]>
Commit | Line | Data |
---|---|---|
5d4a6003 | 1 | #include "builtin.h" |
c323ac7d | 2 | #include "cache.h" |
a74db82e | 3 | #include "attr.h" |
c323ac7d | 4 | #include "object.h" |
8e440259 PE |
5 | #include "blob.h" |
6 | #include "commit.h" | |
7 | #include "tag.h" | |
8 | #include "tree.h" | |
c323ac7d | 9 | #include "delta.h" |
a733cb60 | 10 | #include "pack.h" |
3449f8c4 | 11 | #include "pack-revindex.h" |
c38138cd | 12 | #include "csum-file.h" |
1b0c7174 | 13 | #include "tree-walk.h" |
b5d97e6b JH |
14 | #include "diff.h" |
15 | #include "revision.h" | |
16 | #include "list-objects.h" | |
96a02f8f | 17 | #include "progress.h" |
f0a24aa5 | 18 | #include "refs.h" |
93749194 | 19 | #include "thread-utils.h" |
8ecce684 | 20 | |
812bdbc3 | 21 | static const char pack_usage[] = |
0adda936 | 22 | "git pack-objects [ -q | --progress | --all-progress ]\n" |
4f366275 | 23 | " [--all-progress-implied]\n" |
62b4698e ŠN |
24 | " [--max-pack-size=<n>] [--local] [--incremental]\n" |
25 | " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n" | |
812bdbc3 | 26 | " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n" |
62b4698e | 27 | " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n" |
812bdbc3 | 28 | " [--reflog] [--stdout | base-name] [--include-tag]\n" |
86953531 | 29 | " [--keep-unreachable | --unpack-unreachable]\n" |
88422065 | 30 | " [< ref-list | < object-list]"; |
c323ac7d | 31 | |
c323ac7d | 32 | struct object_entry { |
aa7e44bf | 33 | struct pack_idx_entry idx; |
3f9ac8d2 | 34 | unsigned long size; /* uncompressed size */ |
3f9ac8d2 | 35 | struct packed_git *in_pack; /* already in pack */ |
6777a59f | 36 | off_t in_pack_offset; |
a3fbf4df | 37 | struct object_entry *delta; /* delta base object */ |
82e5a82f | 38 | struct object_entry *delta_child; /* deltified objects who bases me */ |
15b4d577 JH |
39 | struct object_entry *delta_sibling; /* other deltified objects who |
40 | * uses the same base as me | |
41 | */ | |
074b2eea | 42 | void *delta_data; /* cached delta (uncompressed) */ |
a3fbf4df | 43 | unsigned long delta_size; /* delta data size (uncompressed) */ |
ed4a9031 | 44 | unsigned long z_delta_size; /* delta data size (compressed) */ |
5a235b5e | 45 | unsigned int hash; /* name hint hash */ |
a3fbf4df NP |
46 | enum object_type type; |
47 | enum object_type in_pack_type; /* could be delta */ | |
48 | unsigned char in_pack_header_size; | |
49 | unsigned char preferred_base; /* we do not pack this, but is available | |
a74db82e | 50 | * to be used as the base object to delta |
a3fbf4df NP |
51 | * objects against. |
52 | */ | |
a74db82e | 53 | unsigned char no_try_delta; |
1b4bb16b JH |
54 | unsigned char tagged; /* near the very tip of refs */ |
55 | unsigned char filled; /* assigned write-order */ | |
c323ac7d LT |
56 | }; |
57 | ||
3f9ac8d2 | 58 | /* |
82e5a82f | 59 | * Objects we are going to pack are collected in objects array (dynamically |
3f9ac8d2 JH |
60 | * expanded). nr_objects & nr_alloc controls this array. They are stored |
61 | * in the order we see -- typically rev-list --objects order that gives us | |
62 | * nice "minimum seek" order. | |
3f9ac8d2 | 63 | */ |
e4d58311 | 64 | static struct object_entry *objects; |
79814f42 | 65 | static struct pack_idx_entry **written_list; |
d01fb92f | 66 | static uint32_t nr_objects, nr_alloc, nr_result, nr_written; |
3f9ac8d2 | 67 | |
96f1e58f | 68 | static int non_empty; |
a7de7130 | 69 | static int reuse_delta = 1, reuse_object = 1; |
e5e9714a | 70 | static int keep_unreachable, unpack_unreachable, include_tag; |
96f1e58f DR |
71 | static int local; |
72 | static int incremental; | |
e96fb9b8 | 73 | static int ignore_packed_keep; |
be6b1914 | 74 | static int allow_ofs_delta; |
