Merge branch 'maint'
[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, 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 static int write_one(struct sha1file *f,
413 struct object_entry *e,
414 off_t *offset)
415 {
416 unsigned long size;
417
418 /* offset is non zero if object is written already. */
419 if (e->idx.offset || e->preferred_base)
420 return -1;
421
422 /* if we are deltified, write out base object first. */
423 if (e->delta && !write_one(f, e->delta, offset))
424 return 0;
425
426 e->idx.offset = *offset;
427 size = write_object(f, e, *offset);
428 if (!size) {
429 e->idx.offset = 0;
430 return 0;
431 }
432 written_list[nr_written++] = &e->idx;
433
434 /* make sure off_t is sufficiently large not to wrap */
435 if (signed_add_overflows(*offset, size))
436 die("pack too large for current definition of off_t");
437 *offset += size;
438 return 1;
439 }
440
441 static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
442 void *cb_data)
443 {
444 unsigned char peeled[20];
445 struct object_entry *entry = locate_object_entry(sha1);
446
447 if (entry)
448 entry->tagged = 1;
449 if (!peel_ref(path, peeled)) {
450 entry = locate_object_entry(peeled);
451 if (entry)
452 entry->tagged = 1;
453 }
454 return 0;
455 }
456
457 static void add_to_write_order(struct object_entry **wo,
458 int *endp,
459 struct object_entry *e)
460 {
461 if (e->filled)
462 return;
463 wo[(*endp)++] = e;
464 e->filled = 1;
465 }
466
467 static void add_descendants_to_write_order(struct object_entry **wo,
468 int *endp,
469 struct object_entry *e)
470 {
471 struct object_entry *child;
472
473 for (child = e->delta_child; child; child = child->delta_sibling)
474 add_to_write_order(wo, endp, child);
475 for (child = e->delta_child; child; child = child->delta_sibling)
476 add_descendants_to_write_order(wo, endp, child);
477 }
478
479 static void add_family_to_write_order(struct object_entry **wo,
480 int *endp,
481 struct object_entry *e)
482 {
483 struct object_entry *root;
484
485 for (root = e; root->delta; root = root->delta)
486 ; /* nothing */
487 add_to_write_order(wo, endp, root);
488 add_descendants_to_write_order(wo, endp, root);
489 }
490
491 static struct object_entry **compute_write_order(void)
492 {
493 int i, wo_end;
494
495 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));
496
497 for (i = 0; i < nr_objects; i++) {
498 objects[i].tagged = 0;
499 objects[i].filled = 0;
500 objects[i].delta_child = NULL;
501 objects[i].delta_sibling = NULL;
502 }
503
504 /*
505 * Fully connect delta_child/delta_sibling network.
506 * Make sure delta_sibling is sorted in the original
507 * recency order.
508 */
509 for (i = nr_objects - 1; 0 <= i; i--) {
510 struct object_entry *e = &objects[i];
511 if (!e->delta)
512 continue;
513 /* Mark me as the first child */
514 e->delta_sibling = e->delta->delta_child;
515 e->delta->delta_child = e;
516 }
517
518 /*
519 * Mark objects that are at the tip of tags.
520 */
521 for_each_tag_ref(mark_tagged, NULL);
522
523 /*
524 * Give the commits in the original recency order until
525 * we see a tagged tip.
526 */
527 for (i = wo_end = 0; i < nr_objects; i++) {
528 if (objects[i].tagged)
529 break;
530 add_to_write_order(wo, &wo_end, &objects[i]);
531 }
532
533 /*
534 * Then fill all the tagged tips.
535 */
536 for (; i < nr_objects; i++) {
537 if (objects[i].tagged)
538 add_to_write_order(wo, &wo_end, &objects[i]);
539 }
540
541 /*
542 * And then all remaining commits and tags.
543 */
544 for (i = 0; i < nr_objects; i++) {
545 if (objects[i].type != OBJ_COMMIT &&
546 objects[i].type != OBJ_TAG)
547 continue;
548 add_to_write_order(wo, &wo_end, &objects[i]);
549 }
550
551 /*
552 * And then all the trees.
553 */
554 for (i = 0; i < nr_objects; i++) {
555 if (objects[i].type != OBJ_TREE)
556 continue;
557 add_to_write_order(wo, &wo_end, &objects[i]);
558 }
559
560 /*
561 * Finally all the rest in really tight order
562 */
563 for (i = 0; i < nr_objects; i++)
564 add_family_to_write_order(wo, &wo_end, &objects[i]);
565
566 return wo;
567 }
568
569 static void write_pack_file(void)
570 {
571 uint32_t i = 0, j;
572 struct sha1file *f;
573 off_t offset;
574 struct pack_header hdr;
575 uint32_t nr_remaining = nr_result;
576 time_t last_mtime = 0;
577 struct object_entry **write_order;
578
579 if (progress > pack_to_stdout)
580 progress_state = start_progress("Writing objects", nr_result);
581 written_list = xmalloc(nr_objects * sizeof(*written_list));
582 write_order = compute_write_order();
583
584 do {
585 unsigned char sha1[20];
586 char *pack_tmp_name = NULL;
587
588 if (pack_to_stdout) {
589 f = sha1fd_throughput(1, "<stdout>", progress_state);
590 } else {
591 char tmpname[PATH_MAX];
592 int fd;
593 fd = odb_mkstemp(tmpname, sizeof(tmpname),
594 "pack/tmp_pack_XXXXXX");
595 pack_tmp_name = xstrdup(tmpname);
596 f = sha1fd(fd, pack_tmp_name);
597 }
598
599 hdr.hdr_signature = htonl(PACK_SIGNATURE);
600 hdr.hdr_version = htonl(PACK_VERSION);
601 hdr.hdr_entries = htonl(nr_remaining);
602 sha1write(f, &hdr, sizeof(hdr));
603 offset = sizeof(hdr);
604 nr_written = 0;
605 for (; i < nr_objects; i++) {
606 struct object_entry *e = write_order[i];
607 if (!write_one(f, e, &offset))
608 break;
609 display_progress(progress_state, written);
610 }
611
612 /*
613 * Did we write the wrong # entries in the header?
614 * If so, rewrite it like in fast-import
615 */
616 if (pack_to_stdout) {
617 sha1close(f, sha1, CSUM_CLOSE);
618 } else if (nr_written == nr_remaining) {
619 sha1close(f, sha1, CSUM_FSYNC);
620 } else {
621 int fd = sha1close(f, sha1, 0);
622 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
623 nr_written, sha1, offset);
624 close(fd);
625 }
626
627 if (!pack_to_stdout) {
628 struct stat st;
629 const char *idx_tmp_name;
630 char tmpname[PATH_MAX];
631
632 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
633 &pack_idx_opts, sha1);
634
635 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
636 base_name, sha1_to_hex(sha1));
637 free_pack_by_name(tmpname);
638 if (adjust_shared_perm(pack_tmp_name))
639 die_errno("unable to make temporary pack file readable");
640 if (rename(pack_tmp_name, tmpname))
641 die_errno("unable to rename temporary pack file");
642
643 /*
644 * Packs are runtime accessed in their mtime
645 * order since newer packs are more likely to contain
646 * younger objects. So if we are creating multiple
647 * packs then we should modify the mtime of later ones
648 * to preserve this property.
