diff: do not use null sha1 as a sentinel value
[git/git.git] / builtin / pack-objects.c
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;
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 uint32_t nr_remaining = nr_result;
642 time_t last_mtime = 0;
643 struct object_entry **write_order;
644
645 if (progress > pack_to_stdout)
646 progress_state = start_progress("Writing objects", nr_result);
647 written_list = xmalloc(nr_objects * sizeof(*written_list));
648 write_order = compute_write_order();
649
650 do {
651 unsigned char sha1[20];
652 char *pack_tmp_name = NULL;
653
654 if (pack_to_stdout)
655 f = sha1fd_throughput(1, "<stdout>", progress_state);
656 else
657 f = create_tmp_packfile(&pack_tmp_name);
658
659 offset = write_pack_header(f, nr_remaining);
660 if (!offset)
661 die_errno("unable to write pack header");
662 nr_written = 0;
663 for (; i < nr_objects; i++) {
664 struct object_entry *e = write_order[i];
665 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
666 break;
667 display_progress(progress_state, written);
668 }
669
670 /*
671 * Did we write the wrong # entries in the header?
672 * If so, rewrite it like in fast-import
673 */
674 if (pack_to_stdout) {
675 sha1close(f, sha1, CSUM_CLOSE);
676 } else if (nr_written == nr_remaining) {
677 sha1close(f, sha1, CSUM_FSYNC);
678 } else {
679 int fd = sha1close(f, sha1, 0);
680 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
681 nr_written, sha1, offset);
682 close(fd);
683 }
684
685 if (!pack_to_stdout) {
686 struct stat st;
687 char tmpname[PATH_MAX];
688
689 /*
690 * Packs are runtime accessed in their mtime
691 * order since newer packs are more likely to contain
692 * younger objects. So if we are creating multiple
693 * packs then we should modify the mtime of later ones
694 * to preserve this property.
695 */
696 if (stat(pack_tmp_name, &st) < 0) {
697 warning("failed to stat %s: %s",
698 pack_tmp_name, strerror(errno));
699 } else if (!last_mtime) {
700 last_mtime = st.st_mtime;
701 } else {
702 struct utimbuf utb;
703 utb.actime = st.st_atime;
704 utb.modtime = --last_mtime;
705 if (utime(pack_tmp_name, &utb) < 0)
706 warning("failed utime() on %s: %s",
707 tmpname, strerror(errno));
708 }
709
710 /* Enough space for "-<sha-1>.pack"? */
711 if (sizeof(tmpname) <= strlen(base_name) + 50)
712 die("pack base name '%s' too long", base_name);
713 snprintf(tmpname, sizeof(tmpname), "%s-", base_name);
714 finish_tmp_packfile(tmpname, pack_tmp_name,
715 written_list, nr_written,
716 &pack_idx_opts, sha1);
717 free(pack_tmp_name);
718 puts(sha1_to_hex(sha1));
719 }
720
721 /* mark written objects as written to previous pack */
722 for (j = 0; j < nr_written; j++) {
723 written_list[j]->offset = (off_t)-1;
724 }
725 nr_remaining -= nr_written;
726 } while (nr_remaining && i < nr_objects);
727
728 free(written_list);
729 free(write_order);
730 stop_progress(&progress_state);
731 if (written != nr_result)
732 die("wrote %"PRIu32" objects while expecting %"PRIu32,
733 written, nr_result);
734 }
735
736 static int locate_object_entry_hash(const unsigned char *sha1)
737 {
738 int i;
739 unsigned int ui;
740 memcpy(&ui, sha1, sizeof(unsigned int));
741 i = ui % object_ix_hashsz;
742 while (0 < object_ix[i]) {
743 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
744 return i;
745 if (++i == object_ix_hashsz)
746 i = 0;
747 }
748 return -1 - i;
749 }
750
751 static struct object_entry *locate_object_entry(const unsigned char *sha1)
752 {
753 int i;
754
755 if (!object_ix_hashsz)
756 return NULL;
757
758 i = locate_object_entry_hash(sha1);
759 if (0 <= i)
760 return &objects[object_ix[i]-1];
761 return NULL;
762 }
763
764 static void rehash_objects(void)
765 {
766 uint32_t i;
767 struct object_entry *oe;
768
769 object_ix_hashsz = nr_objects * 3;
770 if (object_ix_hashsz < 1024)
771 object_ix_hashsz = 1024;
772 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
773 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
774 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
775 int ix = locate_object_entry_hash(oe->idx.sha1);
776 if (0 <= ix)
777 continue;
778 ix = -1 - ix;
779 object_ix[ix] = i + 1;
780 }
781 }
782
783 static unsigned name_hash(const char *name)
784 {
785 unsigned c, hash = 0;
786
787 if (!name)
788 return 0;
789
790 /*
791 * This effectively just creates a sortable number from the
792 * last sixteen non-whitespace characters. Last characters
793 * count "most", so things that end in ".c" sort together.
794 */
795 while ((c = *name++) != 0) {
796 if (isspace(c))
797 continue;
798 hash = (hash >> 2) + (c << 24);
799 }
800 return hash;
801 }
802
803 static void setup_delta_attr_check(struct git_attr_check *check)
804 {
805 static struct git_attr *attr_delta;
806
807 if (!attr_delta)
808 attr_delta = git_attr("delta");
809
810 check[0].attr = attr_delta;
811 }
812
813 static int no_try_delta(const char *path)
814 {
815 struct git_attr_check check[1];
816
817 setup_delta_attr_check(check);
818 if (git_check_attr(path, ARRAY_SIZE(check), check))
819 return 0;
820 if (ATTR_FALSE(check->value))
821 return 1;
822 return 0;
823 }
824
825 static int add_object_entry(const unsigned char *sha1, enum object_type type,
826 const char *name, int exclude)
827 {
828 struct object_entry *entry;
829 struct packed_git *p, *found_pack = NULL;
830 off_t found_offset = 0;
831 int ix;
832 unsigned hash = name_hash(name);
833
834 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
835 if (ix >= 0) {
836 if (exclude) {
837 entry = objects + object_ix[ix] - 1;
838 if (!entry->preferred_base)
839 nr_result--;
840 entry->preferred_base = 1;
841 }
842 return 0;
843 }
844
845 if (!exclude && local && has_loose_object_nonlocal(sha1))
846 return 0;
847
848 for (p = packed_git; p; p = p->next) {
849 off_t offset = find_pack_entry_one(sha1, p);
850 if (offset) {
851 if (!found_pack) {
852 if (!is_pack_valid(p)) {
853 warning("packfile %s cannot be accessed", p->pack_name);
854 continue;
855 }
856 found_offset = offset;
857 found_pack = p;
858 }
859 if (exclude)
860 break;
861 if (incremental)
862 return 0;
863 if (local && !p->pack_local)
864 return 0;
865 if (ignore_packed_keep && p->pack_local && p->pack_keep)
866 return 0;
867 }
868 }
869
870 if (nr_objects >= nr_alloc) {
871 nr_alloc = (nr_alloc + 1024) * 3 / 2;
872 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
873 }
874
875 entry = objects + nr_objects++;
876 memset(entry, 0, sizeof(*entry));
877 hashcpy(entry->idx.sha1, sha1);
878 entry->hash = hash;
879 if (type)
880 entry->type = type;
881 if (exclude)
882 entry->preferred_base = 1;
883 else
884 nr_result++;
885 if (found_pack) {
886 entry->in_pack = found_pack;
887 entry->in_pack_offset = found_offset;
888 }
889
890 if (object_ix_hashsz * 3 <= nr_objects * 4)
891 rehash_objects();
892 else
893 object_ix[-1 - ix] = nr_objects;
894
895 display_progress(progress_state, nr_objects);
896
897 if (name && no_try_delta(name))
898 entry->no_try_delta = 1;
899
900 return 1;
901 }
902
903 struct pbase_tree_cache {
904 unsigned char sha1[20];
905 int ref;
906 int temporary;
907 void *tree_data;
908 unsigned long tree_size;
909 };
910
911 static struct pbase_tree_cache *(pbase_tree_cache[256]);
912 static int pbase_tree_cache_ix(const unsigned char *sha1)
913 {
914 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
915 }
916 static int pbase_tree_cache_ix_incr(int ix)
917 {
918 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
919 }
920
921 static struct pbase_tree {
922 struct pbase_tree *next;
923 /* This is a phony "cache" entry; we are not
924 * going to evict it nor find it through _get()
925 * mechanism -- this is for the toplevel node that
926 * would almost always change with any commit.