ebcfb379 | 75 | static struct pack_idx_option pack_idx_opts; |
d01fb92f | 76 | static const char *base_name; |
024701f1 | 77 | static int progress = 1; |
4812a93a | 78 | static int window = 10; |
07cf0f24 | 79 | static unsigned long pack_size_limit, pack_size_limit_cfg; |
618e613a | 80 | static int depth = 50; |
43cc2b42 | 81 | static int delta_search_threads; |
df6d6101 | 82 | static int pack_to_stdout; |
8d1d8f83 | 83 | static int num_preferred_base; |
dc6a0757 | 84 | static struct progress *progress_state; |
960ccca6 DH |
85 | static int pack_compression_level = Z_DEFAULT_COMPRESSION; |
86 | static int pack_compression_seen; | |
c323ac7d | 87 | |
074b2eea | 88 | static unsigned long delta_cache_size = 0; |
5749b0b2 | 89 | static unsigned long max_delta_cache_size = 256 * 1024 * 1024; |
e3dfddb3 | 90 | static unsigned long cache_max_small_delta_size = 1000; |
074b2eea | 91 | |
a97773ce BD |
92 | static unsigned long window_memory_limit = 0; |
93 | ||
3f9ac8d2 JH |
94 | /* |
95 | * The object names in objects array are hashed with this hashtable, | |
e4d58311 | 96 | * to help looking up the entry by object name. |
3f9ac8d2 JH |
97 | * This hashtable is built after all the objects are seen. |
98 | */ | |
96f1e58f DR |
99 | static int *object_ix; |
100 | static int object_ix_hashsz; | |
1b4bb16b | 101 | static struct object_entry *locate_object_entry(const unsigned char *sha1); |
3f9ac8d2 | 102 | |
3f9ac8d2 JH |
103 | /* |
104 | * stats | |
105 | */ | |
7cadf491 SP |
106 | static uint32_t written, written_delta; |
107 | static uint32_t reused, reused_delta; | |
3f9ac8d2 | 108 | |
780e6e73 | 109 | |
3613f9b4 | 110 | static void *get_delta(struct object_entry *entry) |
c323ac7d | 111 | { |
3613f9b4 NP |
112 | unsigned long size, base_size, delta_size; |
113 | void *buf, *base_buf, *delta_buf; | |
21666f1a | 114 | enum object_type type; |
c323ac7d | 115 | |
3613f9b4 NP |
116 | buf = read_sha1_file(entry->idx.sha1, &type, &size); |
117 | if (!buf) | |
118 | die("unable to read %s", sha1_to_hex(entry->idx.sha1)); | |
119 | base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size); | |
120 | if (!base_buf) | |
aa7e44bf | 121 | die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1)); |
3613f9b4 | 122 | delta_buf = diff_delta(base_buf, base_size, |
dcde55bc | 123 | buf, size, &delta_size, 0); |
3613f9b4 | 124 | if (!delta_buf || delta_size != entry->delta_size) |
a6080a0a | 125 | die("delta size changed"); |
3613f9b4 NP |
126 | free(buf); |
127 | free(base_buf); | |
c323ac7d LT |
128 | return delta_buf; |
129 | } | |
130 | ||
30ebb40a NP |
131 | static unsigned long do_compress(void **pptr, unsigned long size) |
132 | { | |
ef49a7a0 | 133 | git_zstream stream; |
30ebb40a NP |
134 | void *in, *out; |
135 | unsigned long maxsize; | |
136 | ||
137 | memset(&stream, 0, sizeof(stream)); | |
55bb5c91 | 138 | git_deflate_init(&stream, pack_compression_level); |
225a6f10 | 139 | maxsize = git_deflate_bound(&stream, size); |
30ebb40a NP |
140 | |
141 | in = *pptr; | |
142 | out = xmalloc(maxsize); | |
143 | *pptr = out; | |
144 | ||
145 | stream.next_in = in; | |
146 | stream.avail_in = size; | |
147 | stream.next_out = out; | |
148 | stream.avail_out = maxsize; | |
55bb5c91 | 149 | while (git_deflate(&stream, Z_FINISH) == Z_OK) |
30ebb40a | 150 | ; /* nothing */ |
55bb5c91 | 151 | git_deflate_end(&stream); |
30ebb40a NP |
152 | |
153 | free(in); | |
154 | return stream.total_out; | |
155 | } | |
156 | ||
780e6e73 NP |
157 | /* |
158 | * we are going to reuse the existing object data as is. make | |
159 | * sure it is not corrupt. | |
160 | */ | |
079afb18 SP |
161 | static int check_pack_inflate(struct packed_git *p, |
162 | struct pack_window **w_curs, | |
