![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "blob.h"
5 #include "commit.h"
6 #include "tag.h"
7 #include "tree.h"
8 #include "delta.h"
9 #include "pack.h"
10 #include "csum-file.h"
11 #include "tree-walk.h"
12 #include "diff.h"
13 #include "revision.h"
14 #include "list-objects.h"
16 static const char pack_usage[] = "\
17 git-pack-objects [{ -q | --progress | --all-progress }] \n\
18 [--local] [--incremental] [--window=N] [--depth=N] \n\
19 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
20 [--revs [--unpacked | --all]*] [--reflog] [--stdout | base-name] \n\
21 [<ref-list | <object-list]";
23 struct object_entry {
24 unsigned char sha1[20];
25 unsigned long size; /* uncompressed size */
26 off_t offset; /* offset into the final pack file;
27 * nonzero if already written.
28 */
29 unsigned int depth; /* delta depth */
30 unsigned int delta_limit; /* base adjustment for in-pack delta */
31 unsigned int hash; /* name hint hash */
32 enum object_type type;
33 enum object_type in_pack_type; /* could be delta */
34 unsigned long delta_size; /* delta data size (uncompressed) */
35 #define in_pack_header_size delta_size /* only when reusing pack data */
36 struct object_entry *delta; /* delta base object */
37 struct packed_git *in_pack; /* already in pack */
38 off_t in_pack_offset;
39 struct object_entry *delta_child; /* deltified objects who bases me */
40 struct object_entry *delta_sibling; /* other deltified objects who
41 * uses the same base as me
42 */
43 int preferred_base; /* we do not pack this, but is encouraged to
44 * be used as the base objectto delta huge
45 * objects against.
46 */
47 uint32_t crc32; /* crc of raw pack data for this object */
48 };
50 /*
51 * Objects we are going to pack are collected in objects array (dynamically
52 * expanded). nr_objects & nr_alloc controls this array. They are stored
53 * in the order we see -- typically rev-list --objects order that gives us
54 * nice "minimum seek" order.
55 *
56 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
57 * elements in the objects array. The former is used to build the pack
58 * index (lists object names in the ascending order to help offset lookup),
59 * and the latter is used to group similar things together by try_delta()
60 * heuristics.
61 */
63 static unsigned char object_list_sha1[20];
64 static int non_empty;
65 static int no_reuse_delta;
66 static int local;
67 static int incremental;
68 static int allow_ofs_delta;
70 static struct object_entry **sorted_by_sha, **sorted_by_type;
71 static struct object_entry *objects;
72 static uint32_t nr_objects, nr_alloc, nr_result;
73 static const char *base_name;
74 static unsigned char pack_file_sha1[20];
75 static int progress = 1;
76 static volatile sig_atomic_t progress_update;
77 static int window = 10;
78 static int pack_to_stdout;
79 static int num_preferred_base;
81 /*
82 * The object names in objects array are hashed with this hashtable,
83 * to help looking up the entry by object name. Binary search from
84 * sorted_by_sha is also possible but this was easier to code and faster.
85 * This hashtable is built after all the objects are seen.
86 */
87 static int *object_ix;
88 static int object_ix_hashsz;
90 /*
91 * Pack index for existing packs give us easy access to the offsets into
92 * corresponding pack file where each object's data starts, but the entries
93 * do not store the size of the compressed representation (uncompressed
94 * size is easily available by examining the pack entry header). It is
95 * also rather expensive to find the sha1 for an object given its offset.
96 *
97 * We build a hashtable of existing packs (pack_revindex), and keep reverse
98 * index here -- pack index file is sorted by object name mapping to offset;
99 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
100 * ordered by offset, so if you know the offset of an object, next offset
101 * is where its packed representation ends and the index_nr can be used to
102 * get the object sha1 from the main index.
103 */
104 struct revindex_entry {
105 off_t offset;
106 unsigned int nr;
107 };
108 struct pack_revindex {
109 struct packed_git *p;
110 struct revindex_entry *revindex;
111 };
112 static struct pack_revindex *pack_revindex;
113 static int pack_revindex_hashsz;
115 /*
116 * stats
117 */
118 static uint32_t written, written_delta;
119 static uint32_t reused, reused_delta;
121 static int pack_revindex_ix(struct packed_git *p)
122 {
123 unsigned long ui = (unsigned long)p;
124 int i;
126 ui = ui ^ (ui >> 16); /* defeat structure alignment */
127 i = (int)(ui % pack_revindex_hashsz);
128 while (pack_revindex[i].p) {
129 if (pack_revindex[i].p == p)
130 return i;
131 if (++i == pack_revindex_hashsz)
132 i = 0;
133 }
134 return -1 - i;
135 }
137 static void prepare_pack_ix(void)
138 {
139 int num;
140 struct packed_git *p;
141 for (num = 0, p = packed_git; p; p = p->next)
142 num++;
143 if (!num)
144 return;
145 pack_revindex_hashsz = num * 11;
146 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
147 for (p = packed_git; p; p = p->next) {
148 num = pack_revindex_ix(p);
149 num = - 1 - num;
150 pack_revindex[num].p = p;
151 }
152 /* revindex elements are lazily initialized */
153 }
155 static int cmp_offset(const void *a_, const void *b_)
156 {
157 const struct revindex_entry *a = a_;
158 const struct revindex_entry *b = b_;
159 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
160 }
162 /*
163 * Ordered list of offsets of objects in the pack.
164 */
165 static void prepare_pack_revindex(struct pack_revindex *rix)
166 {
167 struct packed_git *p = rix->p;
168 int num_ent = p->num_objects;
169 int i;
170 const char *index = p->index_data;
172 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
173 index += 4 * 256;
175 if (p->index_version > 1) {
176 const uint32_t *off_32 =
177 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
178 const uint32_t *off_64 = off_32 + p->num_objects;
179 for (i = 0; i < num_ent; i++) {
180 uint32_t off = ntohl(*off_32++);
181 if (!(off & 0x80000000)) {
182 rix->revindex[i].offset = off;
183 } else {
184 rix->revindex[i].offset =
185 ((uint64_t)ntohl(*off_64++)) << 32;
186 rix->revindex[i].offset |=
187 ntohl(*off_64++);
188 }
189 rix->revindex[i].nr = i;
190 }
191 } else {
192 for (i = 0; i < num_ent; i++) {
193 uint32_t hl = *((uint32_t *)(index + 24 * i));
194 rix->revindex[i].offset = ntohl(hl);
195 rix->revindex[i].nr = i;
196 }
197 }
199 /* This knows the pack format -- the 20-byte trailer
200 * follows immediately after the last object data.
