![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
a94ec789716969814e4f406b02ed88e4cdb2ce14
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 "csum-file.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "revision.h"
15 #include "list-objects.h"
16 #include "progress.h"
18 static const char pack_usage[] = "\
19 git-pack-objects [{ -q | --progress | --all-progress }] \n\
20 [--max-pack-size=N] [--local] [--incremental] \n\
21 [--window=N] [--window-memory=N] [--depth=N] \n\
22 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
23 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
24 [--stdout | base-name] [<ref-list | <object-list]";
26 struct object_entry {
27 struct pack_idx_entry idx;
28 unsigned long size; /* uncompressed size */
30 unsigned int hash; /* name hint hash */
31 unsigned int depth; /* delta depth */
32 struct packed_git *in_pack; /* already in pack */
33 off_t in_pack_offset;
34 struct object_entry *delta; /* delta base object */
35 struct object_entry *delta_child; /* deltified objects who bases me */
36 struct object_entry *delta_sibling; /* other deltified objects who
37 * uses the same base as me
38 */
39 void *delta_data; /* cached delta (uncompressed) */
40 unsigned long delta_size; /* delta data size (uncompressed) */
41 enum object_type type;
42 enum object_type in_pack_type; /* could be delta */
43 unsigned char in_pack_header_size;
44 unsigned char preferred_base; /* we do not pack this, but is available
45 * to be used as the base object to delta
46 * objects against.
47 */
48 unsigned char no_try_delta;
49 };
51 /*
52 * Objects we are going to pack are collected in objects array (dynamically
53 * expanded). nr_objects & nr_alloc controls this array. They are stored
54 * in the order we see -- typically rev-list --objects order that gives us
55 * nice "minimum seek" order.
56 */
57 static struct object_entry *objects;
58 static struct object_entry **written_list;
59 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
61 static int non_empty;
62 static int no_reuse_delta, no_reuse_object;
63 static int local;
64 static int incremental;
65 static int allow_ofs_delta;
66 static const char *pack_tmp_name, *idx_tmp_name;
67 static char tmpname[PATH_MAX];
68 static const char *base_name;
69 static int progress = 1;
70 static int window = 10;
71 static uint32_t pack_size_limit;
72 static int depth = 50;
73 static int pack_to_stdout;
74 static int num_preferred_base;
75 static struct progress progress_state;
76 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
77 static int pack_compression_seen;
79 static unsigned long delta_cache_size = 0;
80 static unsigned long max_delta_cache_size = 0;
81 static unsigned long cache_max_small_delta_size = 1000;
83 static unsigned long window_memory_usage = 0;
84 static unsigned long window_memory_limit = 0;
86 /*
87 * The object names in objects array are hashed with this hashtable,
88 * to help looking up the entry by object name.
89 * This hashtable is built after all the objects are seen.
90 */
91 static int *object_ix;
92 static int object_ix_hashsz;
94 /*
95 * Pack index for existing packs give us easy access to the offsets into
96 * corresponding pack file where each object's data starts, but the entries
97 * do not store the size of the compressed representation (uncompressed
98 * size is easily available by examining the pack entry header). It is
99 * also rather expensive to find the sha1 for an object given its offset.
100 *
101 * We build a hashtable of existing packs (pack_revindex), and keep reverse
102 * index here -- pack index file is sorted by object name mapping to offset;
103 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
104 * ordered by offset, so if you know the offset of an object, next offset
105 * is where its packed representation ends and the index_nr can be used to
106 * get the object sha1 from the main index.
107 */
108 struct revindex_entry {
109 off_t offset;
110 unsigned int nr;
111 };
112 struct pack_revindex {
113 struct packed_git *p;
114 struct revindex_entry *revindex;
115 };
116 static struct pack_revindex *pack_revindex;
117 static int pack_revindex_hashsz;
119 /*
120 * stats
121 */
122 static uint32_t written, written_delta;
123 static uint32_t reused, reused_delta;
125 static int pack_revindex_ix(struct packed_git *p)
126 {
127 unsigned long ui = (unsigned long)p;
128 int i;
130 ui = ui ^ (ui >> 16); /* defeat structure alignment */
131 i = (int)(ui % pack_revindex_hashsz);
132 while (pack_revindex[i].p) {
133 if (pack_revindex[i].p == p)
134 return i;
135 if (++i == pack_revindex_hashsz)
136 i = 0;
137 }
138 return -1 - i;
139 }
141 static void prepare_pack_ix(void)
142 {
143 int num;
144 struct packed_git *p;
145 for (num = 0, p = packed_git; p; p = p->next)
146 num++;
147 if (!num)
148 return;
149 pack_revindex_hashsz = num * 11;
150 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
151 for (p = packed_git; p; p = p->next) {
152 num = pack_revindex_ix(p);
153 num = - 1 - num;
154 pack_revindex[num].p = p;
155 }
156 /* revindex elements are lazily initialized */
157 }
159 static int cmp_offset(const void *a_, const void *b_)
160 {
161 const struct revindex_entry *a = a_;
162 const struct revindex_entry *b = b_;
163 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
164 }
166 /*
167 * Ordered list of offsets of objects in the pack.
168 */
169 static void prepare_pack_revindex(struct pack_revindex *rix)
170 {
171 struct packed_git *p = rix->p;
172 int num_ent = p->num_objects;
173 int i;
174 const char *index = p->index_data;
176 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
177 index += 4 * 256;
179 if (p->index_version > 1) {
180 const uint32_t *off_32 =
181 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
182 const uint32_t *off_64 = off_32 + p->num_objects;
183 for (i = 0; i < num_ent; i++) {
184 uint32_t off = ntohl(*off_32++);
185 if (!(off & 0x80000000)) {
186 rix->revindex[i].offset = off;
187 } else {
188 rix->revindex[i].offset =
189 ((uint64_t)ntohl(*off_64++)) << 32;
190 rix->revindex[i].offset |=
191 ntohl(*off_64++);
192 }
193 rix->revindex[i].nr = i;
194 }
195 } else {
196 for (i = 0; i < num_ent; i++) {
197 uint32_t hl = *((uint32_t *)(index + 24 * i));
198 rix->revindex[i].offset = ntohl(hl);
199 rix->revindex[i].nr = i;
200 }
201 }
203 /* This knows the pack format -- the 20-byte trailer
204 * follows immediately after the last object data.
