opts.version = git_config_int(k, v);
2101 if (pack_idx_opts.version > 2)
2102 die("bad pack.indexversion=%"PRIu32,
2103 pack_idx_opts.version);
2104 return 0;
2105 }
2106 if (!strcmp(k, "pack.packsizelimit")) {
2107 pack_size_limit_cfg = git_config_ulong(k, v);
2108 return 0;
2109 }
2110 return git_default_config(k, v, cb);
2111 }
2112
2113 static void read_object_list_from_stdin(void)
2114 {
2115 char line[40 + 1 + PATH_MAX + 2];
2116 unsigned char sha1[20];
2117
2118 for (;;) {
2119 if (!fgets(line, sizeof(line), stdin)) {
2120 if (feof(stdin))
2121 break;
2122 if (!ferror(stdin))
2123 die("fgets returned NULL, not EOF, not error!");
2124 if (errno != EINTR)
2125 die_errno("fgets");
2126 clearerr(stdin);
2127 continue;
2128 }
2129 if (line[0] == '-') {
2130 if (get_sha1_hex(line+1, sha1))
2131 die("expected edge sha1, got garbage:\n %s",
2132 line);
2133 add_preferred_base(sha1);
2134 continue;
2135 }
2136 if (get_sha1_hex(line, sha1))
2137 die("expected sha1, got garbage:\n %s", line);
2138
2139 add_preferred_base_object(line+41);
2140 add_object_entry(sha1, 0, line+41, 0);
2141 }
2142 }
2143
2144 #define OBJECT_ADDED (1u<<20)
2145
2146 static void show_commit(struct commit *commit, void *data)
2147 {
2148 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2149 commit->object.flags |= OBJECT_ADDED;
2150 }
2151
2152 static void show_object(struct object *obj,
2153 const struct name_path *path, const char *last,
2154 void *data)
2155 {
2156 char *name = path_name(path, last);
2157
2158 add_preferred_base_object(name);
2159 add_object_entry(obj->sha1, obj->type, name, 0);
2160 obj->flags |= OBJECT_ADDED;
2161
2162 /*
2163 * We will have generated the hash from the name,
2164 * but not saved a pointer to it - we can free it
2165 */
2166 free((char *)name);
2167 }
2168
2169 static void show_edge(struct commit *commit)
2170 {
2171 add_preferred_base(commit->object.sha1);
2172 }
2173
2174 struct in_pack_object {
2175 off_t offset;
2176 struct object *object;
2177 };
2178
2179 struct in_pack {
2180 int alloc;
2181 int nr;
2182 struct in_pack_object *array;
2183 };
2184
2185 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2186 {
2187 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2188 in_pack->array[in_pack->nr].object = object;
2189 in_pack->nr++;
2190 }
2191
2192 /*
2193 * Compare the objects in the offset order, in order to emulate the
2194 * "git rev-list --objects" output that produced the pack originally.