649 */
650 if (stat(tmpname, &st) < 0) {
651 warning("failed to stat %s: %s",
652 tmpname, strerror(errno));
653 } else if (!last_mtime) {
654 last_mtime = st.st_mtime;
655 } else {
656 struct utimbuf utb;
657 utb.actime = st.st_atime;
658 utb.modtime = --last_mtime;
659 if (utime(tmpname, &utb) < 0)
660 warning("failed utime() on %s: %s",
661 tmpname, strerror(errno));
662 }
663
664 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
665 base_name, sha1_to_hex(sha1));
666 if (adjust_shared_perm(idx_tmp_name))
667 die_errno("unable to make temporary index file readable");
668 if (rename(idx_tmp_name, tmpname))
669 die_errno("unable to rename temporary index file");
670
671 free((void *) idx_tmp_name);
672 free(pack_tmp_name);
673 puts(sha1_to_hex(sha1));
674 }
675
676 /* mark written objects as written to previous pack */
677 for (j = 0; j < nr_written; j++) {
678 written_list[j]->offset = (off_t)-1;
679 }
680 nr_remaining -= nr_written;
681 } while (nr_remaining && i < nr_objects);
682
683 free(written_list);
684 free(write_order);
685 stop_progress(&progress_state);
686 if (written != nr_result)
687 die("wrote %"PRIu32" objects while expecting %"PRIu32,
688 written, nr_result);
689 }
690
691 static int locate_object_entry_hash(const unsigned char *sha1)
692 {
693 int i;
694 unsigned int ui;
695 memcpy(&ui, sha1, sizeof(unsigned int));
696 i = ui % object_ix_hashsz;
697 while (0 < object_ix[i]) {
698 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
699 return i;
700 if (++i == object_ix_hashsz)
701 i = 0;
702 }
703 return -1 - i;
704 }
705
706 static struct object_entry *locate_object_entry(const unsigned char *sha1)
707 {
708 int i;
709
710 if (!object_ix_hashsz)
711 return NULL;
712
713 i = locate_object_entry_hash(sha1);
714 if (0 <= i)
715 return &objects[object_ix[i]-1];
716 return NULL;
717 }
718
719 static void rehash_objects(void)
720 {
721 uint32_t i;
722 struct object_entry *oe;
723
724 object_ix_hashsz = nr_objects * 3;
725 if (object_ix_hashsz < 1024)
726 object_ix_hashsz = 1024;
727 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
728 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
729 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
730 int ix = locate_object_entry_hash(oe->idx.sha1);
731 if (0 <= ix)
732 continue;
733 ix = -1 - ix;
734 object_ix[ix] = i + 1;
735 }
736 }
737
738 static unsigned name_hash(const char *name)
739 {
740 unsigned c, hash = 0;
741
742 if (!name)
743 return 0;
744
745 /*
746 * This effectively just creates a sortable number from the
747 * last sixteen non-whitespace characters. Last characters
748 * count "most", so things that end in ".c" sort together.
749 */
750 while ((c = *name++) != 0) {
751 if (isspace(c))
752 continue;
753 hash = (hash >> 2) + (c << 24);
754 }
755 return hash;
756 }
757
758 static void setup_delta_attr_check(struct git_attr_check *check)
759 {
760 static struct git_attr *attr_delta;
761
762 if (!attr_delta)
763 attr_delta = git_attr("delta");
764
765 check[0].attr = attr_delta;
766 }
767
768 static int no_try_delta(const char *path)
769 {
770 struct git_attr_check check[1];
771
772 setup_delta_attr_check(check);
773 if (git_check_attr(path, ARRAY_SIZE(check), check))
774 return 0;
775 if (ATTR_FALSE(check->value))
776 return 1;
777 return 0;
778 }
779
780 static int add_object_entry(const unsigned char *sha1, enum object_type type,
781 const char *name, int exclude)
782 {
783 struct object_entry *entry;
784 struct packed_git *p, *found_pack = NULL;
785 off_t found_offset = 0;
786 int ix;
787 unsigned hash = name_hash(name);
788
789 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
790 if (ix >= 0) {
791 if (exclude) {
792 entry = objects + object_ix[ix] - 1;
793 if (!entry->preferred_base)
794 nr_result--;
795 entry->preferred_base = 1;
796 }
797 return 0;
798 }
799
800 if (!exclude && local && has_loose_object_nonlocal(sha1))
801 return 0;
802
803 for (p = packed_git; p; p = p->next) {
804 off_t offset = find_pack_entry_one(sha1, p);
805 if (offset) {
806 if (!found_pack) {
807 if (!is_pack_valid(p)) {
808 warning("packfile %s cannot be accessed", p->pack_name);
809 continue;
810 }
811 found_offset = offset;
812 found_pack = p;
813 }
814 if (exclude)
815 break;
816 if (incremental)
817 return 0;
818 if (local && !p->pack_local)
819 return 0;
820 if (ignore_packed_keep && p->pack_local && p->pack_keep)
821 return 0;
822 }
823 }
824
825 if (nr_objects >= nr_alloc) {
826 nr_alloc = (nr_alloc + 1024) * 3 / 2;
827 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
828 }
829
830 entry = objects + nr_objects++;
831 memset(entry, 0, sizeof(*entry));
832 hashcpy(entry->idx.sha1, sha1);
833 entry->hash = hash;
834 if (type)
835 entry->type = type;
836 if (exclude)
837 entry->preferred_base = 1;
838 else
839 nr_result++;
840 if (found_pack) {
841 entry->in_pack = found_pack;
842 entry->in_pack_offset = found_offset;
843 }
844
845 if (object_ix_hashsz * 3 <= nr_objects * 4)
846 rehash_objects();
847 else
848 object_ix[-1 - ix] = nr_objects;
849
850 display_progress(progress_state, nr_objects);
851
852 if (name && no_try_delta(name))
853 entry->no_try_delta = 1;
854
855 return 1;
856 }
857
858 struct pbase_tree_cache {
859 unsigned char sha1[20];
860 int ref;
861 int temporary;
862 void *tree_data;
863 unsigned long tree_size;
864 };
865
866 static struct pbase_tree_cache *(pbase_tree_cache[256]);
867 static int pbase_tree_cache_ix(const unsigned char *sha1)
868 {
869 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
870 }
871 static int pbase_tree_cache_ix_incr(int ix)
872 {
873 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
874 }
875
876 static struct pbase_tree {
877 struct pbase_tree *next;
878 /* This is a phony "cache" entry; we are not
879 * going to evict it nor find it through _get()
880 * mechanism -- this is for the toplevel node that
881 * would almost always change with any commit.
882 */
883 struct pbase_tree_cache pcache;
884 } *pbase_tree;
885
886 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
887 {
888 struct pbase_tree_cache *ent, *nent;
889 void *data;
890 unsigned long size;
891 enum object_type type;
892 int neigh;
893 int my_ix = pbase_tree_cache_ix(sha1);
894 int available_ix = -1;
895
896 /* pbase-tree-cache acts as a limited hashtable.
897 * your object will be found at your index or within a few
898 * slots after that slot if it is cached.
899 */
900 for (neigh = 0; neigh < 8; neigh++) {
901 ent = pbase_tree_cache[my_ix];
902 if (ent && !hashcmp(ent->sha1, sha1)) {
903 ent->ref++;
904 return ent;
905 }
906 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
907 ((0 <= available_ix) &&
908 (!ent && pbase_tree_cache[available_ix])))
909 available_ix = my_ix;
910 if (!ent)
911 break;
912 my_ix = pbase_tree_cache_ix_incr(my_ix);
913 }
914
915 /* Did not find one. Either we got a bogus request or
916 * we need to read and perhaps cache.