205 */
206 rix->revindex[num_ent].offset = p->pack_size - 20;
207 rix->revindex[num_ent].nr = -1;
208 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
209 }
211 static struct revindex_entry * find_packed_object(struct packed_git *p,
212 off_t ofs)
213 {
214 int num;
215 int lo, hi;
216 struct pack_revindex *rix;
217 struct revindex_entry *revindex;
218 num = pack_revindex_ix(p);
219 if (num < 0)
220 die("internal error: pack revindex uninitialized");
221 rix = &pack_revindex[num];
222 if (!rix->revindex)
223 prepare_pack_revindex(rix);
224 revindex = rix->revindex;
225 lo = 0;
226 hi = p->num_objects + 1;
227 do {
228 int mi = (lo + hi) / 2;
229 if (revindex[mi].offset == ofs) {
230 return revindex + mi;
231 }
232 else if (ofs < revindex[mi].offset)
233 hi = mi;
234 else
235 lo = mi + 1;
236 } while (lo < hi);
237 die("internal error: pack revindex corrupt");
238 }
240 static const unsigned char *find_packed_object_name(struct packed_git *p,
241 off_t ofs)
242 {
243 struct revindex_entry *entry = find_packed_object(p, ofs);
244 return nth_packed_object_sha1(p, entry->nr);
245 }
247 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
248 {
249 unsigned long othersize, delta_size;
250 enum object_type type;
251 void *otherbuf = read_sha1_file(entry->delta->idx.sha1, &type, &othersize);
252 void *delta_buf;
254 if (!otherbuf)
255 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
256 delta_buf = diff_delta(otherbuf, othersize,
257 buf, size, &delta_size, 0);
258 if (!delta_buf || delta_size != entry->delta_size)
259 die("delta size changed");
260 free(buf);
261 free(otherbuf);
262 return delta_buf;
263 }
265 /*
266 * The per-object header is a pretty dense thing, which is
267 * - first byte: low four bits are "size", then three bits of "type",
268 * and the high bit is "size continues".
269 * - each byte afterwards: low seven bits are size continuation,
270 * with the high bit being "size continues"
271 */
272 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
273 {
274 int n = 1;
275 unsigned char c;
277 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
278 die("bad type %d", type);
280 c = (type << 4) | (size & 15);
281 size >>= 4;
282 while (size) {
283 *hdr++ = c | 0x80;
284 c = size & 0x7f;
285 size >>= 7;
286 n++;
287 }
288 *hdr = c;
289 return n;
290 }
292 /*
293 * we are going to reuse the existing object data as is. make
294 * sure it is not corrupt.
295 */
296 static int check_pack_inflate(struct packed_git *p,
297 struct pack_window **w_curs,
298 off_t offset,
299 off_t len,
300 unsigned long expect)
301 {
302 z_stream stream;
303 unsigned char fakebuf[4096], *in;
304 int st;
306 memset(&stream, 0, sizeof(stream));
307 inflateInit(&stream);
308 do {
309 in = use_pack(p, w_curs, offset, &stream.avail_in);
310 stream.next_in = in;
311 stream.next_out = fakebuf;
312 stream.avail_out = sizeof(fakebuf);
313 st = inflate(&stream, Z_FINISH);
314 offset += stream.next_in - in;
315 } while (st == Z_OK || st == Z_BUF_ERROR);
316 inflateEnd(&stream);
317 return (st == Z_STREAM_END &&
318 stream.total_out == expect &&
319 stream.total_in == len) ? 0 : -1;
320 }
322 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
323 off_t offset, off_t len, unsigned int nr)
324 {
325 const uint32_t *index_crc;
326 uint32_t data_crc = crc32(0, Z_NULL, 0);
328 do {
329 unsigned int avail;
330 void *data = use_pack(p, w_curs, offset, &avail);
331 if (avail > len)
332 avail = len;
333 data_crc = crc32(data_crc, data, avail);
334 offset += avail;
335 len -= avail;
336 } while (len);
338 index_crc = p->index_data;
339 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
341 return data_crc != ntohl(*index_crc);
342 }
344 static void copy_pack_data(struct sha1file *f,
345 struct packed_git *p,
346 struct pack_window **w_curs,
347 off_t offset,
348 off_t len)
349 {
350 unsigned char *in;
351 unsigned int avail;
353 while (len) {
354 in = use_pack(p, w_curs, offset, &avail);
355 if (avail > len)
356 avail = (unsigned int)len;
357 sha1write(f, in, avail);
358 offset += avail;
359 len -= avail;
360 }
361 }
363 static unsigned long write_object(struct sha1file *f,
364 struct object_entry *entry,
365 off_t write_offset)
366 {
367 unsigned long size;
368 enum object_type type;
369 void *buf;
370 unsigned char header[10];
371 unsigned char dheader[10];
372 unsigned hdrlen;
373 off_t datalen;
374 enum object_type obj_type;
375 int to_reuse = 0;
376 /* write limit if limited packsize and not first object */
377 unsigned long limit = pack_size_limit && nr_written ?
378 pack_size_limit - write_offset : 0;
379 /* no if no delta */
380 int usable_delta = !entry->delta ? 0 :
381 /* yes if unlimited packfile */
382 !pack_size_limit ? 1 :
383 /* no if base written to previous pack */
384 entry->delta->idx.offset == (off_t)-1 ? 0 :
385 /* otherwise double-check written to this
386 * pack, like we do below
387 */
388 entry->delta->idx.offset ? 1 : 0;
390 if (!pack_to_stdout)
391 crc32_begin(f);
393 obj_type = entry->type;
394 if (no_reuse_object)
395 to_reuse = 0; /* explicit */
396 else if (!entry->in_pack)
397 to_reuse = 0; /* can't reuse what we don't have */
398 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
399 /* check_object() decided it for us ... */
400 to_reuse = usable_delta;
401 /* ... but pack split may override that */
402 else if (obj_type != entry->in_pack_type)
403 to_reuse = 0; /* pack has delta which is unusable */
404 else if (entry->delta)
405 to_reuse = 0; /* we want to pack afresh */
406 else
407 to_reuse = 1; /* we have it in-pack undeltified,
408 * and we do not need to deltify it.
409 */
411 if (!to_reuse) {
412 z_stream stream;
413 unsigned long maxsize;
414 void *out;
415 if (!usable_delta) {
416 buf = read_sha1_file(entry->idx.sha1, &obj_type, &size);
417 if (!buf)
418 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
419 } else if (entry->delta_data) {
420 size = entry->delta_size;
421 buf = entry->delta_data;
422 entry->delta_data = NULL;
423 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
424 OBJ_OFS_DELTA : OBJ_REF_DELTA;
425 } else {
426 buf = read_sha1_file(entry->idx.sha1, &type, &size);
427 if (!buf)
428 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
429 buf = delta_against(buf, size, entry);
430 size = entry->delta_size;
431 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
432 OBJ_OFS_DELTA : OBJ_REF_DELTA;
433 }
434 /* compress the data to store and put compressed length in datalen */
435 memset(&stream, 0, sizeof(stream));
436 deflateInit(&stream, pack_compression_level);
437 maxsize = deflateBound(&stream, size);
438 out = xmalloc(maxsize);
439 /* Compress it */
440 stream.next_in = buf;
441 stream.avail_in = size;
442 stream.next_out = out;
443 stream.avail_out = maxsize;
444 while (deflate(&stream, Z_FINISH) == Z_OK)
445 /* nothing */;
446 deflateEnd(&stream);
447 datalen = stream.total_out;
448 deflateEnd(&stream);
449 /*
450 * The object header is a byte of 'type' followed by zero or
451 * more bytes of length.
452 */
453 hdrlen = encode_header(obj_type, size, header);
455 if (obj_type == OBJ_OFS_DELTA) {
456 /*
457 * Deltas with relative base contain an additional
458 * encoding of the relative offset for the delta
459 * base from this object's position in the pack.