201 */
202 rix->revindex[num_ent].offset = p->pack_size - 20;
203 rix->revindex[num_ent].nr = -1;
204 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
205 }
207 static struct revindex_entry * find_packed_object(struct packed_git *p,
208 off_t ofs)
209 {
210 int num;
211 int lo, hi;
212 struct pack_revindex *rix;
213 struct revindex_entry *revindex;
214 num = pack_revindex_ix(p);
215 if (num < 0)
216 die("internal error: pack revindex uninitialized");
217 rix = &pack_revindex[num];
218 if (!rix->revindex)
219 prepare_pack_revindex(rix);
220 revindex = rix->revindex;
221 lo = 0;
222 hi = p->num_objects + 1;
223 do {
224 int mi = (lo + hi) / 2;
225 if (revindex[mi].offset == ofs) {
226 return revindex + mi;
227 }
228 else if (ofs < revindex[mi].offset)
229 hi = mi;
230 else
231 lo = mi + 1;
232 } while (lo < hi);
233 die("internal error: pack revindex corrupt");
234 }
236 static off_t find_packed_object_size(struct packed_git *p, off_t ofs)
237 {
238 struct revindex_entry *entry = find_packed_object(p, ofs);
239 return entry[1].offset - ofs;
240 }
242 static const unsigned char *find_packed_object_name(struct packed_git *p,
243 off_t ofs)
244 {
245 struct revindex_entry *entry = find_packed_object(p, ofs);
246 return nth_packed_object_sha1(p, entry->nr);
247 }
249 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
250 {
251 unsigned long othersize, delta_size;
252 enum object_type type;
253 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
254 void *delta_buf;
256 if (!otherbuf)
257 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
258 delta_buf = diff_delta(otherbuf, othersize,
259 buf, size, &delta_size, 0);
260 if (!delta_buf || delta_size != entry->delta_size)
261 die("delta size changed");
262 free(buf);
263 free(otherbuf);
264 return delta_buf;
265 }
267 /*
268 * The per-object header is a pretty dense thing, which is
269 * - first byte: low four bits are "size", then three bits of "type",
270 * and the high bit is "size continues".
271 * - each byte afterwards: low seven bits are size continuation,
272 * with the high bit being "size continues"
273 */
274 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
275 {
276 int n = 1;
277 unsigned char c;
279 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
280 die("bad type %d", type);
282 c = (type << 4) | (size & 15);
283 size >>= 4;
284 while (size) {
285 *hdr++ = c | 0x80;
286 c = size & 0x7f;
287 size >>= 7;
288 n++;
289 }
290 *hdr = c;
291 return n;
292 }
294 /*
295 * we are going to reuse the existing object data as is. make
296 * sure it is not corrupt.
297 */
298 static int check_pack_inflate(struct packed_git *p,
299 struct pack_window **w_curs,
300 off_t offset,
301 off_t len,
302 unsigned long expect)
303 {
304 z_stream stream;
305 unsigned char fakebuf[4096], *in;
306 int st;
308 memset(&stream, 0, sizeof(stream));
309 inflateInit(&stream);
310 do {
311 in = use_pack(p, w_curs, offset, &stream.avail_in);
312 stream.next_in = in;
313 stream.next_out = fakebuf;
314 stream.avail_out = sizeof(fakebuf);
315 st = inflate(&stream, Z_FINISH);
316 offset += stream.next_in - in;
317 } while (st == Z_OK || st == Z_BUF_ERROR);
318 inflateEnd(&stream);
319 return (st == Z_STREAM_END &&
320 stream.total_out == expect &&
321 stream.total_in == len) ? 0 : -1;
322 }
324 static void copy_pack_data(struct sha1file *f,
325 struct packed_git *p,
326 struct pack_window **w_curs,
327 off_t offset,
328 off_t len)
329 {
330 unsigned char *in;
331 unsigned int avail;
333 while (len) {
334 in = use_pack(p, w_curs, offset, &avail);
335 if (avail > len)
336 avail = (unsigned int)len;
337 sha1write(f, in, avail);
338 offset += avail;
339 len -= avail;
340 }
341 }
343 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
344 {
345 z_stream stream;
346 unsigned char fakebuf[4096];
347 int st;
349 memset(&stream, 0, sizeof(stream));
350 stream.next_in = data;
351 stream.avail_in = len;
352 stream.next_out = fakebuf;
353 stream.avail_out = sizeof(fakebuf);
354 inflateInit(&stream);
356 while (1) {
357 st = inflate(&stream, Z_FINISH);
358 if (st == Z_STREAM_END || st == Z_OK) {
359 st = (stream.total_out == expect &&
360 stream.total_in == len) ? 0 : -1;
361 break;
362 }
363 if (st != Z_BUF_ERROR) {
364 st = -1;
365 break;
366 }
367 stream.next_out = fakebuf;
368 stream.avail_out = sizeof(fakebuf);
369 }
370 inflateEnd(&stream);
371 return st;
372 }
374 static int revalidate_loose_object(struct object_entry *entry,
375 unsigned char *map,
376 unsigned long mapsize)
377 {
378 /* we already know this is a loose object with new type header. */
379 enum object_type type;
380 unsigned long size, used;
382 if (pack_to_stdout)
383 return 0;
385 used = unpack_object_header_gently(map, mapsize, &type, &size);
386 if (!used)
387 return -1;
388 map += used;
389 mapsize -= used;
390 return check_loose_inflate(map, mapsize, size);
391 }
393 static unsigned long write_object(struct sha1file *f,
394 struct object_entry *entry)
395 {
396 unsigned long size;
397 enum object_type type;
398 void *buf;
399 unsigned char header[10];
400 unsigned hdrlen;
401 off_t datalen;
402 enum object_type obj_type;
403 int to_reuse = 0;
405 if (!pack_to_stdout)
406 crc32_begin(f);
408 obj_type = entry->type;
409 if (! entry->in_pack)
410 to_reuse = 0; /* can't reuse what we don't have */
411 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
412 to_reuse = 1; /* check_object() decided it for us */
413 else if (obj_type != entry->in_pack_type)
414 to_reuse = 0; /* pack has delta which is unusable */
415 else if (entry->delta)
416 to_reuse = 0; /* we want to pack afresh */
417 else
418 to_reuse = 1; /* we have it in-pack undeltified,
419 * and we do not need to deltify it.
420 */
422 if (!entry->in_pack && !entry->delta) {
423 unsigned char *map;
424 unsigned long mapsize;
425 map = map_sha1_file(entry->sha1, &mapsize);
426 if (map && !legacy_loose_object(map)) {
427 /* We can copy straight into the pack file */
428 if (revalidate_loose_object(entry, map, mapsize))
429 die("corrupt loose object %s",
430 sha1_to_hex(entry->sha1));
431 sha1write(f, map, mapsize);
432 munmap(map, mapsize);
433 written++;
434 reused++;
435 return mapsize;
436 }
437 if (map)
438 munmap(map, mapsize);
439 }
441 if (!to_reuse) {
442 buf = read_sha1_file(entry->sha1, &type, &size);
443 if (!buf)
444 die("unable to read %s", sha1_to_hex(entry->sha1));
445 if (size != entry->size)
446 die("object %s size inconsistency (%lu vs %lu)",
447 sha1_to_hex(entry->sha1), size, entry->size);
448 if (entry->delta) {
449 buf = delta_against(buf, size, entry);
450 size = entry->delta_size;
451 obj_type = (allow_ofs_delta && entry->delta->offset) ?
