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