2195 */
2196 static int ofscmp(const void *a_, const void *b_)
2197 {
2198 struct in_pack_object *a = (struct in_pack_object *)a_;
2199 struct in_pack_object *b = (struct in_pack_object *)b_;
2200
2201 if (a->offset < b->offset)
2202 return -1;
2203 else if (a->offset > b->offset)
2204 return 1;
2205 else
2206 return hashcmp(a->object->sha1, b->object->sha1);
2207 }
2208
2209 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2210 {
2211 struct packed_git *p;
2212 struct in_pack in_pack;
2213 uint32_t i;
2214
2215 memset(&in_pack, 0, sizeof(in_pack));
2216
2217 for (p = packed_git; p; p = p->next) {
2218 const unsigned char *sha1;
2219 struct object *o;
2220
2221 if (!p->pack_local || p->pack_keep)
2222 continue;
2223 if (open_pack_index(p))
2224 die("cannot open pack index");
2225
2226 ALLOC_GROW(in_pack.array,
2227 in_pack.nr + p->num_objects,
2228 in_pack.alloc);
2229
2230 for (i = 0; i < p->num_objects; i++) {
2231 sha1 = nth_packed_object_sha1(p, i);
2232 o = lookup_unknown_object(sha1);
2233 if (!(o->flags & OBJECT_ADDED))
2234 mark_in_pack_object(o, p, &in_pack);
2235 o->flags |= OBJECT_ADDED;
2236 }
2237 }
2238
2239 if (in_pack.nr) {
2240 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2241 ofscmp);
2242 for (i = 0; i < in_pack.nr; i++) {
2243 struct object *o = in_pack.array[i].object;
2244 add_object_entry(o->sha1, o->type, "", 0);
2245 }
2246 }
2247 free(in_pack.array);
2248 }
2249
2250 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2251 {
2252 static struct packed_git *last_found = (void *)1;
2253 struct packed_git *p;
2254
2255 p = (last_found != (void *)1) ? last_found : packed_git;
2256
2257 while (p) {
2258 if ((!p->pack_local || p->pack_keep) &&
2259 find_pack_entry_one(sha1, p)) {
2260 last_found = p;
2261 return 1;
2262 }
2263 if (p == last_found)
2264 p = packed_git;
2265 else
2266 p = p->next;
2267 if (p == last_found)
2268 p = p->next;
2269 }
2270 return 0;
2271 }
2272
2273 static void loosen_unused_packed_objects(struct rev_info *revs)
2274 {
2275 struct packed_git *p;
2276 uint32_t i;
2277 const unsigned char *sha1;
2278
2279 for (p = packed_git; p; p = p->next) {
2280 if (!p->pack_local || p->pack_keep)
2281 continue;
2282
2283 if (open_pack_index(p))
2284 die("cannot open pack index");
2285
2286 for (i = 0; i < p->num_objects; i++) {
2287 sha1 = nth_packed_object_sha1(p, i);
2288 if (!locate_object_entry(sha1) &&
2289 !has_sha1_pack_kept_or_nonlocal(sha1))
2290 if (force_object_loose(sha1, p->mtime))
2291 die("unable to force loose object");
2292 }
2293 }
2294 }
2295
2296 static void get_object_list(int ac, const char **av)
2297 {
2298 struct rev_info revs;
2299 char line[1000];
2300 int flags = 0;
2301
2302 init_revisions(&revs, NULL);
2303 save_commit_buffer = 0;
2304 setup_revisions(ac, av, &revs, NULL);
2305
2306 while (fgets(line, sizeof(line), stdin) != NULL) {
2307 int len = strlen(line);
2308 if (len && line[len - 1] == '\n')
2309 line[--len] = 0;
2310 if (!len)
2311 break;
2312 if (*line == '-') {
2313 if (!strcmp(line, "--not")) {
2314 flags ^= UNINTERESTING;
2315 continue;
2316 }
2317 die("not a rev '%s'", line);
2318 }
2319 if (handle_revision_arg(line, &revs, flags, 1))
2320 die("bad revision '%s'", line);
2321 }
2322
2323 if (prepare_revision_walk(&revs))
2324 die("revision walk setup failed");
2325 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2326 traverse_commit_list(&revs, show_commit, show_object, NULL);