917 */
918 data = read_sha1_file(sha1, &type, &size);
919 if (!data)
920 return NULL;
921 if (type != OBJ_TREE) {
922 free(data);
923 return NULL;
924 }
925
926 /* We need to either cache or return a throwaway copy */
927
928 if (available_ix < 0)
929 ent = NULL;
930 else {
931 ent = pbase_tree_cache[available_ix];
932 my_ix = available_ix;
933 }
934
935 if (!ent) {
936 nent = xmalloc(sizeof(*nent));
937 nent->temporary = (available_ix < 0);
938 }
939 else {
940 /* evict and reuse */
941 free(ent->tree_data);
942 nent = ent;
943 }
944 hashcpy(nent->sha1, sha1);
945 nent->tree_data = data;
946 nent->tree_size = size;
947 nent->ref = 1;
948 if (!nent->temporary)
949 pbase_tree_cache[my_ix] = nent;
950 return nent;
951 }
952
953 static void pbase_tree_put(struct pbase_tree_cache *cache)
954 {
955 if (!cache->temporary) {
956 cache->ref--;
957 return;
958 }
959 free(cache->tree_data);
960 free(cache);
961 }
962
963 static int name_cmp_len(const char *name)
964 {
965 int i;
966 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
967 ;
968 return i;
969 }
970
971 static void add_pbase_object(struct tree_desc *tree,
972 const char *name,
973 int cmplen,
974 const char *fullname)
975 {
976 struct name_entry entry;
977 int cmp;
978
979 while (tree_entry(tree,&entry)) {
980 if (S_ISGITLINK(entry.mode))
981 continue;
982 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
983 memcmp(name, entry.path, cmplen);
984 if (cmp > 0)
985 continue;
986 if (cmp < 0)
987 return;
988 if (name[cmplen] != '/') {
989 add_object_entry(entry.sha1,
990 object_type(entry.mode),
991 fullname, 1);
992 return;
993 }
994 if (S_ISDIR(entry.mode)) {
995 struct tree_desc sub;
996 struct pbase_tree_cache *tree;
997 const char *down = name+cmplen+1;
998 int downlen = name_cmp_len(down);
999
1000 tree = pbase_tree_get(entry.sha1);
1001 if (!tree)
1002 return;
1003 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1004
1005 add_pbase_object(&sub, down, downlen, fullname);
1006 pbase_tree_put(tree);
1007 }
1008 }
1009 }
1010
1011 static unsigned *done_pbase_paths;
1012 static int done_pbase_paths_num;
1013 static int done_pbase_paths_alloc;
1014 static int done_pbase_path_pos(unsigned hash)
1015 {
1016 int lo = 0;
1017 int hi = done_pbase_paths_num;
1018 while (lo < hi) {
1019 int mi = (hi + lo) / 2;
1020 if (done_pbase_paths[mi] == hash)
1021 return mi;
1022 if (done_pbase_paths[mi] < hash)
1023 hi = mi;
1024 else
1025 lo = mi + 1;
1026 }
1027 return -lo-1;
1028 }
1029
1030 static int check_pbase_path(unsigned hash)
1031 {
1032 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1033 if (0 <= pos)
1034 return 1;
1035 pos = -pos - 1;
1036 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1037 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1038 done_pbase_paths = xrealloc(done_pbase_paths,
1039 done_pbase_paths_alloc *
1040 sizeof(unsigned));
1041 }
1042 done_pbase_paths_num++;
1043 if (pos < done_pbase_paths_num)
1044 memmove(done_pbase_paths + pos + 1,
1045 done_pbase_paths + pos,
1046 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1047 done_pbase_paths[pos] = hash;
1048 return 0;
1049 }
1050
1051 static void add_preferred_base_object(const char *name)
1052 {
1053 struct pbase_tree *it;
1054 int cmplen;
1055 unsigned hash = name_hash(name);
1056
1057 if (!num_preferred_base || check_pbase_path(hash))
1058 return;
1059
1060 cmplen = name_cmp_len(name);
1061 for (it = pbase_tree; it; it = it->next) {
1062 if (cmplen == 0) {
1063 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1064 }
1065 else {
1066 struct tree_desc tree;
1067 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1068 add_pbase_object(&tree, name, cmplen, name);
1069 }
1070 }
1071 }
1072
1073 static void add_preferred_base(unsigned char *sha1)
1074 {
1075 struct pbase_tree *it;
1076 void *data;
1077 unsigned long size;
1078 unsigned char tree_sha1[20];
1079
1080 if (window <= num_preferred_base++)
1081 return;
1082
1083 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1084 if (!data)
1085 return;
1086
1087 for (it = pbase_tree; it; it = it->next) {
1088 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1089 free(data);
1090 return;
1091 }
1092 }
1093
1094 it = xcalloc(1, sizeof(*it));
1095 it->next = pbase_tree;
1096 pbase_tree = it;
1097
1098 hashcpy(it->pcache.sha1, tree_sha1);
1099 it->pcache.tree_data = data;
1100 it->pcache.tree_size = size;
1101 }
1102
1103 static void cleanup_preferred_base(void)
1104 {
1105 struct pbase_tree *it;
1106 unsigned i;
1107
1108 it = pbase_tree;
1109 pbase_tree = NULL;
1110 while (it) {
1111 struct pbase_tree *this = it;
1112 it = this->next;
1113 free(this->pcache.tree_data);
1114 free(this);
1115 }
1116
1117 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1118 if (!pbase_tree_cache[i])
1119 continue;
1120 free(pbase_tree_cache[i]->tree_data);
1121 free(pbase_tree_cache[i]);
1122 pbase_tree_cache[i] = NULL;
1123 }
1124
1125 free(done_pbase_paths);
1126 done_pbase_paths = NULL;
1127 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1128 }
1129
1130 static void check_object(struct object_entry *entry)
1131 {
1132 if (entry->in_pack) {
1133 struct packed_git *p = entry->in_pack;
1134 struct pack_window *w_curs = NULL;
1135 const unsigned char *base_ref = NULL;
1136 struct object_entry *base_entry;
1137 unsigned long used, used_0;
1138 unsigned long avail;
1139 off_t ofs;
1140 unsigned char *buf, c;
1141
1142 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1143
1144 /*
1145 * We want in_pack_type even if we do not reuse delta
1146 * since non-delta representations could still be reused.
1147 */
1148 used = unpack_object_header_buffer(buf, avail,
1149 &entry->in_pack_type,
1150 &entry->size);
1151 if (used == 0)
1152 goto give_up;
1153
1154 /*
1155 * Determine if this is a delta and if so whether we can
1156 * reuse it or not. Otherwise let's find out as cheaply as
1157 * possible what the actual type and size for this object is.
1158 */
1159 switch (entry->in_pack_type) {
1160 default:
1161 /* Not a delta hence we've already got all we need. */
1162 entry->type = entry->in_pack_type;
1163 entry->in_pack_header_size = used;
1164 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1165 goto give_up;
1166 unuse_pack(&w_curs);
1167 return;
1168 case OBJ_REF_DELTA:
1169 if (reuse_delta && !entry->preferred_base)
1170 base_ref = use_pack(p, &w_curs,
1171 entry->in_pack_offset + used, NULL);
1172 entry->in_pack_header_size = used + 20;
1173 break;
1174 case OBJ_OFS_DELTA:
1175 buf = use_pack(p, &w_curs,
1176 entry->in_pack_offset + used, NULL);
1177 used_0 = 0;
1178 c = buf[used_0++];
1179 ofs = c & 127;
1180 while (c & 128) {
1181 ofs += 1;
1182 if (!ofs || MSB(ofs, 7)) {
1183 error("delta base offset overflow in pack for %s",
1184 sha1_to_hex(entry->idx.sha1));
1185 goto give_up;
1186 }
1187 c = buf[used_0++];
1188 ofs = (ofs << 7) + (c & 127);
1189 }
1190 ofs = entry->in_pack_offset - ofs;
1191 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1192 error("delta base offset out of bound for %s",
1193 sha1_to_hex(entry->idx.sha1));
1194 goto give_up;
1195 }
1196 if (reuse_delta && !entry->preferred_base) {
1197 struct revindex_entry *revidx;
1198 revidx = find_pack_revindex(p, ofs);
1199 if (!revidx)
1200 goto give_up;
1201 base_ref = nth_packed_object_sha1(p, revidx->nr);
1202 }
1203 entry->in_pack_header_size = used + used_0;
1204 break;
1205 }
1206
1207 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1208 /*
1209 * If base_ref was set above that means we wish to
1210 * reuse delta data, and we even found that base
1211 * in the list of objects we want to pack. Goodie!
1212 *
1213 * Depth value does not matter - find_deltas() will
1214 * never consider reused delta as the base object to
1215 * deltify other objects against, in order to avoid
1216 * circular deltas.
1217 */
1218 entry->type = entry->in_pack_type;
1219 entry->delta = base_entry;
1220 entry->delta_size = entry->size;
1221 entry->delta_sibling = base_entry->delta_child;
1222 base_entry->delta_child = entry;
1223 unuse_pack(&w_curs);
1224 return;
1225 }
1226
1227 if (entry->type) {
1228 /*
1229 * This must be a delta and we already know what the
1230 * final object type is. Let's extract the actual
1231 * object size from the delta header.