6777a59f SP |
163 | off_t offset, |
164 | off_t len, | |
079afb18 SP |
165 | unsigned long expect) |
166 | { | |
ef49a7a0 | 167 | git_zstream stream; |
079afb18 SP |
168 | unsigned char fakebuf[4096], *in; |
169 | int st; | |
170 | ||
171 | memset(&stream, 0, sizeof(stream)); | |
39c68542 | 172 | git_inflate_init(&stream); |
079afb18 SP |
173 | do { |
174 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
175 | stream.next_in = in; | |
176 | stream.next_out = fakebuf; | |
177 | stream.avail_out = sizeof(fakebuf); | |
39c68542 | 178 | st = git_inflate(&stream, Z_FINISH); |
079afb18 SP |
179 | offset += stream.next_in - in; |
180 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
39c68542 | 181 | git_inflate_end(&stream); |
079afb18 SP |
182 | return (st == Z_STREAM_END && |
183 | stream.total_out == expect && | |
184 | stream.total_in == len) ? 0 : -1; | |
185 | } | |
186 | ||
187 | static void copy_pack_data(struct sha1file *f, | |
188 | struct packed_git *p, | |
189 | struct pack_window **w_curs, | |
6777a59f SP |
190 | off_t offset, |
191 | off_t len) | |
079afb18 SP |
192 | { |
193 | unsigned char *in; | |
ef49a7a0 | 194 | unsigned long avail; |
079afb18 SP |
195 | |
196 | while (len) { | |
197 | in = use_pack(p, w_curs, offset, &avail); | |
198 | if (avail > len) | |
ef49a7a0 | 199 | avail = (unsigned long)len; |
079afb18 SP |
200 | sha1write(f, in, avail); |
201 | offset += avail; | |
202 | len -= avail; | |
203 | } | |
204 | } | |
205 | ||
1b4bb16b | 206 | /* Return 0 if we will bust the pack-size limit */ |
d7dd0223 | 207 | static unsigned long write_object(struct sha1file *f, |
17b08f2c DH |
208 | struct object_entry *entry, |
209 | off_t write_offset) | |
c323ac7d | 210 | { |
30ebb40a | 211 | unsigned long size, limit, datalen; |
3f9ac8d2 | 212 | void *buf; |
2c5ef824 | 213 | unsigned char header[10], dheader[10]; |
6777a59f | 214 | unsigned hdrlen; |
2c5ef824 NP |
215 | enum object_type type; |
216 | int usable_delta, to_reuse; | |
c323ac7d | 217 | |
78d1e84f NP |
218 | if (!pack_to_stdout) |
219 | crc32_begin(f); | |
220 | ||
2c5ef824 NP |
221 | type = entry->type; |
222 | ||
a2430dde | 223 | /* apply size limit if limited packsize and not first object */ |
a1e4760f NP |
224 | if (!pack_size_limit || !nr_written) |
225 | limit = 0; | |
226 | else if (pack_size_limit <= write_offset) | |
227 | /* | |
228 | * the earlier object did not fit the limit; avoid | |
229 | * mistaking this with unlimited (i.e. limit = 0). | |
230 | */ | |
231 | limit = 1; | |
232 | else | |
233 | limit = pack_size_limit - write_offset; | |
2c5ef824 NP |
234 | |
235 | if (!entry->delta) | |
236 | usable_delta = 0; /* no delta */ | |
237 | else if (!pack_size_limit) | |
238 | usable_delta = 1; /* unlimited packfile */ | |
239 | else if (entry->delta->idx.offset == (off_t)-1) | |
240 | usable_delta = 0; /* base was written to another pack */ | |
241 | else if (entry->delta->idx.offset) | |
242 | usable_delta = 1; /* base already exists in this pack */ | |
243 | else | |
244 | usable_delta = 0; /* base could end up in another pack */ | |
245 | ||
a7de7130 | 246 | if (!reuse_object) |
fa736f72 NP |
247 | to_reuse = 0; /* explicit */ |
248 | else if (!entry->in_pack) | |
ab7cd7bb | 249 | to_reuse = 0; /* can't reuse what we don't have */ |
2c5ef824 | 250 | else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA) |
17b08f2c DH |
251 | /* check_object() decided it for us ... */ |
252 | to_reuse = usable_delta; | |
253 | /* ... but pack split may override that */ | |
2c5ef824 | 254 | else if (type != entry->in_pack_type) |
ab7cd7bb JH |
255 | to_reuse = 0; /* pack has delta which is unusable */ |
256 | else if (entry->delta) | |
257 | to_reuse = 0; /* we want to pack afresh */ | |