460 */
461 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
462 unsigned pos = sizeof(dheader) - 1;
463 dheader[pos] = ofs & 127;
464 while (ofs >>= 7)
465 dheader[--pos] = 128 | (--ofs & 127);
466 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
467 free(out);
468 free(buf);
469 return 0;
470 }
471 sha1write(f, header, hdrlen);
472 sha1write(f, dheader + pos, sizeof(dheader) - pos);
473 hdrlen += sizeof(dheader) - 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 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
480 free(out);
481 free(buf);
482 return 0;
483 }
484 sha1write(f, header, hdrlen);
485 sha1write(f, entry->delta->idx.sha1, 20);
486 hdrlen += 20;
487 } else {
488 if (limit && hdrlen + datalen + 20 >= limit) {
489 free(out);
490 free(buf);
491 return 0;
492 }
493 sha1write(f, header, hdrlen);
494 }
495 sha1write(f, out, datalen);
496 free(out);
497 free(buf);
498 }
499 else {
500 struct packed_git *p = entry->in_pack;
501 struct pack_window *w_curs = NULL;
502 struct revindex_entry *revidx;
503 off_t offset;
505 if (entry->delta) {
506 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
507 OBJ_OFS_DELTA : OBJ_REF_DELTA;
508 reused_delta++;
509 }
510 hdrlen = encode_header(obj_type, entry->size, header);
511 offset = entry->in_pack_offset;
512 revidx = find_packed_object(p, offset);
513 datalen = revidx[1].offset - offset;
514 if (!pack_to_stdout && p->index_version > 1 &&
515 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
516 die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
517 offset += entry->in_pack_header_size;
518 datalen -= entry->in_pack_header_size;
519 if (obj_type == OBJ_OFS_DELTA) {
520 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
521 unsigned pos = sizeof(dheader) - 1;
522 dheader[pos] = ofs & 127;
523 while (ofs >>= 7)
524 dheader[--pos] = 128 | (--ofs & 127);
525 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
526 return 0;
527 sha1write(f, header, hdrlen);
528 sha1write(f, dheader + pos, sizeof(dheader) - pos);
529 hdrlen += sizeof(dheader) - pos;
530 } else if (obj_type == OBJ_REF_DELTA) {
531 if (limit && hdrlen + 20 + datalen + 20 >= limit)
532 return 0;
533 sha1write(f, header, hdrlen);
534 sha1write(f, entry->delta->idx.sha1, 20);
535 hdrlen += 20;
536 } else {
537 if (limit && hdrlen + datalen + 20 >= limit)
538 return 0;
539 sha1write(f, header, hdrlen);
540 }
542 if (!pack_to_stdout && p->index_version == 1 &&
543 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
544 die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
545 copy_pack_data(f, p, &w_curs, offset, datalen);
546 unuse_pack(&w_curs);
547 reused++;
548 }
549 if (usable_delta)
550 written_delta++;
551 written++;
552 if (!pack_to_stdout)
553 entry->idx.crc32 = crc32_end(f);
554 return hdrlen + datalen;
555 }
557 static off_t write_one(struct sha1file *f,
558 struct object_entry *e,
559 off_t offset)
560 {
561 unsigned long size;
563 /* offset is non zero if object is written already. */
564 if (e->idx.offset || e->preferred_base)
565 return offset;
567 /* if we are deltified, write out base object first. */
568 if (e->delta) {
569 offset = write_one(f, e->delta, offset);
570 if (!offset)
571 return 0;
572 }
574 e->idx.offset = offset;
575 size = write_object(f, e, offset);
576 if (!size) {
577 e->idx.offset = 0;
578 return 0;
579 }
580 written_list[nr_written++] = e;
582 /* make sure off_t is sufficiently large not to wrap */
583 if (offset > offset + size)
584 die("pack too large for current definition of off_t");
585 return offset + size;
586 }
588 static int open_object_dir_tmp(const char *path)
589 {
590 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
591 return mkstemp(tmpname);
592 }
594 /* forward declaration for write_pack_file */
595 static int adjust_perm(const char *path, mode_t mode);
597 static void write_pack_file(void)
598 {
599 uint32_t i = 0, j;
600 struct sha1file *f;
601 off_t offset, offset_one, last_obj_offset = 0;
602 struct pack_header hdr;
603 int do_progress = progress >> pack_to_stdout;
604 uint32_t nr_remaining = nr_result;
606 if (do_progress)
607 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
608 written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
610 do {
611 unsigned char sha1[20];
613 if (pack_to_stdout) {
614 f = sha1fd(1, "<stdout>");
615 } else {
616 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
617 if (fd < 0)
618 die("unable to create %s: %s\n", tmpname, strerror(errno));
619 pack_tmp_name = xstrdup(tmpname);
620 f = sha1fd(fd, pack_tmp_name);
621 }
623 hdr.hdr_signature = htonl(PACK_SIGNATURE);
624 hdr.hdr_version = htonl(PACK_VERSION);
625 hdr.hdr_entries = htonl(nr_remaining);
626 sha1write(f, &hdr, sizeof(hdr));
627 offset = sizeof(hdr);
628 nr_written = 0;
629 for (; i < nr_objects; i++) {
630 last_obj_offset = offset;
631 offset_one = write_one(f, objects + i, offset);
632 if (!offset_one)
633 break;
634 offset = offset_one;
635 if (do_progress)
636 display_progress(&progress_state, written);
637 }
639 /*
640 * Did we write the wrong # entries in the header?
641 * If so, rewrite it like in fast-import
642 */
643 if (pack_to_stdout || nr_written == nr_remaining) {
644 sha1close(f, sha1, 1);
645 } else {
646 sha1close(f, sha1, 0);
647 fixup_pack_header_footer(f->fd, sha1, pack_tmp_name, nr_written);
648 close(f->fd);
649 }
651 if (!pack_to_stdout) {
652 mode_t mode = umask(0);
654 umask(mode);
655 mode = 0444 & ~mode;
657 idx_tmp_name = write_idx_file(NULL,
658 (struct pack_idx_entry **) written_list, nr_written, sha1);
659 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
660 base_name, sha1_to_hex(sha1));
661 if (adjust_perm(pack_tmp_name, mode))
662 die("unable to make temporary pack file readable: %s",
663 strerror(errno));
664 if (rename(pack_tmp_name, tmpname))
665 die("unable to rename temporary pack file: %s",
666 strerror(errno));
667 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
668 base_name, sha1_to_hex(sha1));
669 if (adjust_perm(idx_tmp_name, mode))
670 die("unable to make temporary index file readable: %s",
671 strerror(errno));
672 if (rename(idx_tmp_name, tmpname))
673 die("unable to rename temporary index file: %s",
674 strerror(errno));
675 puts(sha1_to_hex(sha1));
676 }
678 /* mark written objects as written to previous pack */
679 for (j = 0; j < nr_written; j++) {
680 written_list[j]->idx.offset = (off_t)-1;
681 }
682 nr_remaining -= nr_written;
683 } while (nr_remaining && i < nr_objects);
685 free(written_list);
686 if (do_progress)
687 stop_progress(&progress_state);
688 if (written != nr_result)
689 die("wrote %u objects while expecting %u", written, nr_result);
690 /*
691 * We have scanned through [0 ... i). Since we have written
692 * the correct number of objects, the remaining [i ... nr_objects)
693 * items must be either already written (due to out-of-order delta base)
694 * or a preferred base. Count those which are neither and complain if any.