1232 */
1233 entry->size = get_size_from_delta(p, &w_curs,
1234 entry->in_pack_offset + entry->in_pack_header_size);
1235 if (entry->size == 0)
1236 goto give_up;
1237 unuse_pack(&w_curs);
1238 return;
1239 }
1240
1241 /*
1242 * No choice but to fall back to the recursive delta walk
1243 * with sha1_object_info() to find about the object type
1244 * at this point...
1245 */
1246 give_up:
1247 unuse_pack(&w_curs);
1248 }
1249
1250 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1251 /*
1252 * The error condition is checked in prepare_pack(). This is
1253 * to permit a missing preferred base object to be ignored
1254 * as a preferred base. Doing so can result in a larger
1255 * pack file, but the transfer will still take place.
1256 */
1257 }
1258
1259 static int pack_offset_sort(const void *_a, const void *_b)
1260 {
1261 const struct object_entry *a = *(struct object_entry **)_a;
1262 const struct object_entry *b = *(struct object_entry **)_b;
1263
1264 /* avoid filesystem trashing with loose objects */
1265 if (!a->in_pack && !b->in_pack)
1266 return hashcmp(a->idx.sha1, b->idx.sha1);
1267
1268 if (a->in_pack < b->in_pack)
1269 return -1;
1270 if (a->in_pack > b->in_pack)
1271 return 1;
1272 return a->in_pack_offset < b->in_pack_offset ? -1 :
1273 (a->in_pack_offset > b->in_pack_offset);
1274 }
1275
1276 static void get_object_details(void)
1277 {
1278 uint32_t i;
1279 struct object_entry **sorted_by_offset;
1280
1281 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1282 for (i = 0; i < nr_objects; i++)
1283 sorted_by_offset[i] = objects + i;
1284 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1285
1286 for (i = 0; i < nr_objects; i++) {
1287 struct object_entry *entry = sorted_by_offset[i];
1288 check_object(entry);
1289 if (big_file_threshold <= entry->size)
1290 entry->no_try_delta = 1;
1291 }
1292
1293 free(sorted_by_offset);
1294 }
1295
1296 /*
1297 * We search for deltas in a list sorted by type, by filename hash, and then
1298 * by size, so that we see progressively smaller and smaller files.
1299 * That's because we prefer deltas to be from the bigger file
1300 * to the smaller -- deletes are potentially cheaper, but perhaps
1301 * more importantly, the bigger file is likely the more recent
1302 * one. The deepest deltas are therefore the oldest objects which are
1303 * less susceptible to be accessed often.
1304 */
1305 static int type_size_sort(const void *_a, const void *_b)
1306 {
1307 const struct object_entry *a = *(struct object_entry **)_a;
1308 const struct object_entry *b = *(struct object_entry **)_b;
1309
1310 if (a->type > b->type)
1311 return -1;
1312 if (a->type < b->type)
1313 return 1;
1314 if (a->hash > b->hash)
1315 return -1;
1316 if (a->hash < b->hash)
1317 return 1;
1318 if (a->preferred_base > b->preferred_base)
1319 return -1;
1320 if (a->preferred_base < b->preferred_base)
1321 return 1;
1322 if (a->size > b->size)
1323 return -1;
1324 if (a->size < b->size)
1325 return 1;
1326 return a < b ? -1 : (a > b); /* newest first */
1327 }
1328
1329 struct unpacked {
1330 struct object_entry *entry;
1331 void *data;
1332 struct delta_index *index;
1333 unsigned depth;
1334 };
1335
1336 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1337 unsigned long delta_size)
1338 {
1339 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1340 return 0;
1341
1342 if (delta_size < cache_max_small_delta_size)
1343 return 1;
1344
1345 /* cache delta, if objects are large enough compared to delta size */
1346 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1347 return 1;
1348
1349 return 0;
1350 }
1351
1352 #ifndef NO_PTHREADS
1353
1354 static pthread_mutex_t read_mutex;
1355 #define read_lock() pthread_mutex_lock(&read_mutex)
1356 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1357
1358 static pthread_mutex_t cache_mutex;
1359 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1360 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1361
1362 static pthread_mutex_t progress_mutex;
1363 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1364 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1365
1366 #else
1367
1368 #define read_lock() (void)0
1369 #define read_unlock() (void)0
1370 #define cache_lock() (void)0
1371 #define cache_unlock() (void)0
1372 #define progress_lock() (void)0
1373 #define progress_unlock() (void)0
1374
1375 #endif
1376
1377 static int try_delta(struct unpacked *trg, struct unpacked *src,
1378 unsigned max_depth, unsigned long *mem_usage)
1379 {
1380 struct object_entry *trg_entry = trg->entry;
1381 struct object_entry *src_entry = src->entry;
1382 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1383 unsigned ref_depth;
1384 enum object_type type;
1385 void *delta_buf;
1386
1387 /* Don't bother doing diffs between different types */
1388 if (trg_entry->type != src_entry->type)
1389 return -1;
1390
1391 /*
1392 * We do not bother to try a delta that we discarded
1393 * on an earlier try, but only when reusing delta data.
1394 */
1395 if (reuse_delta && trg_entry->in_pack &&
1396 trg_entry->in_pack == src_entry->in_pack &&
1397 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1398 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1399 return 0;
1400
1401 /* Let's not bust the allowed depth. */
1402 if (src->depth >= max_depth)
1403 return 0;
1404
1405 /* Now some size filtering heuristics. */
1406 trg_size = trg_entry->size;
1407 if (!trg_entry->delta) {
1408 max_size = trg_size/2 - 20;
1409 ref_depth = 1;
1410 } else {
1411 max_size = trg_entry->delta_size;
1412 ref_depth = trg->depth;
1413 }
1414 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1415 (max_depth - ref_depth + 1);
1416 if (max_size == 0)
1417 return 0;
1418 src_size = src_entry->size;
1419 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1420 if (sizediff >= max_size)
1421 return 0;
1422 if (trg_size < src_size / 32)
1423 return 0;
1424
1425 /* Load data if not already done */
1426 if (!trg->data) {
1427 read_lock();
1428 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1429 read_unlock();
1430 if (!trg->data)
1431 die("object %s cannot be read",
1432 sha1_to_hex(trg_entry->idx.sha1));
1433 if (sz != trg_size)
1434 die("object %s inconsistent object length (%lu vs %lu)",
1435 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1436 *mem_usage += sz;
1437 }
1438 if (!src->data) {
1439 read_lock();
1440 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1441 read_unlock();
1442 if (!src->data) {
1443 if (src_entry->preferred_base) {
1444 static int warned = 0;
1445 if (!warned++)
1446 warning("object %s cannot be read",
1447 sha1_to_hex(src_entry->idx.sha1));
1448 /*
1449 * Those objects are not included in the
1450 * resulting pack. Be resilient and ignore
1451 * them if they can't be read, in case the
1452 * pack could be created nevertheless.
1453 */
1454 return 0;
1455 }
1456 die("object %s cannot be read",
1457 sha1_to_hex(src_entry->idx.sha1));
1458 }
1459 if (sz != src_size)
1460 die("object %s inconsistent object length (%lu vs %lu)",
1461 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1462 *mem_usage += sz;
1463 }
1464 if (!src->index) {
1465 src->index = create_delta_index(src->data, src_size);
1466 if (!src->index) {
1467 static int warned = 0;
1468 if (!warned++)
1469 warning("suboptimal pack - out of memory");
1470 return 0;
1471 }
1472 *mem_usage += sizeof_delta_index(src->index);
1473 }
1474
1475 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1476 if (!delta_buf)
1477 return 0;
1478
1479 if (trg_entry->delta) {
1480 /* Prefer only shallower same-sized deltas. */
1481 if (delta_size == trg_entry->delta_size &&
1482 src->depth + 1 >= trg->depth) {
1483 free(delta_buf);
1484 return 0;
1485 }
1486 }
1487
1488 /*
1489 * Handle memory allocation outside of the cache
1490 * accounting lock. Compiler will optimize the strangeness
1491 * away when NO_PTHREADS is defined.