452 OBJ_OFS_DELTA : OBJ_REF_DELTA;
453 }
454 /*
455 * The object header is a byte of 'type' followed by zero or
456 * more bytes of length.
457 */
458 hdrlen = encode_header(obj_type, size, header);
459 sha1write(f, header, hdrlen);
461 if (obj_type == OBJ_OFS_DELTA) {
462 /*
463 * Deltas with relative base contain an additional
464 * encoding of the relative offset for the delta
465 * base from this object's position in the pack.
466 */
467 off_t ofs = entry->offset - entry->delta->offset;
468 unsigned pos = sizeof(header) - 1;
469 header[pos] = ofs & 127;
470 while (ofs >>= 7)
471 header[--pos] = 128 | (--ofs & 127);
472 sha1write(f, header + pos, sizeof(header) - pos);
473 hdrlen += sizeof(header) - pos;
474 } else if (obj_type == OBJ_REF_DELTA) {
475 /*
476 * Deltas with a base reference contain
477 * an additional 20 bytes for the base sha1.
478 */
479 sha1write(f, entry->delta->sha1, 20);
480 hdrlen += 20;
481 }
482 datalen = sha1write_compressed(f, buf, size);
483 free(buf);
484 }
485 else {
486 struct packed_git *p = entry->in_pack;
487 struct pack_window *w_curs = NULL;
488 off_t offset;
490 if (entry->delta) {
491 obj_type = (allow_ofs_delta && entry->delta->offset) ?
492 OBJ_OFS_DELTA : OBJ_REF_DELTA;
493 reused_delta++;
494 }
495 hdrlen = encode_header(obj_type, entry->size, header);
496 sha1write(f, header, hdrlen);
497 if (obj_type == OBJ_OFS_DELTA) {
498 off_t ofs = entry->offset - entry->delta->offset;
499 unsigned pos = sizeof(header) - 1;
500 header[pos] = ofs & 127;
501 while (ofs >>= 7)
502 header[--pos] = 128 | (--ofs & 127);
503 sha1write(f, header + pos, sizeof(header) - pos);
504 hdrlen += sizeof(header) - pos;
505 } else if (obj_type == OBJ_REF_DELTA) {
506 sha1write(f, entry->delta->sha1, 20);
507 hdrlen += 20;
508 }
510 offset = entry->in_pack_offset + entry->in_pack_header_size;
511 datalen = find_packed_object_size(p, entry->in_pack_offset)
512 - entry->in_pack_header_size;
513 if (!pack_to_stdout && check_pack_inflate(p, &w_curs,
514 offset, datalen, entry->size))
515 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
516 copy_pack_data(f, p, &w_curs, offset, datalen);
517 unuse_pack(&w_curs);
518 reused++;
519 }
520 if (entry->delta)
521 written_delta++;
522 written++;
523 if (!pack_to_stdout)
524 entry->crc32 = crc32_end(f);
525 return hdrlen + datalen;
526 }
528 static off_t write_one(struct sha1file *f,
529 struct object_entry *e,
530 off_t offset)
531 {
532 unsigned long size;
534 /* offset is non zero if object is written already. */
535 if (e->offset || e->preferred_base)
536 return offset;
538 /* if we are deltified, write out base object first. */
539 if (e->delta)
540 offset = write_one(f, e->delta, offset);
542 e->offset = offset;
543 size = write_object(f, e);
545 /* make sure off_t is sufficiently large not to wrap */
546 if (offset > offset + size)
547 die("pack too large for current definition of off_t");
548 return offset + size;
549 }
551 static off_t write_pack_file(void)
552 {
553 uint32_t i;
554 struct sha1file *f;
555 off_t offset, last_obj_offset = 0;
556 struct pack_header hdr;
557 unsigned last_percent = 999;
558 int do_progress = progress;
560 if (!base_name) {
561 f = sha1fd(1, "<stdout>");
562 do_progress >>= 1;
563 }
564 else
565 f = sha1create("%s-%s.%s", base_name,
566 sha1_to_hex(object_list_sha1), "pack");
567 if (do_progress)
568 fprintf(stderr, "Writing %u objects.\n", nr_result);
570 hdr.hdr_signature = htonl(PACK_SIGNATURE);
571 hdr.hdr_version = htonl(PACK_VERSION);
572 hdr.hdr_entries = htonl(nr_result);
573 sha1write(f, &hdr, sizeof(hdr));
574 offset = sizeof(hdr);
575 if (!nr_result)
576 goto done;
577 for (i = 0; i < nr_objects; i++) {
578 last_obj_offset = offset;
579 offset = write_one(f, objects + i, offset);
580 if (do_progress) {
581 unsigned percent = written * 100 / nr_result;
582 if (progress_update || percent != last_percent) {
583 fprintf(stderr, "%4u%% (%u/%u) done\r",
584 percent, written, nr_result);
585 progress_update = 0;
586 last_percent = percent;
587 }
588 }
589 }
590 if (do_progress)
591 fputc('\n', stderr);
592 done:
593 if (written != nr_result)
594 die("wrote %u objects while expecting %u", written, nr_result);
595 sha1close(f, pack_file_sha1, 1);
597 return last_obj_offset;
598 }
600 static void write_index_file(off_t last_obj_offset)
601 {
602 uint32_t i;
603 struct sha1file *f = sha1create("%s-%s.%s", base_name,
604 sha1_to_hex(object_list_sha1), "idx");
605 struct object_entry **list = sorted_by_sha;
606 struct object_entry **last = list + nr_result;
607 uint32_t array[256];
608 uint32_t index_version;
610 /* if last object's offset is >= 2^31 we should use index V2 */
611 index_version = (last_obj_offset >> 31) ? 2 : 1;
613 /* index versions 2 and above need a header */
614 if (index_version >= 2) {
615 struct pack_idx_header hdr;
616 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
617 hdr.idx_version = htonl(index_version);
618 sha1write(f, &hdr, sizeof(hdr));
619 }
621 /*
622 * Write the first-level table (the list is sorted,
623 * but we use a 256-entry lookup to be able to avoid
624 * having to do eight extra binary search iterations).
625 */
626 for (i = 0; i < 256; i++) {
627 struct object_entry **next = list;
628 while (next < last) {
629 struct object_entry *entry = *next;
630 if (entry->sha1[0] != i)
631 break;
632 next++;
633 }
634 array[i] = htonl(next - sorted_by_sha);
635 list = next;
636 }
637 sha1write(f, array, 256 * 4);
639 /*
640 * Write the actual SHA1 entries..