927 */
928 struct pbase_tree_cache pcache;
929 } *pbase_tree;
930
931 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
932 {
933 struct pbase_tree_cache *ent, *nent;
934 void *data;
935 unsigned long size;
936 enum object_type type;
937 int neigh;
938 int my_ix = pbase_tree_cache_ix(sha1);
939 int available_ix = -1;
940
941 /* pbase-tree-cache acts as a limited hashtable.
942 * your object will be found at your index or within a few
943 * slots after that slot if it is cached.
944 */
945 for (neigh = 0; neigh < 8; neigh++) {
946 ent = pbase_tree_cache[my_ix];
947 if (ent && !hashcmp(ent->sha1, sha1)) {
948 ent->ref++;
949 return ent;
950 }
951 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
952 ((0 <= available_ix) &&
953 (!ent && pbase_tree_cache[available_ix])))
954 available_ix = my_ix;
955 if (!ent)
956 break;
957 my_ix = pbase_tree_cache_ix_incr(my_ix);
958 }
959
960 /* Did not find one. Either we got a bogus request or
961 * we need to read and perhaps cache.
962 */
963 data = read_sha1_file(sha1, &type, &size);
964 if (!data)
965 return NULL;
966 if (type != OBJ_TREE) {
967 free(data);
968 return NULL;
969 }
970
971 /* We need to either cache or return a throwaway copy */
972
973 if (available_ix < 0)
974 ent = NULL;
975 else {
976 ent = pbase_tree_cache[available_ix];
977 my_ix = available_ix;
978 }
979
980 if (!ent) {
981 nent = xmalloc(sizeof(*nent));
982 nent->temporary = (available_ix < 0);
983 }
984 else {
985 /* evict and reuse */
986 free(ent->tree_data);
987 nent = ent;
988 }
989 hashcpy(nent->sha1, sha1);
990 nent->tree_data = data;
991 nent->tree_size = size;
992 nent->ref = 1;
993 if (!nent->temporary)
994 pbase_tree_cache[my_ix] = nent;
995 return nent;
996 }
997
998 static void pbase_tree_put(struct pbase_tree_cache *cache)
999 {
1000 if (!cache->temporary) {
1001 cache->ref--;
1002 return;
1003 }
1004 free(cache->tree_data);
1005 free(cache);
1006 }
1007
1008 static int name_cmp_len(const char *name)
1009 {
1010 int i;
1011 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1012 ;
1013 return i;
1014 }
1015
1016 static void add_pbase_object(struct tree_desc *tree,
1017 const char *name,
1018 int cmplen,
1019 const char *fullname)
1020 {
1021 struct name_entry entry;
1022 int cmp;
1023
1024 while (tree_entry(tree,&entry)) {
1025 if (S_ISGITLINK(entry.mode))
1026 continue;
1027 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1028 memcmp(name, entry.path, cmplen);
1029 if (cmp > 0)
1030 continue;
1031 if (cmp < 0)
1032 return;
1033 if (name[cmplen] != '/') {
1034 add_object_entry(entry.sha1,
1035 object_type(entry.mode),
1036 fullname, 1);
1037 return;
1038 }
1039 if (S_ISDIR(entry.mode)) {
1040 struct tree_desc sub;
1041 struct pbase_tree_cache *tree;
1042 const char *down = name+cmplen+1;
1043 int downlen = name_cmp_len(down);
1044
1045 tree = pbase_tree_get(entry.sha1);
1046 if (!tree)
1047 return;
1048 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1049
1050 add_pbase_object(&sub, down, downlen, fullname);
1051 pbase_tree_put(tree);
1052 }
1053 }
1054 }
1055
1056 static unsigned *done_pbase_paths;
1057 static int done_pbase_paths_num;
1058 static int done_pbase_paths_alloc;
1059 static int done_pbase_path_pos(unsigned hash)
1060 {
1061 int lo = 0;
1062 int hi = done_pbase_paths_num;
1063 while (lo < hi) {
1064 int mi = (hi + lo) / 2;
1065 if (done_pbase_paths[mi] == hash)
1066 return mi;
1067 if (done_pbase_paths[mi] < hash)
1068 hi = mi;
1069 else
1070 lo = mi + 1;
1071 }
1072 return -lo-1;
1073 }
1074
1075 static int check_pbase_path(unsigned hash)
1076 {
1077 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1078 if (0 <= pos)
1079 return 1;
1080 pos = -pos - 1;
1081 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1082 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1083 done_pbase_paths = xrealloc(done_pbase_paths,
1084 done_pbase_paths_alloc *
1085 sizeof(unsigned));
1086 }
1087 done_pbase_paths_num++;
1088 if (pos < done_pbase_paths_num)
1089 memmove(done_pbase_paths + pos + 1,
1090 done_pbase_paths + pos,
1091 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1092 done_pbase_paths[pos] = hash;
1093 return 0;
1094 }
1095
1096 static void add_preferred_base_object(const char *name)
1097 {
1098 struct pbase_tree *it;
1099 int cmplen;
1100 unsigned hash = name_hash(name);
1101
1102 if (!num_preferred_base || check_pbase_path(hash))
1103 return;
1104
1105 cmplen = name_cmp_len(name);
1106 for (it = pbase_tree; it; it = it->next) {
1107 if (cmplen == 0) {
1108 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1109 }
1110 else {
1111 struct tree_desc tree;
1112 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1113 add_pbase_object(&tree, name, cmplen, name);
1114 }
1115 }
1116 }
1117
1118 static void add_preferred_base(unsigned char *sha1)
1119 {
1120 struct pbase_tree *it;
1121 void *data;
1122 unsigned long size;
1123 unsigned char tree_sha1[20];
1124
1125 if (window <= num_preferred_base++)
1126 return;
1127
1128 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1129 if (!data)
1130 return;
1131
1132 for (it = pbase_tree; it; it = it->next) {
1133 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1134 free(data);
1135 return;
1136 }
1137 }
1138
1139 it = xcalloc(1, sizeof(*it));
1140 it->next = pbase_tree;
1141 pbase_tree = it;
1142
1143 hashcpy(it->pcache.sha1, tree_sha1);
1144 it->pcache.tree_data = data;
1145 it->pcache.tree_size = size;
1146 }
1147
1148 static void cleanup_preferred_base(void)
1149 {
1150 struct pbase_tree *it;
1151 unsigned i;
1152
1153 it = pbase_tree;
1154 pbase_tree = NULL;
1155 while (it) {
1156 struct pbase_tree *this = it;
1157 it = this->next;
1158 free(this->pcache.tree_data);
1159 free(this);
1160 }
1161
1162 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1163 if (!pbase_tree_cache[i])
1164 continue;
1165 free(pbase_tree_cache[i]->tree_data);
1166 free(pbase_tree_cache[i]);
1167 pbase_tree_cache[i] = NULL;
1168 }
1169
1170 free(done_pbase_paths);
1171 done_pbase_paths = NULL;
1172 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1173 }
1174
1175 static void check_object(struct object_entry *entry)
1176 {
1177 if (entry->in_pack) {
1178 struct packed_git *p = entry->in_pack;
1179 struct pack_window *w_curs = NULL;
1180 const unsigned char *base_ref = NULL;
1181 struct object_entry *base_entry;
1182 unsigned long used, used_0;
1183 unsigned long avail;
1184 off_t ofs;
1185 unsigned char *buf, c;
1186
1187 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1188
1189 /*
1190 * We want in_pack_type even if we do not reuse delta
1191 * since non-delta representations could still be reused.