1492 */
1493 free(trg_entry->delta_data);
1494 cache_lock();
1495 if (trg_entry->delta_data) {
1496 delta_cache_size -= trg_entry->delta_size;
1497 trg_entry->delta_data = NULL;
1498 }
1499 if (delta_cacheable(src_size, trg_size, delta_size)) {
1500 delta_cache_size += delta_size;
1501 cache_unlock();
1502 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1503 } else {
1504 cache_unlock();
1505 free(delta_buf);
1506 }
1507
1508 trg_entry->delta = src_entry;
1509 trg_entry->delta_size = delta_size;
1510 trg->depth = src->depth + 1;
1511
1512 return 1;
1513 }
1514
1515 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1516 {
1517 struct object_entry *child = me->delta_child;
1518 unsigned int m = n;
1519 while (child) {
1520 unsigned int c = check_delta_limit(child, n + 1);
1521 if (m < c)
1522 m = c;
1523 child = child->delta_sibling;
1524 }
1525 return m;
1526 }
1527
1528 static unsigned long free_unpacked(struct unpacked *n)
1529 {
1530 unsigned long freed_mem = sizeof_delta_index(n->index);
1531 free_delta_index(n->index);
1532 n->index = NULL;
1533 if (n->data) {
1534 freed_mem += n->entry->size;
1535 free(n->data);
1536 n->data = NULL;
1537 }
1538 n->entry = NULL;
1539 n->depth = 0;
1540 return freed_mem;
1541 }
1542
1543 static void find_deltas(struct object_entry **list, unsigned *list_size,
1544 int window, int depth, unsigned *processed)
1545 {
1546 uint32_t i, idx = 0, count = 0;
1547 struct unpacked *array;
1548 unsigned long mem_usage = 0;
1549
1550 array = xcalloc(window, sizeof(struct unpacked));
1551
1552 for (;;) {
1553 struct object_entry *entry;
1554 struct unpacked *n = array + idx;
1555 int j, max_depth, best_base = -1;
1556
1557 progress_lock();
1558 if (!*list_size) {
1559 progress_unlock();
1560 break;
1561 }
1562 entry = *list++;
1563 (*list_size)--;
1564 if (!entry->preferred_base) {
1565 (*processed)++;
1566 display_progress(progress_state, *processed);
1567 }
1568 progress_unlock();
1569
1570 mem_usage -= free_unpacked(n);
1571 n->entry = entry;
1572
1573 while (window_memory_limit &&
1574 mem_usage > window_memory_limit &&
1575 count > 1) {
1576 uint32_t tail = (idx + window - count) % window;
1577 mem_usage -= free_unpacked(array + tail);
1578 count--;
1579 }
1580
1581 /* We do not compute delta to *create* objects we are not
1582 * going to pack.
1583 */
1584 if (entry->preferred_base)
1585 goto next;
1586
1587 /*
1588 * If the current object is at pack edge, take the depth the
1589 * objects that depend on the current object into account
1590 * otherwise they would become too deep.
1591 */
1592 max_depth = depth;
1593 if (entry->delta_child) {
1594 max_depth -= check_delta_limit(entry, 0);
1595 if (max_depth <= 0)
1596 goto next;
1597 }
1598
1599 j = window;
1600 while (--j > 0) {
1601 int ret;
1602 uint32_t other_idx = idx + j;
1603 struct unpacked *m;
1604 if (other_idx >= window)
1605 other_idx -= window;
1606 m = array + other_idx;
1607 if (!m->entry)
1608 break;
1609 ret = try_delta(n, m, max_depth, &mem_usage);
1610 if (ret < 0)
1611 break;
1612 else if (ret > 0)
1613 best_base = other_idx;
1614 }
1615
1616 /*
1617 * If we decided to cache the delta data, then it is best
1618 * to compress it right away. First because we have to do
1619 * it anyway, and doing it here while we're threaded will
1620 * save a lot of time in the non threaded write phase,
1621 * as well as allow for caching more deltas within
1622 * the same cache size limit.
1623 * ...
1624 * But only if not writing to stdout, since in that case
1625 * the network is most likely throttling writes anyway,
1626 * and therefore it is best to go to the write phase ASAP
1627 * instead, as we can afford spending more time compressing
1628 * between writes at that moment.
1629 */
1630 if (entry->delta_data && !pack_to_stdout) {
1631 entry->z_delta_size = do_compress(&entry->delta_data,
1632 entry->delta_size);
1633 cache_lock();
1634 delta_cache_size -= entry->delta_size;
1635 delta_cache_size += entry->z_delta_size;
1636 cache_unlock();
1637 }
1638
1639 /* if we made n a delta, and if n is already at max
1640 * depth, leaving it in the window is pointless. we
1641 * should evict it first.
1642 */
1643 if (entry->delta && max_depth <= n->depth)
1644 continue;
1645
1646 /*
1647 * Move the best delta base up in the window, after the
1648 * currently deltified object, to keep it longer. It will
1649 * be the first base object to be attempted next.
1650 */
1651 if (entry->delta) {
1652 struct unpacked swap = array[best_base];
1653 int dist = (window + idx - best_base) % window;
1654 int dst = best_base;
1655 while (dist--) {
1656 int src = (dst + 1) % window;
1657 array[dst] = array[src];
1658 dst = src;
1659 }
1660 array[dst] = swap;
1661 }
1662
1663 next:
1664 idx++;
1665 if (count + 1 < window)
1666 count++;
1667 if (idx >= window)
1668 idx = 0;
1669 }
1670
1671 for (i = 0; i < window; ++i) {
1672 free_delta_index(array[i].index);
1673 free(array[i].data);
1674 }
1675 free(array);
1676 }
1677
1678 #ifndef NO_PTHREADS
1679
1680 static void try_to_free_from_threads(size_t size)
1681 {
1682 read_lock();
1683 release_pack_memory(size, -1);
1684 read_unlock();
1685 }
1686
1687 static try_to_free_t old_try_to_free_routine;
1688
1689 /*
1690 * The main thread waits on the condition that (at least) one of the workers
1691 * has stopped working (which is indicated in the .working member of
1692 * struct thread_params).
1693 * When a work thread has completed its work, it sets .working to 0 and
1694 * signals the main thread and waits on the condition that .data_ready
1695 * becomes 1.
1696 */
1697
1698 struct thread_params {
1699 pthread_t thread;
1700 struct object_entry **list;
1701 unsigned list_size;
1702 unsigned remaining;
1703 int window;
1704 int depth;
1705 int working;
1706 int data_ready;
1707 pthread_mutex_t mutex;
1708 pthread_cond_t cond;
1709 unsigned *processed;
1710 };
1711
1712 static pthread_cond_t progress_cond;
1713
1714 /*
1715 * Mutex and conditional variable can't be statically-initialized on Windows.
1716 */
1717 static void init_threaded_search(void)
1718 {
1719 init_recursive_mutex(&read_mutex);
1720 pthread_mutex_init(&cache_mutex, NULL);
1721 pthread_mutex_init(&progress_mutex, NULL);
1722 pthread_cond_init(&progress_cond, NULL);
1723 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1724 }
1725
1726 static void cleanup_threaded_search(void)
1727 {
1728 set_try_to_free_routine(old_try_to_free_routine);
1729 pthread_cond_destroy(&progress_cond);
1730 pthread_mutex_destroy(&read_mutex);
1731 pthread_mutex_destroy(&cache_mutex);
1732 pthread_mutex_destroy(&progress_mutex);
1733 }
1734
1735 static void *threaded_find_deltas(void *arg)
1736 {
1737 struct thread_params *me = arg;
1738
1739 while (me->remaining) {
1740 find_deltas(me->list, &me->remaining,
1741 me->window, me->depth, me->processed);
1742
1743 progress_lock();
1744 me->working = 0;
1745 pthread_cond_signal(&progress_cond);
1746 progress_unlock();
1747
1748 /*
1749 * We must not set ->data_ready before we wait on the
1750 * condition because the main thread may have set it to 1
1751 * before we get here. In order to be sure that new
1752 * work is available if we see 1 in ->data_ready, it
1753 * was initialized to 0 before this thread was spawned
1754 * and we reset it to 0 right away.