695 */
696 for (j = 0; i < nr_objects; i++) {
697 struct object_entry *e = objects + i;
698 j += !e->idx.offset && !e->preferred_base;
699 }
700 if (j)
701 die("wrote %u objects as expected but %u unwritten", written, j);
702 }
704 static int locate_object_entry_hash(const unsigned char *sha1)
705 {
706 int i;
707 unsigned int ui;
708 memcpy(&ui, sha1, sizeof(unsigned int));
709 i = ui % object_ix_hashsz;
710 while (0 < object_ix[i]) {
711 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
712 return i;
713 if (++i == object_ix_hashsz)
714 i = 0;
715 }
716 return -1 - i;
717 }
719 static struct object_entry *locate_object_entry(const unsigned char *sha1)
720 {
721 int i;
723 if (!object_ix_hashsz)
724 return NULL;
726 i = locate_object_entry_hash(sha1);
727 if (0 <= i)
728 return &objects[object_ix[i]-1];
729 return NULL;
730 }
732 static void rehash_objects(void)
733 {
734 uint32_t i;
735 struct object_entry *oe;
737 object_ix_hashsz = nr_objects * 3;
738 if (object_ix_hashsz < 1024)
739 object_ix_hashsz = 1024;
740 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
741 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
742 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
743 int ix = locate_object_entry_hash(oe->idx.sha1);
744 if (0 <= ix)
745 continue;
746 ix = -1 - ix;
747 object_ix[ix] = i + 1;
748 }
749 }
751 static unsigned name_hash(const char *name)
752 {
753 unsigned char c;
754 unsigned hash = 0;
756 if (!name)
757 return 0;
759 /*
760 * This effectively just creates a sortable number from the
761 * last sixteen non-whitespace characters. Last characters
762 * count "most", so things that end in ".c" sort together.
763 */
764 while ((c = *name++) != 0) {
765 if (isspace(c))
766 continue;
767 hash = (hash >> 2) + (c << 24);
768 }
769 return hash;
770 }
772 static void setup_delta_attr_check(struct git_attr_check *check)
773 {
774 static struct git_attr *attr_delta;
776 if (!attr_delta)
777 attr_delta = git_attr("delta", 5);
779 check[0].attr = attr_delta;
780 }
782 static int no_try_delta(const char *path)
783 {
784 struct git_attr_check check[1];
786 setup_delta_attr_check(check);
787 if (git_checkattr(path, ARRAY_SIZE(check), check))
788 return 0;
789 if (ATTR_FALSE(check->value))
790 return 1;
791 return 0;
792 }
794 static int add_object_entry(const unsigned char *sha1, enum object_type type,
795 const char *name, int exclude)
796 {
797 struct object_entry *entry;
798 struct packed_git *p, *found_pack = NULL;
799 off_t found_offset = 0;
800 int ix;
801 unsigned hash = name_hash(name);
803 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
804 if (ix >= 0) {
805 if (exclude) {
806 entry = objects + object_ix[ix] - 1;
807 if (!entry->preferred_base)
808 nr_result--;
809 entry->preferred_base = 1;
810 }
811 return 0;
812 }
814 for (p = packed_git; p; p = p->next) {
815 off_t offset = find_pack_entry_one(sha1, p);
816 if (offset) {
817 if (!found_pack) {
818 found_offset = offset;
819 found_pack = p;
820 }
821 if (exclude)
822 break;
823 if (incremental)
824 return 0;
825 if (local && !p->pack_local)
826 return 0;
827 }
828 }
830 if (nr_objects >= nr_alloc) {
831 nr_alloc = (nr_alloc + 1024) * 3 / 2;
832 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
833 }
835 entry = objects + nr_objects++;
836 memset(entry, 0, sizeof(*entry));
837 hashcpy(entry->idx.sha1, sha1);
838 entry->hash = hash;
839 if (type)
840 entry->type = type;
841 if (exclude)
842 entry->preferred_base = 1;
843 else
844 nr_result++;
845 if (found_pack) {
846 entry->in_pack = found_pack;
847 entry->in_pack_offset = found_offset;
848 }
850 if (object_ix_hashsz * 3 <= nr_objects * 4)
851 rehash_objects();
852 else
853 object_ix[-1 - ix] = nr_objects;
855 if (progress)
856 display_progress(&progress_state, nr_objects);
858 if (name && no_try_delta(name))
859 entry->no_try_delta = 1;
861 return 1;
862 }
864 struct pbase_tree_cache {
865 unsigned char sha1[20];
866 int ref;
867 int temporary;
868 void *tree_data;
869 unsigned long tree_size;
870 };
872 static struct pbase_tree_cache *(pbase_tree_cache[256]);
873 static int pbase_tree_cache_ix(const unsigned char *sha1)
874 {
875 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
876 }
877 static int pbase_tree_cache_ix_incr(int ix)
878 {
879 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
880 }
882 static struct pbase_tree {
883 struct pbase_tree *next;
884 /* This is a phony "cache" entry; we are not
885 * going to evict it nor find it through _get()
886 * mechanism -- this is for the toplevel node that
887 * would almost always change with any commit.
888 */
889 struct pbase_tree_cache pcache;
890 } *pbase_tree;
892 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
893 {
894 struct pbase_tree_cache *ent, *nent;
895 void *data;
896 unsigned long size;
897 enum object_type type;
898 int neigh;
899 int my_ix = pbase_tree_cache_ix(sha1);
900 int available_ix = -1;
902 /* pbase-tree-cache acts as a limited hashtable.
903 * your object will be found at your index or within a few
904 * slots after that slot if it is cached.
905 */
906 for (neigh = 0; neigh < 8; neigh++) {
907 ent = pbase_tree_cache[my_ix];
908 if (ent && !hashcmp(ent->sha1, sha1)) {
909 ent->ref++;
910 return ent;
911 }
912 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
913 ((0 <= available_ix) &&
914 (!ent && pbase_tree_cache[available_ix])))
915 available_ix = my_ix;
916 if (!ent)
917 break;
918 my_ix = pbase_tree_cache_ix_incr(my_ix);
919 }
921 /* Did not find one. Either we got a bogus request or
922 * we need to read and perhaps cache.