2327
2328 if (keep_unreachable)
2329 add_objects_in_unpacked_packs(&revs);
2330 if (unpack_unreachable)
2331 loosen_unused_packed_objects(&revs);
2332 }
2333
2334 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2335 {
2336 int use_internal_rev_list = 0;
2337 int thin = 0;
2338 int all_progress_implied = 0;
2339 uint32_t i;
2340 const char **rp_av;
2341 int rp_ac_alloc = 64;
2342 int rp_ac;
2343
2344 read_replace_refs = 0;
2345
2346 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2347
2348 rp_av[0] = "pack-objects";
2349 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2350 rp_ac = 2;
2351
2352 reset_pack_idx_option(&pack_idx_opts);
2353 git_config(git_pack_config, NULL);
2354 if (!pack_compression_seen && core_compression_seen)
2355 pack_compression_level = core_compression_level;
2356
2357 progress = isatty(2);
2358 for (i = 1; i < argc; i++) {
2359 const char *arg = argv[i];
2360
2361 if (*arg != '-')
2362 break;
2363
2364 if (!strcmp("--non-empty", arg)) {
2365 non_empty = 1;
2366 continue;
2367 }
2368 if (!strcmp("--local", arg)) {
2369 local = 1;
2370 continue;
2371 }
2372 if (!strcmp("--incremental", arg)) {
2373 incremental = 1;
2374 continue;
2375 }
2376 if (!strcmp("--honor-pack-keep", arg)) {
2377 ignore_packed_keep = 1;
2378 continue;
2379 }
2380 if (!prefixcmp(arg, "--compression=")) {
2381 char *end;
2382 int level = strtoul(arg+14, &end, 0);
2383 if (!arg[14] || *end)
2384 usage(pack_usage);
2385 if (level == -1)
2386 level = Z_DEFAULT_COMPRESSION;
2387 else if (level < 0 || level > Z_BEST_COMPRESSION)
2388 die("bad pack compression level %d", level);
2389 pack_compression_level = level;
2390 continue;
2391 }
2392 if (!prefixcmp(arg, "--max-pack-size=")) {
2393 pack_size_limit_cfg = 0;
2394 if (!git_parse_ulong(arg+16, &pack_size_limit))
2395 usage(pack_usage);
2396 continue;
2397 }
2398 if (!prefixcmp(arg, "--window=")) {
2399 char *end;
2400 window = strtoul(arg+9, &end, 0);
2401 if (!arg[9] || *end)
2402 usage(pack_usage);
2403 continue;
2404 }
2405 if (!prefixcmp(arg, "--window-memory=")) {
2406 if (!git_parse_ulong(arg+16, &window_memory_limit))
2407 usage(pack_usage);
2408 continue;
2409 }
2410 if (!prefixcmp(arg, "--threads=")) {
2411 char *end;
2412 delta_search_threads = strtoul(arg+10, &end, 0);
2413 if (!arg[10] || *end || delta_search_threads < 0)
2414 usage(pack_usage);
2415 #ifdef NO_PTHREADS
2416 if (delta_search_threads != 1)
2417 warning("no threads support, "
2418 "ignoring %s", arg);
2419 #endif
2420 continue;
2421 }
2422 if (!prefixcmp(arg, "--depth=")) {
2423 char *end;
2424 depth = strtoul(arg+8, &end, 0);
2425 if (!arg[8] || *end)
2426 usage(pack_usage);
2427 continue;
2428 }
2429 if (!strcmp("--progress", arg)) {
2430 progress = 1;
2431 continue;
2432 }
2433 if (!strcmp("--all-progress", arg)) {
2434 progress = 2;
2435 continue;
2436 }
2437 if (!strcmp("--all-progress-implied", arg)) {
2438 all_progress_implied = 1;
2439 continue;
2440 }
2441 if (!strcmp("-q", arg)) {
2442 progress = 0;
2443 continue;
2444 }
2445 if (!strcmp("--no-reuse-delta", arg)) {
2446 reuse_delta = 0;
2447 continue;
2448 }
2449 if (!strcmp("--no-reuse-object", arg)) {
2450 reuse_object = reuse_delta = 0;
2451 continue;
2452 }
2453 if (!strcmp("--delta-base-offset", arg)) {
2454 allow_ofs_delta = 1;
2455 continue;
2456 }
2457 if (!strcmp("--stdout", arg)) {
2458 pack_to_stdout = 1;
2459 continue;