1755 */
1756 pthread_mutex_lock(&me->mutex);
1757 while (!me->data_ready)
1758 pthread_cond_wait(&me->cond, &me->mutex);
1759 me->data_ready = 0;
1760 pthread_mutex_unlock(&me->mutex);
1761 }
1762 /* leave ->working 1 so that this doesn't get more work assigned */
1763 return NULL;
1764 }
1765
1766 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1767 int window, int depth, unsigned *processed)
1768 {
1769 struct thread_params *p;
1770 int i, ret, active_threads = 0;
1771
1772 init_threaded_search();
1773
1774 if (!delta_search_threads) /* --threads=0 means autodetect */
1775 delta_search_threads = online_cpus();
1776 if (delta_search_threads <= 1) {
1777 find_deltas(list, &list_size, window, depth, processed);
1778 cleanup_threaded_search();
1779 return;
1780 }
1781 if (progress > pack_to_stdout)
1782 fprintf(stderr, "Delta compression using up to %d threads.\n",
1783 delta_search_threads);
1784 p = xcalloc(delta_search_threads, sizeof(*p));
1785
1786 /* Partition the work amongst work threads. */
1787 for (i = 0; i < delta_search_threads; i++) {
1788 unsigned sub_size = list_size / (delta_search_threads - i);
1789
1790 /* don't use too small segments or no deltas will be found */
1791 if (sub_size < 2*window && i+1 < delta_search_threads)
1792 sub_size = 0;
1793
1794 p[i].window = window;
1795 p[i].depth = depth;
1796 p[i].processed = processed;
1797 p[i].working = 1;
1798 p[i].data_ready = 0;
1799
1800 /* try to split chunks on "path" boundaries */
1801 while (sub_size && sub_size < list_size &&
1802 list[sub_size]->hash &&
1803 list[sub_size]->hash == list[sub_size-1]->hash)
1804 sub_size++;
1805
1806 p[i].list = list;
1807 p[i].list_size = sub_size;
1808 p[i].remaining = sub_size;
1809
1810 list += sub_size;
1811 list_size -= sub_size;
1812 }
1813
1814 /* Start work threads. */
1815 for (i = 0; i < delta_search_threads; i++) {
1816 if (!p[i].list_size)
1817 continue;
1818 pthread_mutex_init(&p[i].mutex, NULL);
1819 pthread_cond_init(&p[i].cond, NULL);
1820 ret = pthread_create(&p[i].thread, NULL,
1821 threaded_find_deltas, &p[i]);
1822 if (ret)
1823 die("unable to create thread: %s", strerror(ret));
1824 active_threads++;
1825 }
1826
1827 /*
1828 * Now let's wait for work completion. Each time a thread is done
1829 * with its work, we steal half of the remaining work from the
1830 * thread with the largest number of unprocessed objects and give
1831 * it to that newly idle thread. This ensure good load balancing
1832 * until the remaining object list segments are simply too short
1833 * to be worth splitting anymore.
1834 */
1835 while (active_threads) {
1836 struct thread_params *target = NULL;
1837 struct thread_params *victim = NULL;
1838 unsigned sub_size = 0;
1839
1840 progress_lock();
1841 for (;;) {
1842 for (i = 0; !target && i < delta_search_threads; i++)
1843 if (!p[i].working)
1844 target = &p[i];
1845 if (target)
1846 break;
1847 pthread_cond_wait(&progress_cond, &progress_mutex);
1848 }
1849
1850 for (i = 0; i < delta_search_threads; i++)
1851 if (p[i].remaining > 2*window &&
1852 (!victim || victim->remaining < p[i].remaining))
1853 victim = &p[i];
1854 if (victim) {
1855 sub_size = victim->remaining / 2;
1856 list = victim->list + victim->list_size - sub_size;
1857 while (sub_size && list[0]->hash &&
1858 list[0]->hash == list[-1]->hash) {
1859 list++;
1860 sub_size--;
1861 }
1862 if (!sub_size) {
1863 /*
1864 * It is possible for some "paths" to have
1865 * so many objects that no hash boundary
1866 * might be found. Let's just steal the
1867 * exact half in that case.
1868 */
1869 sub_size = victim->remaining / 2;
1870 list -= sub_size;
1871 }
1872 target->list = list;
1873 victim->list_size -= sub_size;
1874 victim->remaining -= sub_size;
1875 }
1876 target->list_size = sub_size;
1877 target->remaining = sub_size;
1878 target->working = 1;
1879 progress_unlock();
1880
1881 pthread_mutex_lock(&target->mutex);
1882 target->data_ready = 1;
1883 pthread_cond_signal(&target->cond);
1884 pthread_mutex_unlock(&target->mutex);
1885
1886 if (!sub_size) {
1887 pthread_join(target->thread, NULL);
1888 pthread_cond_destroy(&target->cond);
1889 pthread_mutex_destroy(&target->mutex);
1890 active_threads--;
1891 }
1892 }
1893 cleanup_threaded_search();
1894 free(p);
1895 }
1896
1897 #else
1898 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1899 #endif
1900
1901 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1902 {
1903 unsigned char peeled[20];
1904
1905 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1906 !peel_ref(path, peeled) && /* peelable? */
1907 !is_null_sha1(peeled) && /* annotated tag? */
1908 locate_object_entry(peeled)) /* object packed? */
1909 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1910 return 0;
1911 }
1912
1913 static void prepare_pack(int window, int depth)
1914 {
1915 struct object_entry **delta_list;
1916 uint32_t i, nr_deltas;
1917 unsigned n;
1918
1919 get_object_details();
1920
1921 /*
1922 * If we're locally repacking then we need to be doubly careful
1923 * from now on in order to make sure no stealth corruption gets
1924 * propagated to the new pack. Clients receiving streamed packs
1925 * should validate everything they get anyway so no need to incur
1926 * the additional cost here in that case.
1927 */
1928 if (!pack_to_stdout)
1929 do_check_packed_object_crc = 1;
1930
1931 if (!nr_objects || !window || !depth)
1932 return;
1933
1934 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1935 nr_deltas = n = 0;
1936
1937 for (i = 0; i < nr_objects; i++) {
1938 struct object_entry *entry = objects + i;
1939
1940 if (entry->delta)
1941 /* This happens if we decided to reuse existing
1942 * delta from a pack. "reuse_delta &&" is implied.
1943 */
1944 continue;
1945
1946 if (entry->size < 50)
1947 continue;
1948
1949 if (entry->no_try_delta)
1950 continue;
1951
1952 if (!entry->preferred_base) {
1953 nr_deltas++;
1954 if (entry->type < 0)
1955 die("unable to get type of object %s",
1956 sha1_to_hex(entry->idx.sha1));
1957 } else {
1958 if (entry->type < 0) {
1959 /*
1960 * This object is not found, but we
1961 * don't have to include it anyway.