641 */
642 list = sorted_by_sha;
643 for (i = 0; i < nr_result; i++) {
644 struct object_entry *entry = *list++;
645 if (index_version < 2) {
646 uint32_t offset = htonl(entry->offset);
647 sha1write(f, &offset, 4);
648 }
649 sha1write(f, entry->sha1, 20);
650 }
652 if (index_version >= 2) {
653 unsigned int nr_large_offset = 0;
655 /* write the crc32 table */
656 list = sorted_by_sha;
657 for (i = 0; i < nr_objects; i++) {
658 struct object_entry *entry = *list++;
659 uint32_t crc32_val = htonl(entry->crc32);
660 sha1write(f, &crc32_val, 4);
661 }
663 /* write the 32-bit offset table */
664 list = sorted_by_sha;
665 for (i = 0; i < nr_objects; i++) {
666 struct object_entry *entry = *list++;
667 uint32_t offset = (entry->offset <= 0x7fffffff) ?
668 entry->offset : (0x80000000 | nr_large_offset++);
669 offset = htonl(offset);
670 sha1write(f, &offset, 4);
671 }
673 /* write the large offset table */
674 list = sorted_by_sha;
675 while (nr_large_offset) {
676 struct object_entry *entry = *list++;
677 uint64_t offset = entry->offset;
678 if (offset > 0x7fffffff) {
679 uint32_t split[2];
680 split[0] = htonl(offset >> 32);
681 split[1] = htonl(offset & 0xffffffff);
682 sha1write(f, split, 8);
683 nr_large_offset--;
684 }
685 }
686 }
688 sha1write(f, pack_file_sha1, 20);
689 sha1close(f, NULL, 1);
690 }
692 static int locate_object_entry_hash(const unsigned char *sha1)
693 {
694 int i;
695 unsigned int ui;
696 memcpy(&ui, sha1, sizeof(unsigned int));
697 i = ui % object_ix_hashsz;
698 while (0 < object_ix[i]) {
699 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
700 return i;
701 if (++i == object_ix_hashsz)
702 i = 0;
703 }
704 return -1 - i;
705 }
707 static struct object_entry *locate_object_entry(const unsigned char *sha1)
708 {
709 int i;
711 if (!object_ix_hashsz)
712 return NULL;
714 i = locate_object_entry_hash(sha1);
715 if (0 <= i)
716 return &objects[object_ix[i]-1];
717 return NULL;
718 }
720 static void rehash_objects(void)
721 {
722 uint32_t i;
723 struct object_entry *oe;
725 object_ix_hashsz = nr_objects * 3;
726 if (object_ix_hashsz < 1024)
727 object_ix_hashsz = 1024;
728 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
729 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
730 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
731 int ix = locate_object_entry_hash(oe->sha1);
732 if (0 <= ix)
733 continue;
734 ix = -1 - ix;
735 object_ix[ix] = i + 1;
736 }
737 }
739 static unsigned name_hash(const char *name)
740 {
741 unsigned char c;
742 unsigned hash = 0;
744 /*
745 * This effectively just creates a sortable number from the
746 * last sixteen non-whitespace characters. Last characters
747 * count "most", so things that end in ".c" sort together.
748 */
749 while ((c = *name++) != 0) {
750 if (isspace(c))
751 continue;
752 hash = (hash >> 2) + (c << 24);
753 }
754 return hash;
755 }
757 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
758 {
759 uint32_t idx = nr_objects;
760 struct object_entry *entry;
761 struct packed_git *p;
762 off_t found_offset = 0;
763 struct packed_git *found_pack = NULL;
764 int ix, status = 0;
766 if (!exclude) {
767 for (p = packed_git; p; p = p->next) {
768 off_t offset = find_pack_entry_one(sha1, p);
769 if (offset) {
770 if (incremental)
771 return 0;
772 if (local && !p->pack_local)
773 return 0;
774 if (!found_pack) {
775 found_offset = offset;
776 found_pack = p;
777 }
778 }
779 }
780 }
781 if ((entry = locate_object_entry(sha1)) != NULL)
782 goto already_added;
784 if (idx >= nr_alloc) {
785 nr_alloc = (idx + 1024) * 3 / 2;
786 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
787 }
788 entry = objects + idx;
789 nr_objects = idx + 1;
790 memset(entry, 0, sizeof(*entry));
791 hashcpy(entry->sha1, sha1);
792 entry->hash = hash;
794 if (object_ix_hashsz * 3 <= nr_objects * 4)
795 rehash_objects();
796 else {
797 ix = locate_object_entry_hash(entry->sha1);
798 if (0 <= ix)
799 die("internal error in object hashing.");
800 object_ix[-1 - ix] = idx + 1;
801 }
802 status = 1;
804 already_added:
805 if (progress_update) {
806 fprintf(stderr, "Counting objects...%u\r", nr_objects);
807 progress_update = 0;
808 }
809 if (exclude)
810 entry->preferred_base = 1;
811 else {
812 if (found_pack) {
813 entry->in_pack = found_pack;
814 entry->in_pack_offset = found_offset;
815 }
816 }
817 return status;
818 }
820 struct pbase_tree_cache {
821 unsigned char sha1[20];
822 int ref;
823 int temporary;
824 void *tree_data;
825 unsigned long tree_size;
826 };
828 static struct pbase_tree_cache *(pbase_tree_cache[256]);
829 static int pbase_tree_cache_ix(const unsigned char *sha1)
830 {
831 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
832 }
833 static int pbase_tree_cache_ix_incr(int ix)
834 {
835 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
836 }
838 static struct pbase_tree {
839 struct pbase_tree *next;
840 /* This is a phony "cache" entry; we are not
841 * going to evict it nor find it through _get()
842 * mechanism -- this is for the toplevel node that
843 * would almost always change with any commit.
844 */
845 struct pbase_tree_cache pcache;
846 } *pbase_tree;
848 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
849 {
850 struct pbase_tree_cache *ent, *nent;
851 void *data;
852 unsigned long size;
853 enum object_type type;
854 int neigh;
855 int my_ix = pbase_tree_cache_ix(sha1);
856 int available_ix = -1;
858 /* pbase-tree-cache acts as a limited hashtable.
859 * your object will be found at your index or within a few
860 * slots after that slot if it is cached.
861 */
862 for (neigh = 0; neigh < 8; neigh++) {
863 ent = pbase_tree_cache[my_ix];
864 if (ent && !hashcmp(ent->sha1, sha1)) {
865 ent->ref++;
866 return ent;
867 }
868 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
869 ((0 <= available_ix) &&
870 (!ent && pbase_tree_cache[available_ix])))
871 available_ix = my_ix;
872 if (!ent)
873 break;
874 my_ix = pbase_tree_cache_ix_incr(my_ix);
875 }
877 /* Did not find one. Either we got a bogus request or
878 * we need to read and perhaps cache.
