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