1962 */
1963 continue;
1964 }
1965 }
1966
1967 delta_list[n++] = entry;
1968 }
1969
1970 if (nr_deltas && n > 1) {
1971 unsigned nr_done = 0;
1972 if (progress)
1973 progress_state = start_progress("Compressing objects",
1974 nr_deltas);
1975 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1976 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1977 stop_progress(&progress_state);
1978 if (nr_done != nr_deltas)
1979 die("inconsistency with delta count");
1980 }
1981 free(delta_list);
1982 }
1983
1984 static int git_pack_config(const char *k, const char *v, void *cb)
1985 {
1986 if (!strcmp(k, "pack.window")) {
1987 window = git_config_int(k, v);
1988 return 0;
1989 }
1990 if (!strcmp(k, "pack.windowmemory")) {
1991 window_memory_limit = git_config_ulong(k, v);
1992 return 0;
1993 }
1994 if (!strcmp(k, "pack.depth")) {
1995 depth = git_config_int(k, v);
1996 return 0;
1997 }
1998 if (!strcmp(k, "pack.compression")) {
1999 int level = git_config_int(k, v);
2000 if (level == -1)
2001 level = Z_DEFAULT_COMPRESSION;
2002 else if (level < 0 || level > Z_BEST_COMPRESSION)
2003 die("bad pack compression level %d", level);
2004 pack_compression_level = level;
2005 pack_compression_seen = 1;
2006 return 0;
2007 }
2008 if (!strcmp(k, "pack.deltacachesize")) {
2009 max_delta_cache_size = git_config_int(k, v);
2010 return 0;
2011 }
2012 if (!strcmp(k, "pack.deltacachelimit")) {
2013 cache_max_small_delta_size = git_config_int(k, v);
2014 return 0;
2015 }
2016 if (!strcmp(k, "pack.threads")) {
2017 delta_search_threads = git_config_int(k, v);
2018 if (delta_search_threads < 0)
2019 die("invalid number of threads specified (%d)",
2020 delta_search_threads);
2021 #ifdef NO_PTHREADS
2022 if (delta_search_threads != 1)
2023 warning("no threads support, ignoring %s", k);
2024 #endif
2025 return 0;
2026 }
2027 if (!strcmp(k, "pack.indexversion")) {
2028 pack_idx_opts.version = git_config_int(k, v);
2029 if (pack_idx_opts.version > 2)
2030 die("bad pack.indexversion=%"PRIu32,
2031 pack_idx_opts.version);
2032 return 0;
2033 }
2034 if (!strcmp(k, "pack.packsizelimit")) {
2035 pack_size_limit_cfg = git_config_ulong(k, v);
2036 return 0;
2037 }
2038 return git_default_config(k, v, cb);
2039 }
2040
2041 static void read_object_list_from_stdin(void)
2042 {
2043 char line[40 + 1 + PATH_MAX + 2];
2044 unsigned char sha1[20];
2045
2046 for (;;) {
2047 if (!fgets(line, sizeof(line), stdin)) {
2048 if (feof(stdin))
2049 break;
2050 if (!ferror(stdin))
2051 die("fgets returned NULL, not EOF, not error!");
2052 if (errno != EINTR)
2053 die_errno("fgets");
2054 clearerr(stdin);
2055 continue;
2056 }
2057 if (line[0] == '-') {
2058 if (get_sha1_hex(line+1, sha1))
2059 die("expected edge sha1, got garbage:\n %s",
2060 line);
2061 add_preferred_base(sha1);
2062 continue;
2063 }
2064 if (get_sha1_hex(line, sha1))
2065 die("expected sha1, got garbage:\n %s", line);
2066
2067 add_preferred_base_object(line+41);
2068 add_object_entry(sha1, 0, line+41, 0);
2069 }
2070 }
2071
2072 #define OBJECT_ADDED (1u<<20)
2073
2074 static void show_commit(struct commit *commit, void *data)
2075 {
2076 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
2077 commit->object.flags |= OBJECT_ADDED;
2078 }
2079
2080 static void show_object(struct object *obj,
2081 const struct name_path *path, const char *last,
2082 void *data)
2083 {
2084 char *name = path_name(path, last);
2085
2086 add_preferred_base_object(name);
2087 add_object_entry(obj->sha1, obj->type, name, 0);
2088 obj->flags |= OBJECT_ADDED;
2089
2090 /*
2091 * We will have generated the hash from the name,
2092 * but not saved a pointer to it - we can free it
2093 */
2094 free((char *)name);
2095 }
2096
2097 static void show_edge(struct commit *commit)
2098 {
2099 add_preferred_base(commit->object.sha1);
2100 }
2101
2102 struct in_pack_object {
2103 off_t offset;
2104 struct object *object;
2105 };
2106
2107 struct in_pack {
2108 int alloc;
2109 int nr;
2110 struct in_pack_object *array;
2111 };
2112
2113 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2114 {
2115 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2116 in_pack->array[in_pack->nr].object = object;
2117 in_pack->nr++;
2118 }
2119
2120 /*
2121 * Compare the objects in the offset order, in order to emulate the
2122 * "git rev-list --objects" output that produced the pack originally.
2123 */
2124 static int ofscmp(const void *a_, const void *b_)
2125 {
2126 struct in_pack_object *a = (struct in_pack_object *)a_;
2127 struct in_pack_object *b = (struct in_pack_object *)b_;
2128
2129 if (a->offset < b->offset)
2130 return -1;
2131 else if (a->offset > b->offset)
2132 return 1;
2133 else
2134 return hashcmp(a->object->sha1, b->object->sha1);
2135 }
2136
2137 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2138 {
2139 struct packed_git *p;
2140 struct in_pack in_pack;
2141 uint32_t i;
2142
2143 memset(&in_pack, 0, sizeof(in_pack));
2144
2145 for (p = packed_git; p; p = p->next) {
2146 const unsigned char *sha1;
2147 struct object *o;
2148
2149 if (!p->pack_local || p->pack_keep)
2150 continue;
2151 if (open_pack_index(p))
2152 die("cannot open pack index");
2153
2154 ALLOC_GROW(in_pack.array,
2155 in_pack.nr + p->num_objects,
2156 in_pack.alloc);
2157
2158 for (i = 0; i < p->num_objects; i++) {
2159 sha1 = nth_packed_object_sha1(p, i);
2160 o = lookup_unknown_object(sha1);
2161 if (!(o->flags & OBJECT_ADDED))
2162 mark_in_pack_object(o, p, &in_pack);
2163 o->flags |= OBJECT_ADDED;
2164 }
2165 }
2166
2167 if (in_pack.nr) {
2168 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2169 ofscmp);
2170 for (i = 0; i < in_pack.nr; i++) {
2171 struct object *o = in_pack.array[i].object;
2172 add_object_entry(o->sha1, o->type, "", 0);
2173 }
2174 }
2175 free(in_pack.array);
2176 }
2177
2178 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2179 {
2180 static struct packed_git *last_found = (void *)1;
2181 struct packed_git *p;
2182
2183 p = (last_found != (void *)1) ? last_found : packed_git;
2184
2185 while (p) {
2186 if ((!p->pack_local || p->pack_keep) &&
2187 find_pack_entry_one(sha1, p)) {
2188 last_found = p;
2189 return 1;
2190 }
2191 if (p == last_found)
2192 p = packed_git;
2193 else
2194 p = p->next;
2195 if (p == last_found)
2196 p = p->next;
2197 }
2198 return 0;
2199 }
2200
2201 static void loosen_unused_packed_objects(struct rev_info *revs)
2202 {
2203 struct packed_git *p;
2204 uint32_t i;
2205 const unsigned char *sha1;
2206
2207 for (p = packed_git; p; p = p->next) {
2208 if (!p->pack_local || p->pack_keep)
2209 continue;
2210
2211 if (open_pack_index(p))
2212 die("cannot open pack index");
2213
2214 for (i = 0; i < p->num_objects; i++) {
2215 sha1 = nth_packed_object_sha1(p, i);
2216 if (!locate_object_entry(sha1) &&
2217 !has_sha1_pack_kept_or_nonlocal(sha1))
2218 if (force_object_loose(sha1, p->mtime))
2219 die("unable to force loose object");
2220 }
2221 }
2222 }
2223
2224 static void get_object_list(int ac, const char **av)
2225 {
2226 struct rev_info revs;
2227 char line[1000];
2228 int flags = 0;
2229
2230 init_revisions(&revs, NULL);
2231 save_commit_buffer = 0;
2232 setup_revisions(ac, av, &revs, NULL);
2233
2234 while (fgets(line, sizeof(line), stdin) != NULL) {
2235 int len = strlen(line);
2236 if (len && line[len - 1] == '\n')