923 */
924 data = read_sha1_file(sha1, &type, &size);
925 if (!data)
926 return NULL;
927 if (type != OBJ_TREE) {
928 free(data);
929 return NULL;
930 }
932 /* We need to either cache or return a throwaway copy */
934 if (available_ix < 0)
935 ent = NULL;
936 else {
937 ent = pbase_tree_cache[available_ix];
938 my_ix = available_ix;
939 }
941 if (!ent) {
942 nent = xmalloc(sizeof(*nent));
943 nent->temporary = (available_ix < 0);
944 }
945 else {
946 /* evict and reuse */
947 free(ent->tree_data);
948 nent = ent;
949 }
950 hashcpy(nent->sha1, sha1);
951 nent->tree_data = data;
952 nent->tree_size = size;
953 nent->ref = 1;
954 if (!nent->temporary)
955 pbase_tree_cache[my_ix] = nent;
956 return nent;
957 }
959 static void pbase_tree_put(struct pbase_tree_cache *cache)
960 {
961 if (!cache->temporary) {
962 cache->ref--;
963 return;
964 }
965 free(cache->tree_data);
966 free(cache);
967 }
969 static int name_cmp_len(const char *name)
970 {
971 int i;
972 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
973 ;
974 return i;
975 }
977 static void add_pbase_object(struct tree_desc *tree,
978 const char *name,
979 int cmplen,
980 const char *fullname)
981 {
982 struct name_entry entry;
983 int cmp;
985 while (tree_entry(tree,&entry)) {
986 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
987 memcmp(name, entry.path, cmplen);
988 if (cmp > 0)
989 continue;
990 if (cmp < 0)
991 return;
992 if (name[cmplen] != '/') {
993 add_object_entry(entry.sha1,
994 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
995 fullname, 1);
996 return;
997 }
998 if (S_ISDIR(entry.mode)) {
999 struct tree_desc sub;
1000 struct pbase_tree_cache *tree;
1001 const char *down = name+cmplen+1;
1002 int downlen = name_cmp_len(down);
1004 tree = pbase_tree_get(entry.sha1);
1005 if (!tree)
1006 return;
1007 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1009 add_pbase_object(&sub, down, downlen, fullname);
1010 pbase_tree_put(tree);
1011 }
1012 }
1013 }
1015 static unsigned *done_pbase_paths;
1016 static int done_pbase_paths_num;
1017 static int done_pbase_paths_alloc;
1018 static int done_pbase_path_pos(unsigned hash)
1019 {
1020 int lo = 0;
1021 int hi = done_pbase_paths_num;
1022 while (lo < hi) {
1023 int mi = (hi + lo) / 2;
1024 if (done_pbase_paths[mi] == hash)
1025 return mi;
1026 if (done_pbase_paths[mi] < hash)
1027 hi = mi;
1028 else
1029 lo = mi + 1;
1030 }
1031 return -lo-1;
1032 }
1034 static int check_pbase_path(unsigned hash)
1035 {
1036 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1037 if (0 <= pos)
1038 return 1;
1039 pos = -pos - 1;
1040 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1041 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1042 done_pbase_paths = xrealloc(done_pbase_paths,
1043 done_pbase_paths_alloc *
1044 sizeof(unsigned));
1045 }
1046 done_pbase_paths_num++;
1047 if (pos < done_pbase_paths_num)
1048 memmove(done_pbase_paths + pos + 1,
1049 done_pbase_paths + pos,
1050 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1051 done_pbase_paths[pos] = hash;
1052 return 0;
1053 }
1055 static void add_preferred_base_object(const char *name)
1056 {
1057 struct pbase_tree *it;
1058 int cmplen;
1059 unsigned hash = name_hash(name);
1061 if (!num_preferred_base || check_pbase_path(hash))
1062 return;
1064 cmplen = name_cmp_len(name);
1065 for (it = pbase_tree; it; it = it->next) {
1066 if (cmplen == 0) {
1067 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1068 }
1069 else {
1070 struct tree_desc tree;
1071 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1072 add_pbase_object(&tree, name, cmplen, name);
1073 }
1074 }
1075 }
1077 static void add_preferred_base(unsigned char *sha1)
1078 {
1079 struct pbase_tree *it;
1080 void *data;
1081 unsigned long size;
1082 unsigned char tree_sha1[20];
1084 if (window <= num_preferred_base++)
1085 return;
1087 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1088 if (!data)
1089 return;
1091 for (it = pbase_tree; it; it = it->next) {
1092 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1093 free(data);
1094 return;
1095 }
1096 }
1098 it = xcalloc(1, sizeof(*it));
1099 it->next = pbase_tree;
1100 pbase_tree = it;
1102 hashcpy(it->pcache.sha1, tree_sha1);
1103 it->pcache.tree_data = data;
1104 it->pcache.tree_size = size;
1105 }
1107 static void check_object(struct object_entry *entry)
1108 {
1109 if (entry->in_pack) {
1110 struct packed_git *p = entry->in_pack;
1111 struct pack_window *w_curs = NULL;
1112 const unsigned char *base_ref = NULL;
1113 struct object_entry *base_entry;
1114 unsigned long used, used_0;
1115 unsigned int avail;
1116 off_t ofs;
1117 unsigned char *buf, c;
1119 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1121 /*
1122 * We want in_pack_type even if we do not reuse delta
1123 * since non-delta representations could still be reused.
1124 */
1125 used = unpack_object_header_gently(buf, avail,
1126 &entry->in_pack_type,
1127 &entry->size);
1129 /*
1130 * Determine if this is a delta and if so whether we can
1131 * reuse it or not. Otherwise let's find out as cheaply as
1132 * possible what the actual type and size for this object is.
1133 */
1134 switch (entry->in_pack_type) {
1135 default:
1136 /* Not a delta hence we've already got all we need. */
1137 entry->type = entry->in_pack_type;
1138 entry->in_pack_header_size = used;
1139 unuse_pack(&w_curs);
1140 return;
1141 case OBJ_REF_DELTA:
1142 if (!no_reuse_delta && !entry->preferred_base)
1143 base_ref = use_pack(p, &w_curs,
1144 entry->in_pack_offset + used, NULL);
1145 entry->in_pack_header_size = used + 20;
1146 break;
1147 case OBJ_OFS_DELTA:
1148 buf = use_pack(p, &w_curs,
1149 entry->in_pack_offset + used, NULL);
1150 used_0 = 0;
1151 c = buf[used_0++];
1152 ofs = c & 127;
1153 while (c & 128) {
1154 ofs += 1;
1155 if (!ofs || MSB(ofs, 7))
1156 die("delta base offset overflow in pack for %s",
1157 sha1_to_hex(entry->idx.sha1));
1158 c = buf[used_0++];
1159 ofs = (ofs << 7) + (c & 127);
1160 }
1161 if (ofs >= entry->in_pack_offset)
1162 die("delta base offset out of bound for %s",
1163 sha1_to_hex(entry->idx.sha1));
1164 ofs = entry->in_pack_offset - ofs;
1165 if (!no_reuse_delta && !entry->preferred_base)
1166 base_ref = find_packed_object_name(p, ofs);
1167 entry->in_pack_header_size = used + used_0;
1168 break;
1169 }
1171 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1172 /*
1173 * If base_ref was set above that means we wish to
1174 * reuse delta data, and we even found that base
1175 * in the list of objects we want to pack. Goodie!