2460 }
2461 if (!strcmp("--revs", arg)) {
2462 use_internal_rev_list = 1;
2463 continue;
2464 }
2465 if (!strcmp("--keep-unreachable", arg)) {
2466 keep_unreachable = 1;
2467 continue;
2468 }
2469 if (!strcmp("--unpack-unreachable", arg)) {
2470 unpack_unreachable = 1;
2471 continue;
2472 }
2473 if (!strcmp("--include-tag", arg)) {
2474 include_tag = 1;
2475 continue;
2476 }
2477 if (!strcmp("--unpacked", arg) ||
2478 !strcmp("--reflog", arg) ||
2479 !strcmp("--all", arg)) {
2480 use_internal_rev_list = 1;
2481 if (rp_ac >= rp_ac_alloc - 1) {
2482 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2483 rp_av = xrealloc(rp_av,
2484 rp_ac_alloc * sizeof(*rp_av));
2485 }
2486 rp_av[rp_ac++] = arg;
2487 continue;
2488 }
2489 if (!strcmp("--thin", arg)) {
2490 use_internal_rev_list = 1;
2491 thin = 1;
2492 rp_av[1] = "--objects-edge";
2493 continue;
2494 }
2495 if (!prefixcmp(arg, "--index-version=")) {
2496 char *c;
2497 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2498 if (pack_idx_opts.version > 2)
2499 die("bad %s", arg);
2500 if (*c == ',')
2501 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2502 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2503 die("bad %s", arg);
2504 continue;
2505 }
2506 if (!strcmp(arg, "--keep-true-parents")) {
2507 grafts_replace_parents = 0;
2508 continue;
2509 }
2510 usage(pack_usage);
2511 }
2512
2513 /* Traditionally "pack-objects [options] base extra" failed;
2514 * we would however want to take refs parameter that would
2515 * have been given to upstream rev-list ourselves, which means
2516 * we somehow want to say what the base name is. So the
2517 * syntax would be:
2518 *
2519 * pack-objects [options] base <refs...>
2520 *
2521 * in other words, we would treat the first non-option as the
2522 * base_name and send everything else to the internal revision
2523 * walker.
2524 */
2525
2526 if (!pack_to_stdout)
2527 base_name = argv[i++];
2528
2529 if (pack_to_stdout != !base_name)
2530 usage(pack_usage);
2531
2532 if (!pack_to_stdout && !pack_size_limit)
2533 pack_size_limit = pack_size_limit_cfg;
2534 if (pack_to_stdout && pack_size_limit)
2535 die("--max-pack-size cannot be used to build a pack for transfer.");
2536 if (pack_size_limit && pack_size_limit < 1024*1024) {
2537 warning("minimum pack size limit is 1 MiB");
2538 pack_size_limit = 1024*1024;
2539 }
2540
2541 if (!pack_to_stdout && thin)
2542 die("--thin cannot be used to build an indexable pack.");
2543
2544 if (keep_unreachable && unpack_unreachable)
2545 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2546
2547 if (progress && all_progress_implied)
2548 progress = 2;
2549
2550 prepare_packed_git();
2551
2552 if (progress)
2553 progress_state = start_progress("Counting objects", 0);
2554 if (!use_internal_rev_list)
2555 read_object_list_from_stdin();
2556 else {
2557 rp_av[rp_ac] = NULL;
2558 get_object_list(rp_ac, rp_av);
2559 }
2560 cleanup_preferred_base();
2561 if (include_tag && nr_result)
2562 for_each_ref(add_ref_tag, NULL);
2563 stop_progress(&progress_state);
2564
2565 if (non_empty && !nr_result)
2566 return 0;
2567 if (nr_result)
2568 prepare_pack(window, depth);
2569 write_pack_file();
2570 if (progress)
2571 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2572 " reused %"PRIu32" (delta %"PRIu32")\n",
2573 written, written_delta, reused, reused_delta);
2574 return 0;
2575 }
git with Jakub Narębski modifications