2237 line[--len] = 0;
2238 if (!len)
2239 break;
2240 if (*line == '-') {
2241 if (!strcmp(line, "--not")) {
2242 flags ^= UNINTERESTING;
2243 continue;
2244 }
2245 die("not a rev '%s'", line);
2246 }
2247 if (handle_revision_arg(line, &revs, flags, 1))
2248 die("bad revision '%s'", line);
2249 }
2250
2251 if (prepare_revision_walk(&revs))
2252 die("revision walk setup failed");
2253 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2254 traverse_commit_list(&revs, show_commit, show_object, NULL);
2255
2256 if (keep_unreachable)
2257 add_objects_in_unpacked_packs(&revs);
2258 if (unpack_unreachable)
2259 loosen_unused_packed_objects(&revs);
2260 }
2261
2262 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2263 {
2264 int use_internal_rev_list = 0;
2265 int thin = 0;
2266 int all_progress_implied = 0;
2267 uint32_t i;
2268 const char **rp_av;
2269 int rp_ac_alloc = 64;
2270 int rp_ac;
2271
2272 read_replace_refs = 0;
2273
2274 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2275
2276 rp_av[0] = "pack-objects";
2277 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2278 rp_ac = 2;
2279
2280 reset_pack_idx_option(&pack_idx_opts);
2281 git_config(git_pack_config, NULL);
2282 if (!pack_compression_seen && core_compression_seen)
2283 pack_compression_level = core_compression_level;
2284
2285 progress = isatty(2);
2286 for (i = 1; i < argc; i++) {
2287 const char *arg = argv[i];
2288
2289 if (*arg != '-')
2290 break;
2291
2292 if (!strcmp("--non-empty", arg)) {
2293 non_empty = 1;
2294 continue;
2295 }
2296 if (!strcmp("--local", arg)) {
2297 local = 1;
2298 continue;
2299 }
2300 if (!strcmp("--incremental", arg)) {
2301 incremental = 1;
2302 continue;
2303 }
2304 if (!strcmp("--honor-pack-keep", arg)) {
2305 ignore_packed_keep = 1;
2306 continue;
2307 }
2308 if (!prefixcmp(arg, "--compression=")) {
2309 char *end;
2310 int level = strtoul(arg+14, &end, 0);
2311 if (!arg[14] || *end)
2312 usage(pack_usage);
2313 if (level == -1)
2314 level = Z_DEFAULT_COMPRESSION;
2315 else if (level < 0 || level > Z_BEST_COMPRESSION)
2316 die("bad pack compression level %d", level);
2317 pack_compression_level = level;
2318 continue;
2319 }
2320 if (!prefixcmp(arg, "--max-pack-size=")) {
2321 pack_size_limit_cfg = 0;
2322 if (!git_parse_ulong(arg+16, &pack_size_limit))
2323 usage(pack_usage);
2324 continue;
2325 }
2326 if (!prefixcmp(arg, "--window=")) {
2327 char *end;
2328 window = strtoul(arg+9, &end, 0);
2329 if (!arg[9] || *end)
2330 usage(pack_usage);
2331 continue;
2332 }
2333 if (!prefixcmp(arg, "--window-memory=")) {
2334 if (!git_parse_ulong(arg+16, &window_memory_limit))
2335 usage(pack_usage);
2336 continue;
2337 }
2338 if (!prefixcmp(arg, "--threads=")) {
2339 char *end;
2340 delta_search_threads = strtoul(arg+10, &end, 0);
2341 if (!arg[10] || *end || delta_search_threads < 0)
2342 usage(pack_usage);
2343 #ifdef NO_PTHREADS
2344 if (delta_search_threads != 1)
2345 warning("no threads support, "
2346 "ignoring %s", arg);
2347 #endif
2348 continue;
2349 }
2350 if (!prefixcmp(arg, "--depth=")) {
2351 char *end;
2352 depth = strtoul(arg+8, &end, 0);
2353 if (!arg[8] || *end)
2354 usage(pack_usage);
2355 continue;
2356 }
2357 if (!strcmp("--progress", arg)) {
2358 progress = 1;
2359 continue;
2360 }
2361 if (!strcmp("--all-progress", arg)) {
2362 progress = 2;
2363 continue;
2364 }
2365 if (!strcmp("--all-progress-implied", arg)) {
2366 all_progress_implied = 1;
2367 continue;
2368 }
2369 if (!strcmp("-q", arg)) {
2370 progress = 0;
2371 continue;
2372 }
2373 if (!strcmp("--no-reuse-delta", arg)) {
2374 reuse_delta = 0;
2375 continue;
2376 }
2377 if (!strcmp("--no-reuse-object", arg)) {
2378 reuse_object = reuse_delta = 0;
2379 continue;
2380 }
2381 if (!strcmp("--delta-base-offset", arg)) {
2382 allow_ofs_delta = 1;
2383 continue;
2384 }
2385 if (!strcmp("--stdout", arg)) {
2386 pack_to_stdout = 1;
2387 continue;
2388 }
2389 if (!strcmp("--revs", arg)) {
2390 use_internal_rev_list = 1;
2391 continue;
2392 }
2393 if (!strcmp("--keep-unreachable", arg)) {
2394 keep_unreachable = 1;
2395 continue;
2396 }
2397 if (!strcmp("--unpack-unreachable", arg)) {
2398 unpack_unreachable = 1;
2399 continue;
2400 }
2401 if (!strcmp("--include-tag", arg)) {
2402 include_tag = 1;
2403 continue;
2404 }
2405 if (!strcmp("--unpacked", arg) ||
2406 !strcmp("--reflog", arg) ||
2407 !strcmp("--all", arg)) {
2408 use_internal_rev_list = 1;
2409 if (rp_ac >= rp_ac_alloc - 1) {
2410 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2411 rp_av = xrealloc(rp_av,
2412 rp_ac_alloc * sizeof(*rp_av));
2413 }
2414 rp_av[rp_ac++] = arg;
2415 continue;
2416 }
2417 if (!strcmp("--thin", arg)) {
2418 use_internal_rev_list = 1;
2419 thin = 1;
2420 rp_av[1] = "--objects-edge";
2421 continue;
2422 }
2423 if (!prefixcmp(arg, "--index-version=")) {
2424 char *c;
2425 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2426 if (pack_idx_opts.version > 2)
2427 die("bad %s", arg);
2428 if (*c == ',')
2429 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2430 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2431 die("bad %s", arg);
2432 continue;
2433 }
2434 if (!strcmp(arg, "--keep-true-parents")) {
2435 grafts_replace_parents = 0;
2436 continue;
2437 }
2438 usage(pack_usage);
2439 }
2440
2441 /* Traditionally "pack-objects [options] base extra" failed;
2442 * we would however want to take refs parameter that would
2443 * have been given to upstream rev-list ourselves, which means
2444 * we somehow want to say what the base name is. So the
2445 * syntax would be:
2446 *
2447 * pack-objects [options] base <refs...>
2448 *
2449 * in other words, we would treat the first non-option as the
2450 * base_name and send everything else to the internal revision
2451 * walker.
2452 */
2453
2454 if (!pack_to_stdout)
2455 base_name = argv[i++];
2456
2457 if (pack_to_stdout != !base_name)
2458 usage(pack_usage);
2459
2460 if (!pack_to_stdout && !pack_size_limit)
2461 pack_size_limit = pack_size_limit_cfg;
2462 if (pack_to_stdout && pack_size_limit)
2463 die("--max-pack-size cannot be used to build a pack for transfer.");
2464 if (pack_size_limit && pack_size_limit < 1024*1024) {
2465 warning("minimum pack size limit is 1 MiB");
2466 pack_size_limit = 1024*1024;
2467 }
2468
2469 if (!pack_to_stdout && thin)
2470 die("--thin cannot be used to build an indexable pack.");
2471
2472 if (keep_unreachable && unpack_unreachable)
2473 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2474
2475 if (progress && all_progress_implied)
2476 progress = 2;
2477
2478 prepare_packed_git();
2479
2480 if (progress)
2481 progress_state = start_progress("Counting objects", 0);
2482 if (!use_internal_rev_list)
2483 read_object_list_from_stdin();
2484 else {
2485 rp_av[rp_ac] = NULL;
2486 get_object_list(rp_ac, rp_av);
2487 }
2488 cleanup_preferred_base();
2489 if (include_tag && nr_result)
2490 for_each_ref(add_ref_tag, NULL);
2491 stop_progress(&progress_state);
2492
2493 if (non_empty && !nr_result)
2494 return 0;
2495 if (nr_result)
2496 prepare_pack(window, depth);
2497 write_pack_file();
2498 if (progress)
2499 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2500 " reused %"PRIu32" (delta %"PRIu32")\n",
2501 written, written_delta, reused, reused_delta);
2502 return 0;
2503 }