1176 *
1177 * Depth value does not matter - find_deltas() will
1178 * never consider reused delta as the base object to
1179 * deltify other objects against, in order to avoid
1180 * circular deltas.
1181 */
1182 entry->type = entry->in_pack_type;
1183 entry->delta = base_entry;
1184 entry->delta_sibling = base_entry->delta_child;
1185 base_entry->delta_child = entry;
1186 unuse_pack(&w_curs);
1187 return;
1188 }
1190 if (entry->type) {
1191 /*
1192 * This must be a delta and we already know what the
1193 * final object type is. Let's extract the actual
1194 * object size from the delta header.
1195 */
1196 entry->size = get_size_from_delta(p, &w_curs,
1197 entry->in_pack_offset + entry->in_pack_header_size);
1198 unuse_pack(&w_curs);
1199 return;
1200 }
1202 /*
1203 * No choice but to fall back to the recursive delta walk
1204 * with sha1_object_info() to find about the object type
1205 * at this point...
1206 */
1207 unuse_pack(&w_curs);
1208 }
1210 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1211 if (entry->type < 0)
1212 die("unable to get type of object %s",
1213 sha1_to_hex(entry->idx.sha1));
1214 }
1216 static int pack_offset_sort(const void *_a, const void *_b)
1217 {
1218 const struct object_entry *a = *(struct object_entry **)_a;
1219 const struct object_entry *b = *(struct object_entry **)_b;
1221 /* avoid filesystem trashing with loose objects */
1222 if (!a->in_pack && !b->in_pack)
1223 return hashcmp(a->idx.sha1, b->idx.sha1);
1225 if (a->in_pack < b->in_pack)
1226 return -1;
1227 if (a->in_pack > b->in_pack)
1228 return 1;
1229 return a->in_pack_offset < b->in_pack_offset ? -1 :
1230 (a->in_pack_offset > b->in_pack_offset);
1231 }
1233 static void get_object_details(void)
1234 {
1235 uint32_t i;
1236 struct object_entry **sorted_by_offset;
1238 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1239 for (i = 0; i < nr_objects; i++)
1240 sorted_by_offset[i] = objects + i;
1241 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1243 prepare_pack_ix();
1244 for (i = 0; i < nr_objects; i++)
1245 check_object(sorted_by_offset[i]);
1246 free(sorted_by_offset);
1247 }
1249 static int type_size_sort(const void *_a, const void *_b)
1250 {
1251 const struct object_entry *a = *(struct object_entry **)_a;
1252 const struct object_entry *b = *(struct object_entry **)_b;
1254 if (a->type < b->type)
1255 return -1;
1256 if (a->type > b->type)
1257 return 1;
1258 if (a->hash < b->hash)
1259 return -1;
1260 if (a->hash > b->hash)
1261 return 1;
1262 if (a->preferred_base < b->preferred_base)
1263 return -1;
1264 if (a->preferred_base > b->preferred_base)
1265 return 1;
1266 if (a->size < b->size)
1267 return -1;
1268 if (a->size > b->size)
1269 return 1;
1270 return a > b ? -1 : (a < b); /* newest last */
1271 }
1273 struct unpacked {
1274 struct object_entry *entry;
1275 void *data;
1276 struct delta_index *index;
1277 };
1279 static int delta_cacheable(struct unpacked *trg, struct unpacked *src,
1280 unsigned long src_size, unsigned long trg_size,
1281 unsigned long delta_size)
1282 {
1283 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1284 return 0;
1286 if (delta_size < cache_max_small_delta_size)
1287 return 1;
1289 /* cache delta, if objects are large enough compared to delta size */
1290 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1291 return 1;
1293 return 0;
1294 }
1296 /*
1297 * We search for deltas _backwards_ in a list sorted by type and
1298 * by size, so that we see progressively smaller and smaller files.
1299 * That's because we prefer deltas to be from the bigger file
1300 * to the smaller - deletes are potentially cheaper, but perhaps
1301 * more importantly, the bigger file is likely the more recent
1302 * one.
1303 */
1304 static int try_delta(struct unpacked *trg, struct unpacked *src,
1305 unsigned max_depth)
1306 {
1307 struct object_entry *trg_entry = trg->entry;
1308 struct object_entry *src_entry = src->entry;
1309 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1310 unsigned ref_depth;
1311 enum object_type type;
1312 void *delta_buf;
1314 /* Don't bother doing diffs between different types */
1315 if (trg_entry->type != src_entry->type)
1316 return -1;
1318 /* We do not compute delta to *create* objects we are not
1319 * going to pack.
1320 */
1321 if (trg_entry->preferred_base)
1322 return -1;
1324 /*
1325 * We do not bother to try a delta that we discarded
1326 * on an earlier try, but only when reusing delta data.
1327 */
1328 if (!no_reuse_delta && trg_entry->in_pack &&
1329 trg_entry->in_pack == src_entry->in_pack &&
1330 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1331 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1332 return 0;
1334 /* Let's not bust the allowed depth. */
1335 if (src_entry->depth >= max_depth)
1336 return 0;
1338 /* Now some size filtering heuristics. */
1339 trg_size = trg_entry->size;
1340 if (!trg_entry->delta) {
1341 max_size = trg_size/2 - 20;
1342 ref_depth = 1;
1343 } else {
1344 max_size = trg_entry->delta_size;
1345 ref_depth = trg_entry->depth;
1346 }
1347 max_size = max_size * (max_depth - src_entry->depth) /
1348 (max_depth - ref_depth + 1);
1349 if (max_size == 0)
1350 return 0;
1351 src_size = src_entry->size;
1352 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1353 if (sizediff >= max_size)
1354 return 0;
1355 if (trg_size < src_size / 32)
1356 return 0;
1358 /* Load data if not already done */
1359 if (!trg->data) {
1360 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1361 if (sz != trg_size)
1362 die("object %s inconsistent object length (%lu vs %lu)",
1363 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1364 window_memory_usage += sz;
1365 }
1366 if (!src->data) {
1367 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1368 if (sz != src_size)
1369 die("object %s inconsistent object length (%lu vs %lu)",
1370 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1371 window_memory_usage += sz;
1372 }
1373 if (!src->index) {
1374 src->index = create_delta_index(src->data, src_size);
1375 if (!src->index) {
1376 static int warned = 0;
1377 if (!warned++)
1378 warning("suboptimal pack - out of memory");
1379 return 0;
1380 }
1381 window_memory_usage += sizeof_delta_index(src->index);
1382 }
1384 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1385 if (!delta_buf)
1386 return 0;
1388 if (trg_entry->delta_data) {
1389 /* Prefer only shallower same-sized deltas. */
1390 if (delta_size == trg_entry->delta_size &&
1391 src_entry->depth + 1 >= trg_entry->depth) {
1392 free(delta_buf);
1393 return 0;
1394 }
1395 delta_cache_size -= trg_entry->delta_size;
1396 free(trg_entry->delta_data);
1397 }
1398 trg_entry->delta_data = 0;
1399 trg_entry->delta = src_entry;
1400 trg_entry->delta_size = delta_size;
1401 trg_entry->depth = src_entry->depth + 1;
1403 if (delta_cacheable(src, trg, src_size, trg_size, delta_size)) {
1404 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1405 delta_cache_size += trg_entry->delta_size;
1406 } else
1407 free(delta_buf);
1408 return 1;
1409 }
1411 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1412 {
1413 struct object_entry *child = me->delta_child;
1414 unsigned int m = n;
1415 while (child) {
1416 unsigned int c = check_delta_limit(child, n + 1);
1417 if (m < c)
1418 m = c;
1419 child = child->delta_sibling;
1420 }
1421 return m;
1422 }
1424 static void free_unpacked(struct unpacked *n)
1425 {
1426 window_memory_usage -= sizeof_delta_index(n->index);
1427 free_delta_index(n->index);
1428 n->index = NULL;
1429 if (n->data) {
1430 free(n->data);
1431 n->data = NULL;
1432 window_memory_usage -= n->entry->size;
1433 }
1434 n->entry = NULL;
1435 }
1437 static void find_deltas(struct object_entry **list, int window, int depth)
1438 {
1439 uint32_t i = nr_objects, idx = 0, count = 0, processed = 0;
1440 unsigned int array_size = window * sizeof(struct unpacked);
1441 struct unpacked *array;
1442 int max_depth;
1444 if (!nr_objects)
1445 return;
1446 array = xmalloc(array_size);
1447 memset(array, 0, array_size);
1448 if (progress)
1449 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1451 do {
1452 struct object_entry *entry = list[--i];
1453 struct unpacked *n = array + idx;
1454 int j;
1456 if (!entry->preferred_base)
1457 processed++;
1459 if (progress)
1460 display_progress(&progress_state, processed);
1462 if (entry->delta)
1463 /* This happens if we decided to reuse existing
1464 * delta from a pack. "!no_reuse_delta &&" is implied.