879 */
880 data = read_sha1_file(sha1, &type, &size);
881 if (!data)
882 return NULL;
883 if (type != OBJ_TREE) {
884 free(data);
885 return NULL;
886 }
888 /* We need to either cache or return a throwaway copy */
890 if (available_ix < 0)
891 ent = NULL;
892 else {
893 ent = pbase_tree_cache[available_ix];
894 my_ix = available_ix;
895 }
897 if (!ent) {
898 nent = xmalloc(sizeof(*nent));
899 nent->temporary = (available_ix < 0);
900 }
901 else {
902 /* evict and reuse */
903 free(ent->tree_data);
904 nent = ent;
905 }
906 hashcpy(nent->sha1, sha1);
907 nent->tree_data = data;
908 nent->tree_size = size;
909 nent->ref = 1;
910 if (!nent->temporary)
911 pbase_tree_cache[my_ix] = nent;
912 return nent;
913 }
915 static void pbase_tree_put(struct pbase_tree_cache *cache)
916 {
917 if (!cache->temporary) {
918 cache->ref--;
919 return;
920 }
921 free(cache->tree_data);
922 free(cache);
923 }
925 static int name_cmp_len(const char *name)
926 {
927 int i;
928 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
929 ;
930 return i;
931 }
933 static void add_pbase_object(struct tree_desc *tree,
934 const char *name,
935 int cmplen,
936 const char *fullname)
937 {
938 struct name_entry entry;
940 while (tree_entry(tree,&entry)) {
941 unsigned long size;
942 enum object_type type;
944 if (tree_entry_len(entry.path, entry.sha1) != cmplen ||
945 memcmp(entry.path, name, cmplen) ||
946 !has_sha1_file(entry.sha1) ||
947 (type = sha1_object_info(entry.sha1, &size)) < 0)
948 continue;
949 if (name[cmplen] != '/') {
950 unsigned hash = name_hash(fullname);
951 add_object_entry(entry.sha1, hash, 1);
952 return;
953 }
954 if (type == OBJ_TREE) {
955 struct tree_desc sub;
956 struct pbase_tree_cache *tree;
957 const char *down = name+cmplen+1;
958 int downlen = name_cmp_len(down);
960 tree = pbase_tree_get(entry.sha1);
961 if (!tree)
962 return;
963 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
965 add_pbase_object(&sub, down, downlen, fullname);
966 pbase_tree_put(tree);
967 }
968 }
969 }
971 static unsigned *done_pbase_paths;
972 static int done_pbase_paths_num;
973 static int done_pbase_paths_alloc;
974 static int done_pbase_path_pos(unsigned hash)
975 {
976 int lo = 0;
977 int hi = done_pbase_paths_num;
978 while (lo < hi) {
979 int mi = (hi + lo) / 2;
980 if (done_pbase_paths[mi] == hash)
981 return mi;
982 if (done_pbase_paths[mi] < hash)
983 hi = mi;
984 else
985 lo = mi + 1;
986 }
987 return -lo-1;
988 }
990 static int check_pbase_path(unsigned hash)
991 {
992 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
993 if (0 <= pos)
994 return 1;
995 pos = -pos - 1;
996 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
997 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
998 done_pbase_paths = xrealloc(done_pbase_paths,
999 done_pbase_paths_alloc *
1000 sizeof(unsigned));
1001 }
1002 done_pbase_paths_num++;
1003 if (pos < done_pbase_paths_num)
1004 memmove(done_pbase_paths + pos + 1,
1005 done_pbase_paths + pos,
1006 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1007 done_pbase_paths[pos] = hash;
1008 return 0;
1009 }
1011 static void add_preferred_base_object(const char *name, unsigned hash)
1012 {
1013 struct pbase_tree *it;
1014 int cmplen = name_cmp_len(name);
1016 if (check_pbase_path(hash))
1017 return;
1019 for (it = pbase_tree; it; it = it->next) {
1020 if (cmplen == 0) {
1021 hash = name_hash("");
1022 add_object_entry(it->pcache.sha1, hash, 1);
1023 }
1024 else {
1025 struct tree_desc tree;
1026 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1027 add_pbase_object(&tree, name, cmplen, name);
1028 }
1029 }
1030 }
1032 static void add_preferred_base(unsigned char *sha1)
1033 {
1034 struct pbase_tree *it;
1035 void *data;
1036 unsigned long size;
1037 unsigned char tree_sha1[20];
1039 if (window <= num_preferred_base++)
1040 return;
1042 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1043 if (!data)
1044 return;
1046 for (it = pbase_tree; it; it = it->next) {
1047 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1048 free(data);
1049 return;
1050 }
1051 }
1053 it = xcalloc(1, sizeof(*it));
1054 it->next = pbase_tree;
1055 pbase_tree = it;
1057 hashcpy(it->pcache.sha1, tree_sha1);
1058 it->pcache.tree_data = data;
1059 it->pcache.tree_size = size;
1060 }
1062 static void check_object(struct object_entry *entry)
1063 {
1064 if (entry->in_pack && !entry->preferred_base) {
1065 struct packed_git *p = entry->in_pack;
1066 struct pack_window *w_curs = NULL;
1067 unsigned long size, used;
1068 unsigned int avail;
1069 unsigned char *buf;
1070 struct object_entry *base_entry = NULL;
1072 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1074 /* We want in_pack_type even if we do not reuse delta.
1075 * There is no point not reusing non-delta representations.
1076 */
1077 used = unpack_object_header_gently(buf, avail,
1078 &entry->in_pack_type, &size);
1080 /* Check if it is delta, and the base is also an object
1081 * we are going to pack. If so we will reuse the existing
1082 * delta.
1083 */
1084 if (!no_reuse_delta) {
1085 unsigned char c;
1086 const unsigned char *base_name;
1087 off_t ofs;
1088 unsigned long used_0;
1089 /* there is at least 20 bytes left in the pack */
1090 switch (entry->in_pack_type) {
1091 case OBJ_REF_DELTA:
1092 base_name = use_pack(p, &w_curs,
1093 entry->in_pack_offset + used, NULL);
1094 used += 20;
1095 break;
1096 case OBJ_OFS_DELTA:
1097 buf = use_pack(p, &w_curs,
1098 entry->in_pack_offset + used, NULL);
1099 used_0 = 0;
1100 c = buf[used_0++];
1101 ofs = c & 127;
1102 while (c & 128) {
1103 ofs += 1;
1104 if (!ofs || MSB(ofs, 7))
1105 die("delta base offset overflow in pack for %s",
1106 sha1_to_hex(entry->sha1));
1107 c = buf[used_0++];
1108 ofs = (ofs << 7) + (c & 127);
1109 }
1110 if (ofs >= entry->in_pack_offset)
1111 die("delta base offset out of bound for %s",
1112 sha1_to_hex(entry->sha1));
1113 ofs = entry->in_pack_offset - ofs;
1114 base_name = find_packed_object_name(p, ofs);
1115 used += used_0;
1116 break;
1117 default:
1118 base_name = NULL;
1119 }
1120 if (base_name)
1121 base_entry = locate_object_entry(base_name);
1122 }
1123 unuse_pack(&w_curs);
1124 entry->in_pack_header_size = used;
1126 if (base_entry) {
1128 /* Depth value does not matter - find_deltas()
1129 * will never consider reused delta as the
1130 * base object to deltify other objects
1131 * against, in order to avoid circular deltas.