1192 */
1193 used = unpack_object_header_buffer(buf, avail,
1194 &entry->in_pack_type,
1195 &entry->size);
1196 if (used == 0)
1197 goto give_up;
1198
1199 /*
1200 * Determine if this is a delta and if so whether we can
1201 * reuse it or not. Otherwise let's find out as cheaply as
1202 * possible what the actual type and size for this object is.
1203 */
1204 switch (entry->in_pack_type) {
1205 default:
1206 /* Not a delta hence we've already got all we need. */
1207 entry->type = entry->in_pack_type;
1208 entry->in_pack_header_size = used;
1209 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1210 goto give_up;
1211 unuse_pack(&w_curs);
1212 return;
1213 case OBJ_REF_DELTA:
1214 if (reuse_delta && !entry->preferred_base)
1215 base_ref = use_pack(p, &w_curs,
1216 entry->in_pack_offset + used, NULL);
1217 entry->in_pack_header_size = used + 20;
1218 break;
1219 case OBJ_OFS_DELTA:
1220 buf = use_pack(p, &w_curs,
1221 entry->in_pack_offset + used, NULL);
1222 used_0 = 0;
1223 c = buf[used_0++];
1224 ofs = c & 127;
1225 while (c & 128) {
1226 ofs += 1;
1227 if (!ofs || MSB(ofs, 7)) {
1228 error("delta base offset overflow in pack for %s",
1229 sha1_to_hex(entry->idx.sha1));
1230 goto give_up;
1231 }
1232 c = buf[used_0++];
1233 ofs = (ofs << 7) + (c & 127);
1234 }
1235 ofs = entry->in_pack_offset - ofs;
1236 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1237 error("delta base offset out of bound for %s",
1238 sha1_to_hex(entry->idx.sha1));
1239 goto give_up;
1240 }
1241 if (reuse_delta && !entry->preferred_base) {
1242 struct revindex_entry *revidx;
1243 revidx = find_pack_revindex(p, ofs);
1244 if (!revidx)
1245 goto give_up;
1246 base_ref = nth_packed_object_sha1(p, revidx->nr);
1247 }
1248 entry->in_pack_header_size = used + used_0;
1249 break;
1250 }
1251
1252 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1253 /*
1254 * If base_ref was set above that means we wish to
1255 * reuse delta data, and we even found that base
1256 * in the list of objects we want to pack. Goodie!
1257 *
1258 * Depth value does not matter - find_deltas() will
1259 * never consider reused delta as the base object to
1260 * deltify other objects against, in order to avoid
1261 * circular deltas.
1262 */
1263 entry->type = entry->in_pack_type;
1264 entry->delta = base_entry;
1265 entry->delta_size = entry->size;
1266 entry->delta_sibling = base_entry->delta_child;
1267 base_entry->delta_child = entry;
1268 unuse_pack(&w_curs);
1269 return;
1270 }
1271
1272 if (entry->type) {
1273 /*
1274 * This must be a delta and we already know what the
1275 * final object type is. Let's extract the actual
1276 * object size from the delta header.
1277 */
1278 entry->size = get_size_from_delta(p, &w_curs,
1279 entry->in_pack_offset + entry->in_pack_header_size);
1280 if (entry->size == 0)
1281 goto give_up;
1282 unuse_pack(&w_curs);
1283 return;
1284 }
1285
1286 /*
1287 * No choice but to fall back to the recursive delta walk
1288 * with sha1_object_info() to find about the object type
1289 * at this point...
1290 */
1291 give_up:
1292 unuse_pack(&w_curs);
1293 }
1294
1295 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1296 /*
1297 * The error condition is checked in prepare_pack(). This is
1298 * to permit a missing preferred base object to be ignored
1299 * as a preferred base. Doing so can result in a larger
1300 * pack file, but the transfer will still take place.
1301 */
1302 }
1303
1304 static int pack_offset_sort(const void *_a, const void *_b)
1305 {
1306 const struct object_entry *a = *(struct object_entry **)_a;
1307 const struct object_entry *b = *(struct object_entry **)_b;
1308
1309 /* avoid filesystem trashing with loose objects */
1310 if (!a->in_pack && !b->in_pack)
1311 return hashcmp(a->idx.sha1, b->idx.sha1);
1312
1313 if (a->in_pack < b->in_pack)
1314 return -1;
1315 if (a->in_pack > b->in_pack)
1316 return 1;
1317 return a->in_pack_offset < b->in_pack_offset ? -1 :
1318 (a->in_pack_offset > b->in_pack_offset);
1319 }
1320
1321 static void get_object_details(void)
1322 {
1323 uint32_t i;
1324 struct object_entry **sorted_by_offset;
1325
1326 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1327 for (i = 0; i < nr_objects; i++)
1328 sorted_by_offset[i] = objects + i;
1329 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1330
1331 for (i = 0; i < nr_objects; i++) {
1332 struct object_entry *entry = sorted_by_offset[i];
1333 check_object(entry);
1334 if (big_file_threshold <= entry->size)
1335 entry->no_try_delta = 1;
1336 }
1337
1338 free(sorted_by_offset);
1339 }
1340
1341 /*
1342 * We search for deltas in a list sorted by type, by filename hash, and then
1343 * by size, so that we see progressively smaller and smaller files.
1344 * That's because we prefer deltas to be from the bigger file
1345 * to the smaller -- deletes are potentially cheaper, but perhaps
1346 * more importantly, the bigger file is likely the more recent
1347 * one. The deepest deltas are therefore the oldest objects which are
1348 * less susceptible to be accessed often.