1465 */
1466 continue;
1468 if (entry->size < 50)
1469 continue;
1471 if (entry->no_try_delta)
1472 continue;
1474 free_unpacked(n);
1475 n->entry = entry;
1477 while (window_memory_limit &&
1478 window_memory_usage > window_memory_limit &&
1479 count > 1) {
1480 uint32_t tail = (idx + window - count) % window;
1481 free_unpacked(array + tail);
1482 count--;
1483 }
1485 /*
1486 * If the current object is at pack edge, take the depth the
1487 * objects that depend on the current object into account
1488 * otherwise they would become too deep.
1489 */
1490 max_depth = depth;
1491 if (entry->delta_child) {
1492 max_depth -= check_delta_limit(entry, 0);
1493 if (max_depth <= 0)
1494 goto next;
1495 }
1497 j = window;
1498 while (--j > 0) {
1499 uint32_t other_idx = idx + j;
1500 struct unpacked *m;
1501 if (other_idx >= window)
1502 other_idx -= window;
1503 m = array + other_idx;
1504 if (!m->entry)
1505 break;
1506 if (try_delta(n, m, max_depth) < 0)
1507 break;
1508 }
1510 /* if we made n a delta, and if n is already at max
1511 * depth, leaving it in the window is pointless. we
1512 * should evict it first.
1513 */
1514 if (entry->delta && depth <= entry->depth)
1515 continue;
1517 next:
1518 idx++;
1519 if (count + 1 < window)
1520 count++;
1521 if (idx >= window)
1522 idx = 0;
1523 } while (i > 0);
1525 if (progress)
1526 stop_progress(&progress_state);
1528 for (i = 0; i < window; ++i) {
1529 free_delta_index(array[i].index);
1530 free(array[i].data);
1531 }
1532 free(array);
1533 }
1535 static void prepare_pack(int window, int depth)
1536 {
1537 struct object_entry **delta_list;
1538 uint32_t i;
1540 get_object_details();
1542 if (!window || !depth)
1543 return;
1545 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1546 for (i = 0; i < nr_objects; i++)
1547 delta_list[i] = objects + i;
1548 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1549 find_deltas(delta_list, window+1, depth);
1550 free(delta_list);
1551 }
1553 static int git_pack_config(const char *k, const char *v)
1554 {
1555 if(!strcmp(k, "pack.window")) {
1556 window = git_config_int(k, v);
1557 return 0;
1558 }
1559 if (!strcmp(k, "pack.windowmemory")) {
1560 window_memory_limit = git_config_ulong(k, v);
1561 return 0;
1562 }
1563 if (!strcmp(k, "pack.depth")) {
1564 depth = git_config_int(k, v);
1565 return 0;
1566 }
1567 if (!strcmp(k, "pack.compression")) {
1568 int level = git_config_int(k, v);
1569 if (level == -1)
1570 level = Z_DEFAULT_COMPRESSION;
1571 else if (level < 0 || level > Z_BEST_COMPRESSION)
1572 die("bad pack compression level %d", level);
1573 pack_compression_level = level;
1574 pack_compression_seen = 1;
1575 return 0;
1576 }
1577 if (!strcmp(k, "pack.deltacachesize")) {
1578 max_delta_cache_size = git_config_int(k, v);
1579 return 0;
1580 }
1581 if (!strcmp(k, "pack.deltacachelimit")) {
1582 cache_max_small_delta_size = git_config_int(k, v);
1583 return 0;
1584 }
1585 return git_default_config(k, v);
1586 }
1588 static void read_object_list_from_stdin(void)
1589 {
1590 char line[40 + 1 + PATH_MAX + 2];
1591 unsigned char sha1[20];
1593 for (;;) {
1594 if (!fgets(line, sizeof(line), stdin)) {
1595 if (feof(stdin))
1596 break;
1597 if (!ferror(stdin))
1598 die("fgets returned NULL, not EOF, not error!");
1599 if (errno != EINTR)
1600 die("fgets: %s", strerror(errno));
1601 clearerr(stdin);
1602 continue;
1603 }
1604 if (line[0] == '-') {
1605 if (get_sha1_hex(line+1, sha1))
1606 die("expected edge sha1, got garbage:\n %s",
1607 line);
1608 add_preferred_base(sha1);
1609 continue;
1610 }
1611 if (get_sha1_hex(line, sha1))
1612 die("expected sha1, got garbage:\n %s", line);
1614 add_preferred_base_object(line+41);
1615 add_object_entry(sha1, 0, line+41, 0);
1616 }
1617 }
1619 static void show_commit(struct commit *commit)
1620 {
1621 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1622 }
1624 static void show_object(struct object_array_entry *p)
1625 {
1626 add_preferred_base_object(p->name);
1627 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1628 }
1630 static void show_edge(struct commit *commit)
1631 {
1632 add_preferred_base(commit->object.sha1);
1633 }
1635 static void get_object_list(int ac, const char **av)
1636 {
1637 struct rev_info revs;
1638 char line[1000];
1639 int flags = 0;
1641 init_revisions(&revs, NULL);
1642 save_commit_buffer = 0;
1643 track_object_refs = 0;
1644 setup_revisions(ac, av, &revs, NULL);
1646 while (fgets(line, sizeof(line), stdin) != NULL) {
1647 int len = strlen(line);
1648 if (line[len - 1] == '\n')
1649 line[--len] = 0;
1650 if (!len)
1651 break;
1652 if (*line == '-') {
1653 if (!strcmp(line, "--not")) {
1654 flags ^= UNINTERESTING;
1655 continue;
1656 }
1657 die("not a rev '%s'", line);
1658 }
1659 if (handle_revision_arg(line, &revs, flags, 1))
1660 die("bad revision '%s'", line);
1661 }
1663 prepare_revision_walk(&revs);
1664 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1665 traverse_commit_list(&revs, show_commit, show_object);
1666 }
1668 static int adjust_perm(const char *path, mode_t mode)
1669 {
1670 if (chmod(path, mode))
1671 return -1;
1672 return adjust_shared_perm(path);
1673 }