1132 */
1134 /* uncompressed size of the delta data */
1135 entry->size = size;
1136 entry->delta = base_entry;
1137 entry->type = entry->in_pack_type;
1139 entry->delta_sibling = base_entry->delta_child;
1140 base_entry->delta_child = entry;
1142 return;
1143 }
1144 /* Otherwise we would do the usual */
1145 }
1147 entry->type = sha1_object_info(entry->sha1, &entry->size);
1148 if (entry->type < 0)
1149 die("unable to get type of object %s",
1150 sha1_to_hex(entry->sha1));
1151 }
1153 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1154 {
1155 struct object_entry *child = me->delta_child;
1156 unsigned int m = n;
1157 while (child) {
1158 unsigned int c = check_delta_limit(child, n + 1);
1159 if (m < c)
1160 m = c;
1161 child = child->delta_sibling;
1162 }
1163 return m;
1164 }
1166 static void get_object_details(void)
1167 {
1168 uint32_t i;
1169 struct object_entry *entry;
1171 prepare_pack_ix();
1172 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1173 check_object(entry);
1175 if (nr_objects == nr_result) {
1176 /*
1177 * Depth of objects that depend on the entry -- this
1178 * is subtracted from depth-max to break too deep
1179 * delta chain because of delta data reusing.
1180 * However, we loosen this restriction when we know we
1181 * are creating a thin pack -- it will have to be
1182 * expanded on the other end anyway, so do not
1183 * artificially cut the delta chain and let it go as
1184 * deep as it wants.
1185 */
1186 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1187 if (!entry->delta && entry->delta_child)
1188 entry->delta_limit =
1189 check_delta_limit(entry, 1);
1190 }
1191 }
1193 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1195 static entry_sort_t current_sort;
1197 static int sort_comparator(const void *_a, const void *_b)
1198 {
1199 struct object_entry *a = *(struct object_entry **)_a;
1200 struct object_entry *b = *(struct object_entry **)_b;
1201 return current_sort(a,b);
1202 }
1204 static struct object_entry **create_sorted_list(entry_sort_t sort)
1205 {
1206 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1207 uint32_t i;
1209 for (i = 0; i < nr_objects; i++)
1210 list[i] = objects + i;
1211 current_sort = sort;
1212 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1213 return list;
1214 }
1216 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1217 {
1218 return hashcmp(a->sha1, b->sha1);
1219 }
1221 static struct object_entry **create_final_object_list(void)
1222 {
1223 struct object_entry **list;
1224 uint32_t i, j;
1226 for (i = nr_result = 0; i < nr_objects; i++)
1227 if (!objects[i].preferred_base)
1228 nr_result++;
1229 list = xmalloc(nr_result * sizeof(struct object_entry *));
1230 for (i = j = 0; i < nr_objects; i++) {
1231 if (!objects[i].preferred_base)
1232 list[j++] = objects + i;
1233 }
1234 current_sort = sha1_sort;
1235 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1236 return list;
1237 }
1239 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1240 {
1241 if (a->type < b->type)
1242 return -1;
1243 if (a->type > b->type)
1244 return 1;
1245 if (a->hash < b->hash)
1246 return -1;
1247 if (a->hash > b->hash)
1248 return 1;
1249 if (a->preferred_base < b->preferred_base)
1250 return -1;
1251 if (a->preferred_base > b->preferred_base)
1252 return 1;
1253 if (a->size < b->size)
1254 return -1;
1255 if (a->size > b->size)
1256 return 1;
1257 return a < b ? -1 : (a > b);
1258 }
1260 struct unpacked {
1261 struct object_entry *entry;
1262 void *data;
1263 struct delta_index *index;
1264 };
1266 /*
1267 * We search for deltas _backwards_ in a list sorted by type and
1268 * by size, so that we see progressively smaller and smaller files.
1269 * That's because we prefer deltas to be from the bigger file
1270 * to the smaller - deletes are potentially cheaper, but perhaps
1271 * more importantly, the bigger file is likely the more recent
1272 * one.
1273 */
1274 static int try_delta(struct unpacked *trg, struct unpacked *src,
1275 unsigned max_depth)
1276 {
1277 struct object_entry *trg_entry = trg->entry;
1278 struct object_entry *src_entry = src->entry;
1279 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1280 enum object_type type;
1281 void *delta_buf;
1283 /* Don't bother doing diffs between different types */
1284 if (trg_entry->type != src_entry->type)
1285 return -1;
1287 /* We do not compute delta to *create* objects we are not
1288 * going to pack.
1289 */
1290 if (trg_entry->preferred_base)
1291 return -1;
1293 /*
1294 * We do not bother to try a delta that we discarded
1295 * on an earlier try, but only when reusing delta data.
1296 */
1297 if (!no_reuse_delta && trg_entry->in_pack &&
1298 trg_entry->in_pack == src_entry->in_pack &&
1299 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1300 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1301 return 0;
1303 /*
1304 * If the current object is at pack edge, take the depth the
1305 * objects that depend on the current object into account --
1306 * otherwise they would become too deep.