1349 */
1350 static int type_size_sort(const void *_a, const void *_b)
1351 {
1352 const struct object_entry *a = *(struct object_entry **)_a;
1353 const struct object_entry *b = *(struct object_entry **)_b;
1354
1355 if (a->type > b->type)
1356 return -1;
1357 if (a->type < b->type)
1358 return 1;
1359 if (a->hash > b->hash)
1360 return -1;
1361 if (a->hash < b->hash)
1362 return 1;
1363 if (a->preferred_base > b->preferred_base)
1364 return -1;
1365 if (a->preferred_base < b->preferred_base)
1366 return 1;
1367 if (a->size > b->size)
1368 return -1;
1369 if (a->size < b->size)
1370 return 1;
1371 return a < b ? -1 : (a > b); /* newest first */
1372 }
1373
1374 struct unpacked {
1375 struct object_entry *entry;
1376 void *data;
1377 struct delta_index *index;
1378 unsigned depth;
1379 };
1380
1381 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1382 unsigned long delta_size)
1383 {
1384 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1385 return 0;
1386
1387 if (delta_size < cache_max_small_delta_size)
1388 return 1;
1389
1390 /* cache delta, if objects are large enough compared to delta size */
1391 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1392 return 1;
1393
1394 return 0;
1395 }
1396
1397 #ifndef NO_PTHREADS
1398
1399 static pthread_mutex_t read_mutex;
1400 #define read_lock() pthread_mutex_lock(&read_mutex)
1401 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1402
1403 static pthread_mutex_t cache_mutex;
1404 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1405 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1406
1407 static pthread_mutex_t progress_mutex;
1408 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1409 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1410
1411 #else
1412
1413 #define read_lock() (void)0
1414 #define read_unlock() (void)0
1415 #define cache_lock() (void)0
1416 #define cache_unlock() (void)0
1417 #define progress_lock() (void)0
1418 #define progress_unlock() (void)0
1419
1420 #endif
1421
1422 static int try_delta(struct unpacked *trg, struct unpacked *src,
1423 unsigned max_depth, unsigned long *mem_usage)
1424 {
1425 struct object_entry *trg_entry = trg->entry;
1426 struct object_entry *src_entry = src->entry;
1427 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1428 unsigned ref_depth;
1429 enum object_type type;
1430 void *delta_buf;
1431
1432 /* Don't bother doing diffs between different types */
1433 if (trg_entry->type != src_entry->type)
1434 return -1;
1435
1436 /*
1437 * We do not bother to try a delta that we discarded on an
1438 * earlier try, but only when reusing delta data. Note that
1439 * src_entry that is marked as the preferred_base should always
1440 * be considered, as even if we produce a suboptimal delta against
1441 * it, we will still save the transfer cost, as we already know
1442 * the other side has it and we won't send src_entry at all.
1443 */
1444 if (reuse_delta && trg_entry->in_pack &&
1445 trg_entry->in_pack == src_entry->in_pack &&
1446 !src_entry->preferred_base &&
1447 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1448 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1449 return 0;
1450
1451 /* Let's not bust the allowed depth. */
1452 if (src->depth >= max_depth)
1453 return 0;
1454
1455 /* Now some size filtering heuristics. */
1456 trg_size = trg_entry->size;
1457 if (!trg_entry->delta) {
1458 max_size = trg_size/2 - 20;
1459 ref_depth = 1;
1460 } else {
1461 max_size = trg_entry->delta_size;
1462 ref_depth = trg->depth;
1463 }
1464 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1465 (max_depth - ref_depth + 1);
1466 if (max_size == 0)
1467 return 0;
1468 src_size = src_entry->size;
1469 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1470 if (sizediff >= max_size)
1471 return 0;
1472 if (trg_size < src_size / 32)
1473 return 0;
1474
1475 /* Load data if not already done */
1476 if (!trg->data) {
1477 read_lock();
1478 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1479 read_unlock();
1480 if (!trg->data)
1481 die("object %s cannot be read",
1482 sha1_to_hex(trg_entry->idx.sha1));
1483 if (sz != trg_size)
1484 die("object %s inconsistent object length (%lu vs %lu)",
1485 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1486 *mem_usage += sz;
1487 }
1488 if (!src->data) {
1489 read_lock();
1490 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1491 read_unlock();
1492 if (!src->data) {
1493 if (src_entry->preferred_base) {
1494 static int warned = 0;
1495 if (!warned++)
1496 warning("object %s cannot be read",
1497 sha1_to_hex(src_entry->idx.sha1));
1498 /*
1499 * Those objects are not included in the
1500 * resulting pack. Be resilient and ignore
1501 * them if they can't be read, in case the
1502 * pack could be created nevertheless.
1503 */
1504 return 0;
1505 }
1506 die("object %s cannot be read",
1507 sha1_to_hex(src_entry->idx.sha1));
1508 }
1509 if (sz != src_size)
1510 die("object %s inconsistent object length (%lu vs %lu)",
1511 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1512 *mem_usage += sz;
1513 }
1514 if (!src->index) {
1515 src->index = create_delta_index(src->data, src_size);
1516 if (!src->index) {
1517 static int warned = 0;
1518 if (!warned++)
1519 warning("suboptimal pack - out of memory");
1520 return 0;
1521 }
1522 *mem_usage += sizeof_delta_index(src->index);
1523 }
1524
1525 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1526 if (!delta_buf)
1527 return 0;
1528
1529 if (trg_entry->delta) {
1530 /* Prefer only shallower same-sized deltas. */
1531 if (delta_size == trg_entry->delta_size &&
1532 src->depth + 1 >= trg->depth) {
1533 free(delta_buf);
1534 return 0;
1535 }
1536 }
1537
1538 /*
1539 * Handle memory allocation outside of the cache
1540 * accounting lock. Compiler will optimize the strangeness
1541 * away when NO_PTHREADS is defined.
1542 */
1543 free(trg_entry->delta_data);
1544 cache_lock();
1545 if (trg_entry->delta_data) {
1546 delta_cache_size -= trg_entry->delta_size;
1547 trg_entry->delta_data = NULL;
1548 }
1549 if (delta_cacheable(src_size, trg_size, delta_size)) {
1550 delta_cache_size += delta_size;
1551 cache_unlock();
1552 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1553 } else {
1554 cache_unlock();
1555 free(delta_buf);
1556 }
1557
1558 trg_entry->delta = src_entry;
1559 trg_entry->delta_size = delta_size;
1560 trg->depth = src->depth + 1;
1561
1562 return 1;
1563 }
1564
1565 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1566 {
1567 struct object_entry *child = me->delta_child;
1568 unsigned int m = n;
1569 while (child) {
1570 unsigned int c = check_delta_limit(child, n + 1);
1571 if (m < c)
1572 m = c;
1573 child = child->delta_sibling;
1574 }
1575 return m;
1576 }
1577
1578 static unsigned long free_unpacked(struct unpacked *n)
1579 {
1580 unsigned long freed_mem = sizeof_delta_index(n->index);
1581 free_delta_index(n->index);
1582 n->index = NULL;
1583 if (n->data) {
1584 freed_mem += n->entry->size;
1585 free(n->data);
1586 n->data = NULL;
1587 }
1588 n->entry = NULL;
1589 n->depth = 0;
1590 return freed_mem;
1591 }
1592
1593 static void find_deltas(struct object_entry **list, unsigned *list_size,
1594 int window, int depth, unsigned *processed)
1595 {
1596 uint32_t i, idx = 0, count = 0;
1597 struct unpacked *array;
1598 unsigned long mem_usage = 0;
1599
1600 array = xcalloc(window, sizeof(struct unpacked));
1601
1602 for (;;) {
1603 struct object_entry *entry;
1604 struct unpacked *n = array + idx;
1605 int j, max_depth, best_base = -1;
1606
1607 progress_lock();
1608 if (!*list_size) {
1609 progress_unlock();
1610 break;
1611 }
1612 entry = *list++;
1613 (*list_size)--;
1614 if (!entry->preferred_base) {
1615 (*processed)++;
1616 display_progress(progress_state, *processed);
1617 }
1618 progress_unlock();
1619
1620 mem_usage -= free_unpacked(n);
1621 n->entry = entry;
1622
1623 while (window_memory_limit &&
1624 mem_usage > window_memory_limit &&
1625 count > 1) {
1626 uint32_t tail = (idx + window - count) % window;
1627 mem_usage -= free_unpacked(array + tail);
1628 count--;
1629 }
1630
1631 /* We do not compute delta to *create* objects we are not
1632 * going to pack.