1675 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1676 {
1677 int use_internal_rev_list = 0;
1678 int thin = 0;
1679 uint32_t i;
1680 const char **rp_av;
1681 int rp_ac_alloc = 64;
1682 int rp_ac;
1684 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1686 rp_av[0] = "pack-objects";
1687 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1688 rp_ac = 2;
1690 git_config(git_pack_config);
1691 if (!pack_compression_seen && core_compression_seen)
1692 pack_compression_level = core_compression_level;
1694 progress = isatty(2);
1695 for (i = 1; i < argc; i++) {
1696 const char *arg = argv[i];
1698 if (*arg != '-')
1699 break;
1701 if (!strcmp("--non-empty", arg)) {
1702 non_empty = 1;
1703 continue;
1704 }
1705 if (!strcmp("--local", arg)) {
1706 local = 1;
1707 continue;
1708 }
1709 if (!strcmp("--incremental", arg)) {
1710 incremental = 1;
1711 continue;
1712 }
1713 if (!prefixcmp(arg, "--compression=")) {
1714 char *end;
1715 int level = strtoul(arg+14, &end, 0);
1716 if (!arg[14] || *end)
1717 usage(pack_usage);
1718 if (level == -1)
1719 level = Z_DEFAULT_COMPRESSION;
1720 else if (level < 0 || level > Z_BEST_COMPRESSION)
1721 die("bad pack compression level %d", level);
1722 pack_compression_level = level;
1723 continue;
1724 }
1725 if (!prefixcmp(arg, "--max-pack-size=")) {
1726 char *end;
1727 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1728 if (!arg[16] || *end)
1729 usage(pack_usage);
1730 continue;
1731 }
1732 if (!prefixcmp(arg, "--window=")) {
1733 char *end;
1734 window = strtoul(arg+9, &end, 0);
1735 if (!arg[9] || *end)
1736 usage(pack_usage);
1737 continue;
1738 }
1739 if (!prefixcmp(arg, "--window-memory=")) {
1740 if (!git_parse_ulong(arg+16, &window_memory_limit))
1741 usage(pack_usage);
1742 continue;
1743 }
1744 if (!prefixcmp(arg, "--depth=")) {
1745 char *end;
1746 depth = strtoul(arg+8, &end, 0);
1747 if (!arg[8] || *end)
1748 usage(pack_usage);
1749 continue;
1750 }
1751 if (!strcmp("--progress", arg)) {
1752 progress = 1;
1753 continue;
1754 }
1755 if (!strcmp("--all-progress", arg)) {
1756 progress = 2;
1757 continue;
1758 }
1759 if (!strcmp("-q", arg)) {
1760 progress = 0;
1761 continue;
1762 }
1763 if (!strcmp("--no-reuse-delta", arg)) {
1764 no_reuse_delta = 1;
1765 continue;
1766 }
1767 if (!strcmp("--no-reuse-object", arg)) {
1768 no_reuse_object = no_reuse_delta = 1;
1769 continue;
1770 }
1771 if (!strcmp("--delta-base-offset", arg)) {
1772 allow_ofs_delta = 1;
1773 continue;
1774 }
1775 if (!strcmp("--stdout", arg)) {
1776 pack_to_stdout = 1;
1777 continue;
1778 }
1779 if (!strcmp("--revs", arg)) {
1780 use_internal_rev_list = 1;
1781 continue;
1782 }
1783 if (!strcmp("--unpacked", arg) ||
1784 !prefixcmp(arg, "--unpacked=") ||
1785 !strcmp("--reflog", arg) ||
1786 !strcmp("--all", arg)) {
1787 use_internal_rev_list = 1;
1788 if (rp_ac >= rp_ac_alloc - 1) {
1789 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1790 rp_av = xrealloc(rp_av,
1791 rp_ac_alloc * sizeof(*rp_av));
1792 }
1793 rp_av[rp_ac++] = arg;
1794 continue;
1795 }
1796 if (!strcmp("--thin", arg)) {
1797 use_internal_rev_list = 1;
1798 thin = 1;
1799 rp_av[1] = "--objects-edge";
1800 continue;
1801 }
1802 if (!prefixcmp(arg, "--index-version=")) {
1803 char *c;
1804 pack_idx_default_version = strtoul(arg + 16, &c, 10);
1805 if (pack_idx_default_version > 2)
1806 die("bad %s", arg);
1807 if (*c == ',')
1808 pack_idx_off32_limit = strtoul(c+1, &c, 0);
1809 if (*c || pack_idx_off32_limit & 0x80000000)
1810 die("bad %s", arg);
1811 continue;
1812 }
1813 usage(pack_usage);
1814 }
1816 /* Traditionally "pack-objects [options] base extra" failed;
1817 * we would however want to take refs parameter that would
1818 * have been given to upstream rev-list ourselves, which means
1819 * we somehow want to say what the base name is. So the
1820 * syntax would be:
1821 *
1822 * pack-objects [options] base <refs...>
1823 *
1824 * in other words, we would treat the first non-option as the
1825 * base_name and send everything else to the internal revision
1826 * walker.
1827 */
1829 if (!pack_to_stdout)
1830 base_name = argv[i++];
1832 if (pack_to_stdout != !base_name)
1833 usage(pack_usage);
1835 if (pack_to_stdout && pack_size_limit)
1836 die("--max-pack-size cannot be used to build a pack for transfer.");
1838 if (!pack_to_stdout && thin)
1839 die("--thin cannot be used to build an indexable pack.");
1841 prepare_packed_git();
1843 if (progress)
1844 start_progress(&progress_state, "Generating pack...",
1845 "Counting objects: ", 0);
1846 if (!use_internal_rev_list)
1847 read_object_list_from_stdin();
1848 else {
1849 rp_av[rp_ac] = NULL;
1850 get_object_list(rp_ac, rp_av);
1851 }
1852 if (progress) {
1853 stop_progress(&progress_state);
1854 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1855 }
1857 if (non_empty && !nr_result)
1858 return 0;
1859 if (progress && (nr_objects != nr_result))
1860 fprintf(stderr, "Result has %u objects.\n", nr_result);
1861 if (nr_result)
1862 prepare_pack(window, depth);
1863 write_pack_file();
1864 if (progress)
1865 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1866 written, written_delta, reused, reused_delta);
1867 return 0;
1868 }