1307 */
1308 if (trg_entry->delta_child) {
1309 if (max_depth <= trg_entry->delta_limit)
1310 return 0;
1311 max_depth -= trg_entry->delta_limit;
1312 }
1313 if (src_entry->depth >= max_depth)
1314 return 0;
1316 /* Now some size filtering heuristics. */
1317 trg_size = trg_entry->size;
1318 max_size = trg_size/2 - 20;
1319 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1320 if (max_size == 0)
1321 return 0;
1322 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1323 max_size = trg_entry->delta_size-1;
1324 src_size = src_entry->size;
1325 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1326 if (sizediff >= max_size)
1327 return 0;
1329 /* Load data if not already done */
1330 if (!trg->data) {
1331 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1332 if (sz != trg_size)
1333 die("object %s inconsistent object length (%lu vs %lu)",
1334 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1335 }
1336 if (!src->data) {
1337 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1338 if (sz != src_size)
1339 die("object %s inconsistent object length (%lu vs %lu)",
1340 sha1_to_hex(src_entry->sha1), sz, src_size);
1341 }
1342 if (!src->index) {
1343 src->index = create_delta_index(src->data, src_size);
1344 if (!src->index)
1345 die("out of memory");
1346 }
1348 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1349 if (!delta_buf)
1350 return 0;
1352 trg_entry->delta = src_entry;
1353 trg_entry->delta_size = delta_size;
1354 trg_entry->depth = src_entry->depth + 1;
1355 free(delta_buf);
1356 return 1;
1357 }
1359 static void progress_interval(int signum)
1360 {
1361 progress_update = 1;
1362 }
1364 static void find_deltas(struct object_entry **list, int window, int depth)
1365 {
1366 uint32_t i = nr_objects, idx = 0, processed = 0;
1367 unsigned int array_size = window * sizeof(struct unpacked);
1368 struct unpacked *array;
1369 unsigned last_percent = 999;
1371 if (!nr_objects)
1372 return;
1373 array = xmalloc(array_size);
1374 memset(array, 0, array_size);
1375 if (progress)
1376 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
1378 do {
1379 struct object_entry *entry = list[--i];
1380 struct unpacked *n = array + idx;
1381 int j;
1383 if (!entry->preferred_base)
1384 processed++;
1386 if (progress) {
1387 unsigned percent = processed * 100 / nr_result;
1388 if (percent != last_percent || progress_update) {
1389 fprintf(stderr, "%4u%% (%u/%u) done\r",
1390 percent, processed, nr_result);
1391 progress_update = 0;
1392 last_percent = percent;
1393 }
1394 }
1396 if (entry->delta)
1397 /* This happens if we decided to reuse existing
1398 * delta from a pack. "!no_reuse_delta &&" is implied.
1399 */
1400 continue;
1402 if (entry->size < 50)
1403 continue;
1404 free_delta_index(n->index);
1405 n->index = NULL;
1406 free(n->data);
1407 n->data = NULL;
1408 n->entry = entry;
1410 j = window;
1411 while (--j > 0) {
1412 uint32_t other_idx = idx + j;
1413 struct unpacked *m;
1414 if (other_idx >= window)
1415 other_idx -= window;
1416 m = array + other_idx;
1417 if (!m->entry)
1418 break;
1419 if (try_delta(n, m, depth) < 0)
1420 break;
1421 }
1422 /* if we made n a delta, and if n is already at max
1423 * depth, leaving it in the window is pointless. we
1424 * should evict it first.
1425 */
1426 if (entry->delta && depth <= entry->depth)
1427 continue;
1429 idx++;
1430 if (idx >= window)
1431 idx = 0;
1432 } while (i > 0);
1434 if (progress)
1435 fputc('\n', stderr);
1437 for (i = 0; i < window; ++i) {
1438 free_delta_index(array[i].index);
1439 free(array[i].data);
1440 }
1441 free(array);
1442 }
1444 static void prepare_pack(int window, int depth)
1445 {
1446 get_object_details();
1447 sorted_by_type = create_sorted_list(type_size_sort);
1448 if (window && depth)
1449 find_deltas(sorted_by_type, window+1, depth);
1450 }
1452 static int reuse_cached_pack(unsigned char *sha1)
1453 {
1454 static const char cache[] = "pack-cache/pack-%s.%s";
1455 char *cached_pack, *cached_idx;
1456 int ifd, ofd, ifd_ix = -1;
1458 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1459 ifd = open(cached_pack, O_RDONLY);
1460 if (ifd < 0)
1461 return 0;
1463 if (!pack_to_stdout) {
1464 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1465 ifd_ix = open(cached_idx, O_RDONLY);
1466 if (ifd_ix < 0) {
1467 close(ifd);
1468 return 0;
1469 }
1470 }
1472 if (progress)
1473 fprintf(stderr, "Reusing %u objects pack %s\n", nr_objects,
1474 sha1_to_hex(sha1));
1476 if (pack_to_stdout) {
1477 if (copy_fd(ifd, 1))
1478 exit(1);
1479 close(ifd);
1480 }
1481 else {
1482 char name[PATH_MAX];
1483 snprintf(name, sizeof(name),
1484 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1485 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1486 if (ofd < 0)
1487 die("unable to open %s (%s)", name, strerror(errno));
1488 if (copy_fd(ifd, ofd))
1489 exit(1);
1490 close(ifd);
1492 snprintf(name, sizeof(name),
1493 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1494 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1495 if (ofd < 0)
1496 die("unable to open %s (%s)", name, strerror(errno));
1497 if (copy_fd(ifd_ix, ofd))
1498 exit(1);
1499 close(ifd_ix);
1500 puts(sha1_to_hex(sha1));
1501 }
1503 return 1;
1504 }
1506 static void setup_progress_signal(void)
1507 {
1508 struct sigaction sa;
1509 struct itimerval v;
1511 memset(&sa, 0, sizeof(sa));
1512 sa.sa_handler = progress_interval;
1513 sigemptyset(&sa.sa_mask);
1514 sa.sa_flags = SA_RESTART;
1515 sigaction(SIGALRM, &sa, NULL);
1517 v.it_interval.tv_sec = 1;
1518 v.it_interval.tv_usec = 0;
1519 v.it_value = v.it_interval;
1520 setitimer(ITIMER_REAL, &v, NULL);
1521 }
1523 static int git_pack_config(const char *k, const char *v)
1524 {
1525 if(!strcmp(k, "pack.window")) {
1526 window = git_config_int(k, v);
1527 return 0;
1528 }
1529 return git_default_config(k, v);
1530 }
1532 static void read_object_list_from_stdin(void)
1533 {
1534 char line[40 + 1 + PATH_MAX + 2];
1535 unsigned char sha1[20];
1536 unsigned hash;
1538 for (;;) {
1539 if (!fgets(line, sizeof(line), stdin)) {
1540 if (feof(stdin))
1541 break;
1542 if (!ferror(stdin))
1543 die("fgets returned NULL, not EOF, not error!");
1544 if (errno != EINTR)
1545 die("fgets: %s", strerror(errno));
1546 clearerr(stdin);
1547 continue;
1548 }
1549 if (line[0] == '-') {