1633 */
1634 if (entry->preferred_base)
1635 goto next;
1636
1637 /*
1638 * If the current object is at pack edge, take the depth the
1639 * objects that depend on the current object into account
1640 * otherwise they would become too deep.
1641 */
1642 max_depth = depth;
1643 if (entry->delta_child) {
1644 max_depth -= check_delta_limit(entry, 0);
1645 if (max_depth <= 0)
1646 goto next;
1647 }
1648
1649 j = window;
1650 while (--j > 0) {
1651 int ret;
1652 uint32_t other_idx = idx + j;
1653 struct unpacked *m;
1654 if (other_idx >= window)
1655 other_idx -= window;
1656 m = array + other_idx;
1657 if (!m->entry)
1658 break;
1659 ret = try_delta(n, m, max_depth, &mem_usage);
1660 if (ret < 0)
1661 break;
1662 else if (ret > 0)
1663 best_base = other_idx;
1664 }
1665
1666 /*
1667 * If we decided to cache the delta data, then it is best
1668 * to compress it right away. First because we have to do
1669 * it anyway, and doing it here while we're threaded will
1670 * save a lot of time in the non threaded write phase,
1671 * as well as allow for caching more deltas within
1672 * the same cache size limit.
1673 * ...
1674 * But only if not writing to stdout, since in that case
1675 * the network is most likely throttling writes anyway,
1676 * and therefore it is best to go to the write phase ASAP
1677 * instead, as we can afford spending more time compressing
1678 * between writes at that moment.
1679 */
1680 if (entry->delta_data && !pack_to_stdout) {
1681 entry->z_delta_size = do_compress(&entry->delta_data,
1682 entry->delta_size);
1683 cache_lock();
1684 delta_cache_size -= entry->delta_size;
1685 delta_cache_size += entry->z_delta_size;
1686 cache_unlock();
1687 }
1688
1689 /* if we made n a delta, and if n is already at max
1690 * depth, leaving it in the window is pointless. we
1691 * should evict it first.
1692 */
1693 if (entry->delta && max_depth <= n->depth)
1694 continue;
1695
1696 /*
1697 * Move the best delta base up in the window, after the
1698 * currently deltified object, to keep it longer. It will
1699 * be the first base object to be attempted next.
1700 */
1701 if (entry->delta) {
1702 struct unpacked swap = array[best_base];
1703 int dist = (window + idx - best_base) % window;
1704 int dst = best_base;
1705 while (dist--) {
1706 int src = (dst + 1) % window;
1707 array[dst] = array[src];
1708 dst = src;
1709 }
1710 array[dst] = swap;
1711 }
1712
1713 next:
1714 idx++;
1715 if (count + 1 < window)
1716 count++;
1717 if (idx >= window)
1718 idx = 0;
1719 }
1720
1721 for (i = 0; i < window; ++i) {
1722 free_delta_index(array[i].index);
1723 free(array[i].data);
1724 }
1725 free(array);
1726 }
1727
1728 #ifndef NO_PTHREADS
1729
1730 static void try_to_free_from_threads(size_t size)
1731 {
1732 read_lock();
1733 release_pack_memory(size, -1);
1734 read_unlock();
1735 }
1736
1737 static try_to_free_t old_try_to_free_routine;
1738
1739 /*
1740 * The main thread waits on the condition that (at least) one of the workers
1741 * has stopped working (which is indicated in the .working member of
1742 * struct thread_params).
1743 * When a work thread has completed its work, it sets .working to 0 and
1744 * signals the main thread and waits on the condition that .data_ready
1745 * becomes 1.
1746 */
1747
1748 struct thread_params {
1749 pthread_t thread;
1750 struct object_entry **list;
1751 unsigned list_size;
1752 unsigned remaining;
1753 int window;
1754 int depth;
1755 int working;
1756 int data_ready;
1757 pthread_mutex_t mutex;
1758 pthread_cond_t cond;
1759 unsigned *processed;
1760 };
1761
1762 static pthread_cond_t progress_cond;
1763
1764 /*
1765 * Mutex and conditional variable can't be statically-initialized on Windows.
1766 */
1767 static void init_threaded_search(void)
1768 {
1769 init_recursive_mutex(&read_mutex);
1770 pthread_mutex_init(&cache_mutex, NULL);
1771 pthread_mutex_init(&progress_mutex, NULL);
1772 pthread_cond_init(&progress_cond, NULL);
1773 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1774 }
1775
1776 static void cleanup_threaded_search(void)
1777 {
1778 set_try_to_free_routine(old_try_to_free_routine);
1779 pthread_cond_destroy(&progress_cond);
1780 pthread_mutex_destroy(&read_mutex);
1781 pthread_mutex_destroy(&cache_mutex);
1782 pthread_mutex_destroy(&progress_mutex);
1783 }
1784
1785 static void *threaded_find_deltas(void *arg)
1786 {
1787 struct thread_params *me = arg;
1788
1789 while (me->remaining) {
1790 find_deltas(me->list, &me->remaining,
1791 me->window, me->depth, me->processed);
1792
1793 progress_lock();
1794 me->working = 0;
1795 pthread_cond_signal(&progress_cond);
1796 progress_unlock();
1797
1798 /*
1799 * We must not set ->data_ready before we wait on the
1800 * condition because the main thread may have set it to 1
1801 * before we get here. In order to be sure that new
1802 * work is available if we see 1 in ->data_ready, it
1803 * was initialized to 0 before this thread was spawned
1804 * and we reset it to 0 right away.
1805 */
1806 pthread_mutex_lock(&me->mutex);
1807 while (!me->data_ready)
1808 pthread_cond_wait(&me->cond, &me->mutex);
1809 me->data_ready = 0;
1810 pthread_mutex_unlock(&me->mutex);
1811 }
1812 /* leave ->working 1 so that this doesn't get more work assigned */
1813 return NULL;
1814 }
1815
1816 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1817 int window, int depth, unsigned *processed)
1818 {
1819 struct thread_params *p;
1820 int i, ret, active_threads = 0;
1821
1822 init_threaded_search();
1823
1824 if (!delta_search_threads) /* --threads=0 means autodetect */
1825 delta_search_threads = online_cpus();
1826 if (delta_search_threads <= 1) {
1827 find_deltas(list, &list_size, window, depth, processed);
1828 cleanup_threaded_search();
1829 return;
1830 }
1831 if (progress > pack_to_stdout)
1832 fprintf(stderr, "Delta compression using up to %d threads.\n",
1833 delta_search_threads);
1834 p = xcalloc(delta_search_threads, sizeof(*p));
1835
1836 /* Partition the work amongst work threads. */
1837 for (i = 0; i < delta_search_threads; i++) {
1838 unsigned sub_size = list_size / (delta_search_threads - i);
1839
1840 /* don't use too small segments or no deltas will be found */
1841 if (sub_size < 2*window && i+1 < delta_search_threads)
1842 sub_size = 0;
1843
1844 p[i].window = window;
1845 p[i].depth = depth;
1846 p[i].processed = processed;
1847 p[i].working = 1;
1848 p[i].data_ready = 0;
1849
1850 /* try to split chunks on "path" boundaries */
1851 while (sub_size && sub_size < list_size &&
1852 list[sub_size]->hash &&
1853 list[sub_size]->hash == list[sub_size-1]->hash)
1854 sub_size++;
1855
1856 p[i].list = list;
1857 p[i].list_size = sub_size;
1858 p[i].remaining = sub_size;
1859
1860 list += sub_size;
1861 list_size -= sub_size;
1862 }
1863
1864 /* Start work threads. */
1865 for (i = 0; i < delta_search_threads; i++) {
1866 if (!p[i].list_size)
1867 continue;
1868 pthread_mutex_init(&p[i].mutex, NULL);
1869 pthread_cond_init(&p[i].cond, NULL);
1870 ret = pthread_create(&p[i].thread, NULL,
1871 threaded_find_deltas, &p[i]);
1872 if (ret)
1873 die("unable to create thread: %s", strerror(ret));
1874 active_threads++;
1875 }
1876
1877 /*
1878 * Now let's wait for work completion. Each time a thread is done
1879 * with its work, we steal half of the remaining work from the
1880 * thread with the largest number of unprocessed objects and give
1881 * it to that newly idle thread. This ensure good load balancing
1882 * until the remaining object list segments are simply too short
1883 * to be worth splitting anymore.
1884 */
1885 while (active_threads) {
1886 struct thread_params *target = NULL;
1887 struct thread_params *victim = NULL;
1888 unsigned sub_size = 0;
1889
1890 progress_lock();
1891 for (;;) {
1892 for (i = 0; !target && i < delta_search_threads; i++)
1893 if (!p[i].working)
1894 target = &p[i];
1895 if (target)
1896 break;
1897 pthread_cond_wait(&progress_cond, &progress_mutex);
1898 }
1899
1900 for (i = 0; i < delta_search_threads; i++)
1901 if (p[i].remaining > 2*window &&
1902 (!victim || victim->remaining < p[i].remaining))
1903 victim = &p[i];
1904 if (victim) {
1905 sub_size = victim->remaining / 2;
1906 list = victim->list + victim->list_size - sub_size;
1907 while (sub_size && list[0]->hash &&
1908 list[0]->hash == list[-1]->hash) {
1909 list++;
1910 sub_size--;
1911 }
1912 if (!sub_size) {
1913 /*
1914 * It is possible for some "paths" to have
1915 * so many objects that no hash boundary
1916 * might be found. Let's just steal the
1917 * exact half in that case.
1918 */
1919 sub_size = victim->remaining / 2;
1920 list -= sub_size;
1921 }
1922 target->list = list;
1923 victim->list_size -= sub_size;
1924 victim->remaining -= sub_size;
1925 }
1926 target->list_size = sub_size;
1927 target->remaining = sub_size;
1928 target->working = 1;
1929 progress_unlock();
1930
1931 pthread_mutex_lock(&target->mutex);
1932 target->data_ready = 1;
1933 pthread_cond_signal(&target->cond);
1934 pthread_mutex_unlock(&target->mutex);
1935
1936 if (!sub_size) {
1937 pthread_join(target->thread, NULL);
1938 pthread_cond_destroy(&target->cond);
1939 pthread_mutex_destroy(&target->mutex);
1940 active_threads--;
1941 }
1942 }
1943 cleanup_threaded_search();
1944 free(p);
1945 }
1946
1947 #else
1948 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1949 #endif
1950
1951 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1952 {
1953 unsigned char peeled[20];
1954
1955 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1956 !peel_ref(path, peeled) && /* peelable? */
1957 !is_null_sha1(peeled) && /* annotated tag? */
1958 locate_object_entry(peeled)) /* object packed? */
1959 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1960 return 0;
1961 }
1962
1963 static void prepare_pack(int window, int depth)
1964 {
1965 struct object_entry **delta_list;
1966 uint32_t i, nr_deltas;
1967 unsigned n;
1968
1969 get_object_details();
1970
1971 /*
1972 * If we're locally repacking then we need to be doubly careful
1973 * from now on in order to make sure no stealth corruption gets
1974 * propagated to the new pack. Clients receiving streamed packs
1975 * should validate everything they get anyway so no need to incur
1976 * the additional cost here in that case.
1977 */
1978 if (!pack_to_stdout)
1979 do_check_packed_object_crc = 1;
1980
1981 if (!nr_objects || !window || !depth)
1982 return;
1983
1984 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1985 nr_deltas = n = 0;
1986
1987 for (i = 0; i < nr_objects; i++) {
1988 struct object_entry *entry = objects + i;
1989
1990 if (entry->delta)
1991 /* This happens if we decided to reuse existing
1992 * delta from a pack. "reuse_delta &&" is implied.
1993 */
1994 continue;
1995
1996 if (entry->size < 50)
1997 continue;
1998
1999 if (entry->no_try_delta)
2000 continue;
2001
2002 if (!entry->preferred_base) {
2003 nr_deltas++;
2004 if (entry->type < 0)
2005 die("unable to get type of object %s",
2006 sha1_to_hex(entry->idx.sha1));
2007 } else {
2008 if (entry->type < 0) {
2009 /*
2010 * This object is not found, but we
2011 * don't have to include it anyway.
2012 */
2013 continue;
2014 }
2015 }
2016
2017 delta_list[n++] = entry;
2018 }
2019
2020 if (nr_deltas && n > 1) {
2021 unsigned nr_done = 0;
2022 if (progress)
2023 progress_state = start_progress("Compressing objects",
2024 nr_deltas);
2025 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2026 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2027 stop_progress(&progress_state);
2028 if (nr_done != nr_deltas)
2029 die("inconsistency with delta count");
2030 }
2031 free(delta_list);
2032 }
2033
2034 static int git_pack_config(const char *k, const char *v, void *cb)
2035 {
2036 if (!strcmp(k, "pack.window")) {
2037 window = git_config_int(k, v);
2038 return 0;
2039 }
2040 if (!strcmp(k, "pack.windowmemory")) {
2041 window_memory_limit = git_config_ulong(k, v);
2042 return 0;
2043 }
2044 if (!strcmp(k, "pack.depth")) {
2045 depth = git_config_int(k, v);
2046 return 0;
2047 }
2048 if (!strcmp(k, "pack.compression")) {
2049 int level = git_config_int(k, v);
2050 if (level == -1)
2051 level = Z_DEFAULT_COMPRESSION;
2052 else if (level < 0 || level > Z_BEST_COMPRESSION)
2053 die("bad pack compression level %d", level);
2054 pack_compression_level = level;
2055 pack_compression_seen = 1;
2056 return 0;
2057 }
2058 if (!strcmp(k, "pack.deltacachesize")) {
2059 max_delta_cache_size = git_config_int(k, v);
2060 return 0;
2061 }
2062 if (!strcmp(k, "pack.deltacachelimit")) {
2063 cache_max_small_delta_size = git_config_int(k, v);
2064 return 0;
2065 }
2066 if (!strcmp(k, "pack.threads")) {
2067 delta_search_threads = git_config_int(k, v);
2068 if (delta_search_threads < 0)
2069 die("invalid number of threads specified (%d)",
2070 delta_search_threads);
2071 #ifdef NO_PTHREADS
2072 if (delta_search_threads != 1)
2073 warning("no threads support, ignoring %s", k);
2074 #endif
2075 return 0;
2076 }
2077 if (!strcmp(k, "pack.indexversion")) {
2078 pack_idx_opts.version = git_config_int(k, v);
2079 if (pack_idx_opts.version > 2)
2080 die("bad pack.indexversion=%"PRIu32,
2081 pack_idx_opts.version);
2082 return 0;
2083 }
2084 return git_default_config(k, v, cb);
2085 }
2086
2087 static void read_object_list_from_stdin(void)
2088 {
2089 char line[40 + 1 + PATH_MAX + 2];
2090 unsigned char sha1[20];
2091
2092 for (;;) {
2093 if (!fgets(line, sizeof(line), stdin)) {
2094 if (feof(stdin))
2095 break;
2096 if (!ferror(stdin))
2097 die("fgets returned NULL, not EOF, not error!");
2098 if (errno != EINTR)