1550 if (get_sha1_hex(line+1, sha1))
1551 die("expected edge sha1, got garbage:\n %s",
1552 line);
1553 add_preferred_base(sha1);
1554 continue;
1555 }
1556 if (get_sha1_hex(line, sha1))
1557 die("expected sha1, got garbage:\n %s", line);
1559 hash = name_hash(line+41);
1560 add_preferred_base_object(line+41, hash);
1561 add_object_entry(sha1, hash, 0);
1562 }
1563 }
1565 static void show_commit(struct commit *commit)
1566 {
1567 unsigned hash = name_hash("");
1568 add_preferred_base_object("", hash);
1569 add_object_entry(commit->object.sha1, hash, 0);
1570 }
1572 static void show_object(struct object_array_entry *p)
1573 {
1574 unsigned hash = name_hash(p->name);
1575 add_preferred_base_object(p->name, hash);
1576 add_object_entry(p->item->sha1, hash, 0);
1577 }
1579 static void show_edge(struct commit *commit)
1580 {
1581 add_preferred_base(commit->object.sha1);
1582 }
1584 static void get_object_list(int ac, const char **av)
1585 {
1586 struct rev_info revs;
1587 char line[1000];
1588 int flags = 0;
1590 init_revisions(&revs, NULL);
1591 save_commit_buffer = 0;
1592 track_object_refs = 0;
1593 setup_revisions(ac, av, &revs, NULL);
1595 while (fgets(line, sizeof(line), stdin) != NULL) {
1596 int len = strlen(line);
1597 if (line[len - 1] == '\n')
1598 line[--len] = 0;
1599 if (!len)
1600 break;
1601 if (*line == '-') {
1602 if (!strcmp(line, "--not")) {
1603 flags ^= UNINTERESTING;
1604 continue;
1605 }
1606 die("not a rev '%s'", line);
1607 }
1608 if (handle_revision_arg(line, &revs, flags, 1))
1609 die("bad revision '%s'", line);
1610 }
1612 prepare_revision_walk(&revs);
1613 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1614 traverse_commit_list(&revs, show_commit, show_object);
1615 }
1617 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1618 {
1619 SHA_CTX ctx;
1620 int depth = 10;
1621 struct object_entry **list;
1622 int use_internal_rev_list = 0;
1623 int thin = 0;
1624 uint32_t i;
1625 const char **rp_av;
1626 int rp_ac_alloc = 64;
1627 int rp_ac;
1629 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1631 rp_av[0] = "pack-objects";
1632 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1633 rp_ac = 2;
1635 git_config(git_pack_config);
1637 progress = isatty(2);
1638 for (i = 1; i < argc; i++) {
1639 const char *arg = argv[i];
1641 if (*arg != '-')
1642 break;
1644 if (!strcmp("--non-empty", arg)) {
1645 non_empty = 1;
1646 continue;
1647 }
1648 if (!strcmp("--local", arg)) {
1649 local = 1;
1650 continue;
1651 }
1652 if (!strcmp("--incremental", arg)) {
1653 incremental = 1;
1654 continue;
1655 }
1656 if (!prefixcmp(arg, "--window=")) {
1657 char *end;
1658 window = strtoul(arg+9, &end, 0);
1659 if (!arg[9] || *end)
1660 usage(pack_usage);
1661 continue;
1662 }
1663 if (!prefixcmp(arg, "--depth=")) {
1664 char *end;
1665 depth = strtoul(arg+8, &end, 0);
1666 if (!arg[8] || *end)
1667 usage(pack_usage);
1668 continue;
1669 }
1670 if (!strcmp("--progress", arg)) {
1671 progress = 1;
1672 continue;
1673 }
1674 if (!strcmp("--all-progress", arg)) {
1675 progress = 2;
1676 continue;
1677 }
1678 if (!strcmp("-q", arg)) {
1679 progress = 0;
1680 continue;
1681 }
1682 if (!strcmp("--no-reuse-delta", arg)) {
1683 no_reuse_delta = 1;
1684 continue;
1685 }
1686 if (!strcmp("--delta-base-offset", arg)) {
1687 allow_ofs_delta = 1;
1688 continue;
1689 }
1690 if (!strcmp("--stdout", arg)) {
1691 pack_to_stdout = 1;
1692 continue;
1693 }
1694 if (!strcmp("--revs", arg)) {
1695 use_internal_rev_list = 1;
1696 continue;
1697 }
1698 if (!strcmp("--unpacked", arg) ||
1699 !prefixcmp(arg, "--unpacked=") ||
1700 !strcmp("--reflog", arg) ||
1701 !strcmp("--all", arg)) {
1702 use_internal_rev_list = 1;
1703 if (rp_ac >= rp_ac_alloc - 1) {
1704 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1705 rp_av = xrealloc(rp_av,
1706 rp_ac_alloc * sizeof(*rp_av));
1707 }
1708 rp_av[rp_ac++] = arg;
1709 continue;
1710 }
1711 if (!strcmp("--thin", arg)) {
1712 use_internal_rev_list = 1;
1713 thin = 1;
1714 rp_av[1] = "--objects-edge";
1715 continue;
1716 }
1717 usage(pack_usage);
1718 }
1720 /* Traditionally "pack-objects [options] base extra" failed;
1721 * we would however want to take refs parameter that would
1722 * have been given to upstream rev-list ourselves, which means
1723 * we somehow want to say what the base name is. So the
1724 * syntax would be:
1725 *
1726 * pack-objects [options] base <refs...>
1727 *
1728 * in other words, we would treat the first non-option as the
1729 * base_name and send everything else to the internal revision
1730 * walker.
1731 */
1733 if (!pack_to_stdout)
1734 base_name = argv[i++];
1736 if (pack_to_stdout != !base_name)
1737 usage(pack_usage);
1739 if (!pack_to_stdout && thin)
1740 die("--thin cannot be used to build an indexable pack.");
1742 prepare_packed_git();
1744 if (progress) {
1745 fprintf(stderr, "Generating pack...\n");
1746 setup_progress_signal();
1747 }
1749 if (!use_internal_rev_list)
1750 read_object_list_from_stdin();
1751 else {
1752 rp_av[rp_ac] = NULL;
1753 get_object_list(rp_ac, rp_av);
1754 }
1756 if (progress)
1757 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1758 sorted_by_sha = create_final_object_list();
1759 if (non_empty && !nr_result)
1760 return 0;
1762 SHA1_Init(&ctx);
1763 list = sorted_by_sha;
1764 for (i = 0; i < nr_result; i++) {
1765 struct object_entry *entry = *list++;
1766 SHA1_Update(&ctx, entry->sha1, 20);
1767 }
1768 SHA1_Final(object_list_sha1, &ctx);
1769 if (progress && (nr_objects != nr_result))
1770 fprintf(stderr, "Result has %u objects.\n", nr_result);
1772 if (reuse_cached_pack(object_list_sha1))
1773 ;
1774 else {
1775 off_t last_obj_offset;
1776 if (nr_result)
1777 prepare_pack(window, depth);
1778 if (progress == 1 && pack_to_stdout) {
1779 /* the other end usually displays progress itself */
1780 struct itimerval v = {{0,},};
1781 setitimer(ITIMER_REAL, &v, NULL);
1782 signal(SIGALRM, SIG_IGN );
1783 progress_update = 0;
1784 }
1785 last_obj_offset = write_pack_file();
1786 if (!pack_to_stdout) {
1787 write_index_file(last_obj_offset);
1788 puts(sha1_to_hex(object_list_sha1));
1789 }
1790 }
1791 if (progress)
1792 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1793 written, written_delta, reused, reused_delta);
1794 return 0;
1795 }