2099 die_errno("fgets");
2100 clearerr(stdin);
2101 continue;
2102 }
2103 if (line[0] == '-') {
2104 if (get_sha1_hex(line+1, sha1))
2105 die("expected edge sha1, got garbage:\n %s",
2106 line);
2107 add_preferred_base(sha1);
2108 continue;
2109 }
2110 if (get_sha1_hex(line, sha1))
2111 die("expected sha1, got garbage:\n %s", line);
2112
2113 add_preferred_base_object(line+41);
2114 add_object_entry(sha1, 0, line+41, 0);
2115 }
2116 }
2117
2118 #define OBJECT_ADDED (1u<<20)
2119
2120 static void show_commit(struct commit *commit, void *data)
2121 {
2122 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2123 commit->object.flags |= OBJECT_ADDED;
2124 }
2125
2126 static void show_object(struct object *obj,
2127 const struct name_path *path, const char *last,
2128 void *data)
2129 {
2130 char *name = path_name(path, last);
2131
2132 add_preferred_base_object(name);
2133 add_object_entry(obj->sha1, obj->type, name, 0);
2134 obj->flags |= OBJECT_ADDED;
2135
2136 /*
2137 * We will have generated the hash from the name,
2138 * but not saved a pointer to it - we can free it
2139 */
2140 free((char *)name);
2141 }
2142
2143 static void show_edge(struct commit *commit)
2144 {
2145 add_preferred_base(commit->object.sha1);
2146 }
2147
2148 struct in_pack_object {
2149 off_t offset;
2150 struct object *object;
2151 };
2152
2153 struct in_pack {
2154 int alloc;
2155 int nr;
2156 struct in_pack_object *array;
2157 };
2158
2159 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2160 {
2161 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2162 in_pack->array[in_pack->nr].object = object;
2163 in_pack->nr++;
2164 }
2165
2166 /*
2167 * Compare the objects in the offset order, in order to emulate the
2168 * "git rev-list --objects" output that produced the pack originally.
2169 */
2170 static int ofscmp(const void *a_, const void *b_)
2171 {
2172 struct in_pack_object *a = (struct in_pack_object *)a_;
2173 struct in_pack_object *b = (struct in_pack_object *)b_;
2174
2175 if (a->offset < b->offset)
2176 return -1;
2177 else if (a->offset > b->offset)
2178 return 1;
2179 else
2180 return hashcmp(a->object->sha1, b->object->sha1);
2181 }
2182
2183 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2184 {
2185 struct packed_git *p;
2186 struct in_pack in_pack;
2187 uint32_t i;
2188
2189 memset(&in_pack, 0, sizeof(in_pack));
2190
2191 for (p = packed_git; p; p = p->next) {
2192 const unsigned char *sha1;
2193 struct object *o;
2194
2195 if (!p->pack_local || p->pack_keep)
2196 continue;
2197 if (open_pack_index(p))
2198 die("cannot open pack index");
2199
2200 ALLOC_GROW(in_pack.array,
2201 in_pack.nr + p->num_objects,
2202 in_pack.alloc);
2203
2204 for (i = 0; i < p->num_objects; i++) {
2205 sha1 = nth_packed_object_sha1(p, i);
2206 o = lookup_unknown_object(sha1);
2207 if (!(o->flags & OBJECT_ADDED))
2208 mark_in_pack_object(o, p, &in_pack);
2209 o->flags |= OBJECT_ADDED;
2210 }
2211 }
2212
2213 if (in_pack.nr) {
2214 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2215 ofscmp);
2216 for (i = 0; i < in_pack.nr; i++) {
2217 struct object *o = in_pack.array[i].object;
2218 add_object_entry(o->sha1, o->type, "", 0);
2219 }
2220 }
2221 free(in_pack.array);
2222 }
2223
2224 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2225 {
2226 static struct packed_git *last_found = (void *)1;
2227 struct packed_git *p;
2228
2229 p = (last_found != (void *)1) ? last_found : packed_git;
2230
2231 while (p) {
2232 if ((!p->pack_local || p->pack_keep) &&
2233 find_pack_entry_one(sha1, p)) {
2234 last_found = p;
2235 return 1;
2236 }
2237 if (p == last_found)
2238 p = packed_git;
2239 else
2240 p = p->next;
2241 if (p == last_found)
2242 p = p->next;
2243 }
2244 return 0;
2245 }
2246
2247 static void loosen_unused_packed_objects(struct rev_info *revs)
2248 {
2249 struct packed_git *p;
2250 uint32_t i;
2251 const unsigned char *sha1;
2252
2253 for (p = packed_git; p; p = p->next) {
2254 if (!p->pack_local || p->pack_keep)
2255 continue;
2256
2257 if (open_pack_index(p))
2258 die("cannot open pack index");
2259
2260 for (i = 0; i < p->num_objects; i++) {
2261 sha1 = nth_packed_object_sha1(p, i);
2262 if (!locate_object_entry(sha1) &&
2263 !has_sha1_pack_kept_or_nonlocal(sha1))
2264 if (force_object_loose(sha1, p->mtime))
2265 die("unable to force loose object");
2266 }
2267 }
2268 }
2269
2270 static void get_object_list(int ac, const char **av)
2271 {
2272 struct rev_info revs;
2273 char line[1000];
2274 int flags = 0;
2275
2276 init_revisions(&revs, NULL);
2277 save_commit_buffer = 0;
2278 setup_revisions(ac, av, &revs, NULL);
2279
2280 while (fgets(line, sizeof(line), stdin) != NULL) {
2281 int len = strlen(line);
2282 if (len && line[len - 1] == '\n')
2283 line[--len] = 0;
2284 if (!len)
2285 break;
2286 if (*line == '-') {
2287 if (!strcmp(line, "--not")) {
2288 flags ^= UNINTERESTING;
2289 continue;
2290 }
2291 die("not a rev '%s'", line);
2292 }
2293 if (handle_revision_arg(line, &revs, flags, 1))
2294 die("bad revision '%s'", line);
2295 }
2296
2297 if (prepare_revision_walk(&revs))
2298 die("revision walk setup failed");
2299 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2300 traverse_commit_list(&revs, show_commit, show_object, NULL);
2301
2302 if (keep_unreachable)
2303 add_objects_in_unpacked_packs(&revs);
2304 if (unpack_unreachable)
2305 loosen_unused_packed_objects(&revs);
2306 }
2307
2308 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2309 {
2310 int use_internal_rev_list = 0;
2311 int thin = 0;
2312 int all_progress_implied = 0;
2313 uint32_t i;
2314 const char **rp_av;
2315 int rp_ac_alloc = 64;
2316 int rp_ac;
2317
2318 read_replace_refs = 0;
2319
2320 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2321
2322 rp_av[0] = "pack-objects";
2323 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2324 rp_ac = 2;
2325
2326 reset_pack_idx_option(&pack_idx_opts);
2327 git_config(git_pack_config, NULL);
2328 if (!pack_compression_seen && core_compression_seen)
2329 pack_compression_level = core_compression_level;
2330
2331 progress = isatty(2);
2332 for (i = 1; i < argc; i++) {
2333 const char *arg = argv[i];
2334
2335 if (*arg != '-')
2336 break;
2337
2338 if (!strcmp("--non-empty", arg)) {
2339 non_empty = 1;
2340 continue;
2341 }
2342 if (!strcmp("--local", arg)) {
2343 local = 1;
2344 continue;
2345 }
2346 if (!strcmp("--incremental", arg)) {
2347 incremental = 1;
2348 continue;
2349 }
2350 if (!strcmp("--honor-pack-keep", arg)) {
2351 ignore_packed_keep = 1;
2352 continue;
2353 }
2354 if (!prefixcmp(arg, "--compression=")) {
2355 char *end;
2356 int level = strtoul(arg+14, &end, 0);
2357 if (!arg[14] || *end)
2358 usage(pack_usage);
2359 if (level == -1)
2360 level = Z_DEFAULT_COMPRESSION;
2361 else if (level < 0 || level > Z_BEST_COMPRESSION)
2362 die("bad pack compression level %d", level);
2363 pack_compression_level = level;
2364 continue;
2365 }
2366 if (!prefixcmp(arg, "--max-pack-size=")) {
2367 pack_size_limit_cfg = 0;
2368 if (!git_parse_ulong(arg+16, &pack_size_limit))
2369 usage(pack_usage);
2370 continue;
2371 }
2372 if (!prefixcmp(arg, "--window=")) {
2373 char *end;
2374 window = strtoul(arg+9, &end, 0);
2375 if (!arg[9] || *end)
2376 usage(pack_usage);
2377 continue;
2378 }
2379 if (!prefixcmp(arg, "--window-memory=")) {
2380 if (!git_parse_ulong(arg+16, &window_memory_limit))
2381 usage(pack_usage);
2382 continue;
2383 }
2384 if (!prefixcmp(arg, "--threads=")) {
2385 char *end;
2386 delta_search_threads = strtoul(arg+10, &end, 0);
2387 if (!arg[10] || *end || delta_search_threads < 0)
2388 usage(pack_usage);
2389 #ifdef NO_PTHREADS
2390 if (delta_search_threads != 1)
2391 warning("no threads support, "
2392 "ignoring %s", arg);
2393 #endif
2394 continue;
2395 }
2396 if (!prefixcmp(arg, "--depth=")) {
2397 char *end;
2398 depth = strtoul(arg+8, &end, 0);
2399 if (!arg[8] || *end)
2400 usage(pack_usage);
2401 continue;
2402 }
2403 if (!strcmp("--progress", arg)) {
2404 progress = 1;
2405 continue;
2406 }
2407 if (!strcmp("--all-progress", arg)) {
2408 progress = 2;
2409 continue;
2410 }
2411 if (!strcmp("--all-progress-implied", arg)) {
2412 all_progress_implied = 1;
2413 continue;
2414 }
2415 if (!strcmp("-q", arg)) {
2416 progress = 0;
2417 continue;
2418 }
2419 if (!strcmp("--no-reuse-delta", arg)) {
2420 reuse_delta = 0;
2421 continue;
2422 }
2423 if (!strcmp("--no-reuse-object", arg)) {
2424 reuse_object = reuse_delta = 0;
2425 continue;
2426 }
2427 if (!strcmp("--delta-base-offset", arg)) {
2428 allow_ofs_delta = 1;
2429 continue;
2430 }
2431 if (!strcmp("--stdout", arg)) {
2432 pack_to_stdout = 1;
2433 continue;
2434 }
2435 if (!strcmp("--revs", arg)) {
2436 use_internal_rev_list = 1;
2437 continue;
2438 }
2439 if (!strcmp("--keep-unreachable", arg)) {
2440 keep_unreachable = 1;
2441 continue;
2442 }
2443 if (!strcmp("--unpack-unreachable", arg)) {
2444 unpack_unreachable = 1;
2445 continue;
2446 }
2447 if (!strcmp("--include-tag", arg)) {
2448 include_tag = 1;
2449 continue;
2450 }
2451 if (!strcmp("--unpacked", arg) ||
2452 !strcmp("--reflog", arg) ||
2453 !strcmp("--all", arg)) {
2454 use_internal_rev_list = 1;
2455 if (rp_ac >= rp_ac_alloc - 1) {
2456 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2457 rp_av = xrealloc(rp_av,
2458 rp_ac_alloc * sizeof(*rp_av));
2459 }
2460 rp_av[rp_ac++] = arg;
2461 continue;
2462 }
2463 if (!strcmp("--thin", arg)) {
2464 use_internal_rev_list = 1;
2465 thin = 1;
2466 rp_av[1] = "--objects-edge";
2467 continue;
2468 }
2469 if (!prefixcmp(arg, "--index-version=")) {
2470 char *c;
2471 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2472 if (pack_idx_opts.version > 2)
2473 die("bad %s", arg);
2474 if (*c == ',')
2475 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2476 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2477 die("bad %s", arg);
2478 continue;
2479 }
2480 if (!strcmp(arg, "--keep-true-parents")) {
2481 grafts_replace_parents = 0;
2482 continue;
2483 }
2484 usage(pack_usage);
2485 }
2486
2487 /* Traditionally "pack-objects [options] base extra" failed;
2488 * we would however want to take refs parameter that would
2489 * have been given to upstream rev-list ourselves, which means
2490 * we somehow want to say what the base name is. So the
2491 * syntax would be:
2492 *
2493 * pack-objects [options] base <refs...>
2494 *
2495 * in other words, we would treat the first non-option as the
2496 * base_name and send everything else to the internal revision
2497 * walker.
2498 */
2499
2500 if (!pack_to_stdout)
2501 base_name = argv[i++];
2502
2503 if (pack_to_stdout != !base_name)
2504 usage(pack_usage);
2505
2506 if (!pack_to_stdout && !pack_size_limit)
2507 pack_size_limit = pack_size_limit_cfg;
2508 if (pack_to_stdout && pack_size_limit)
2509 die("--max-pack-size cannot be used to build a pack for transfer.");
2510 if (pack_size_limit && pack_size_limit < 1024*1024) {
2511 warning("minimum pack size limit is 1 MiB");
2512 pack_size_limit = 1024*1024;
2513 }
2514
2515 if (!pack_to_stdout && thin)
2516 die("--thin cannot be used to build an indexable pack.");
2517
2518 if (keep_unreachable && unpack_unreachable)
2519 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2520
2521 if (progress && all_progress_implied)
2522 progress = 2;
2523
2524 prepare_packed_git();
2525
2526 if (progress)
2527 progress_state = start_progress("Counting objects", 0);
2528 if (!use_internal_rev_list)
2529 read_object_list_from_stdin();
2530 else {
2531 rp_av[rp_ac] = NULL;
2532 get_object_list(rp_ac, rp_av);
2533 }
2534 cleanup_preferred_base();
2535 if (include_tag && nr_result)
2536 for_each_ref(add_ref_tag, NULL);
2537 stop_progress(&progress_state);
2538
2539 if (non_empty && !nr_result)
2540 return 0;
2541 if (nr_result)
2542 prepare_pack(window, depth);
2543 write_pack_file();
2544 if (progress)
2545 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2546 " reused %"PRIu32" (delta %"PRIu32")\n",
2547 written, written_delta, reused, reused_delta);
2548 return 0;
2549 }