258 | else | |
259 | to_reuse = 1; /* we have it in-pack undeltified, | |
260 | * and we do not need to deltify it. | |
261 | */ | |
262 | ||
df6d6101 | 263 | if (!to_reuse) { |
64bd76b1 | 264 | no_reuse: |
5476a8ad | 265 | if (!usable_delta) { |
2c5ef824 | 266 | buf = read_sha1_file(entry->idx.sha1, &type, &size); |
5476a8ad | 267 | if (!buf) |
aa7e44bf | 268 | die("unable to read %s", sha1_to_hex(entry->idx.sha1)); |
ed4a9031 NP |
269 | /* |
270 | * make sure no cached delta data remains from a | |
3ea3c215 | 271 | * previous attempt before a pack split occurred. |
ed4a9031 NP |
272 | */ |
273 | free(entry->delta_data); | |
274 | entry->delta_data = NULL; | |
275 | entry->z_delta_size = 0; | |
5476a8ad | 276 | } else if (entry->delta_data) { |
3f9ac8d2 | 277 | size = entry->delta_size; |
5476a8ad NP |
278 | buf = entry->delta_data; |
279 | entry->delta_data = NULL; | |
2c5ef824 | 280 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
be6b1914 | 281 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
17b08f2c | 282 | } else { |
3613f9b4 | 283 | buf = get_delta(entry); |
5476a8ad | 284 | size = entry->delta_size; |
2c5ef824 | 285 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
5476a8ad | 286 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
3f9ac8d2 | 287 | } |
ed4a9031 NP |
288 | |
289 | if (entry->z_delta_size) | |
290 | datalen = entry->z_delta_size; | |
291 | else | |
292 | datalen = do_compress(&buf, size); | |
8c3c7b2a | 293 | |
3f9ac8d2 JH |
294 | /* |
295 | * The object header is a byte of 'type' followed by zero or | |
be6b1914 | 296 | * more bytes of length. |
3f9ac8d2 | 297 | */ |
1b22b6c8 | 298 | hdrlen = encode_in_pack_object_header(type, size, header); |
3f9ac8d2 | 299 | |
2c5ef824 | 300 | if (type == OBJ_OFS_DELTA) { |
be6b1914 NP |
301 | /* |
302 | * Deltas with relative base contain an additional | |
303 | * encoding of the relative offset for the delta | |
304 | * base from this object's position in the pack. | |
305 | */ | |
aa7e44bf | 306 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; |
17b08f2c DH |
307 | unsigned pos = sizeof(dheader) - 1; |
308 | dheader[pos] = ofs & 127; | |
be6b1914 | 309 | while (ofs >>= 7) |
17b08f2c DH |
310 | dheader[--pos] = 128 | (--ofs & 127); |
311 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) { | |
17b08f2c DH |
312 | free(buf); |
313 | return 0; | |
314 | } | |
315 | sha1write(f, header, hdrlen); | |
316 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
317 | hdrlen += sizeof(dheader) - pos; | |
2c5ef824 | 318 | } else if (type == OBJ_REF_DELTA) { |
be6b1914 NP |
319 | /* |
320 | * Deltas with a base reference contain | |
321 | * an additional 20 bytes for the base sha1. | |
322 | */ | |
17b08f2c | 323 | if (limit && hdrlen + 20 + datalen + 20 >= limit) { |
17b08f2c DH |
324 | free(buf); |
325 | return 0; | |
326 | } | |
327 | sha1write(f, header, hdrlen); | |
aa7e44bf | 328 | sha1write(f, entry->delta->idx.sha1, 20); |
3f9ac8d2 | 329 | hdrlen += 20; |
17b08f2c DH |
330 | } else { |
331 | if (limit && hdrlen + datalen + 20 >= limit) { | |
17b08f2c DH |
332 | free(buf); |
333 | return 0; | |
334 | } | |
335 | sha1write(f, header, hdrlen); | |
3f9ac8d2 | 336 | } |
30ebb40a | 337 | sha1write(f, buf, datalen); |
3f9ac8d2 | 338 | free(buf); |
c323ac7d | 339 | } |
3f9ac8d2 JH |
340 | else { |
341 | struct packed_git *p = entry->in_pack; | |
03e79c88 | 342 | struct pack_window *w_curs = NULL; |
91ecbeca | 343 | struct revindex_entry *revidx; |
6777a59f | 344 | off_t offset; |
3f9ac8d2 | 345 | |
64bd76b1 | 346 | if (entry->delta) |
2c5ef824 | 347 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
780e6e73 | 348 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
1b22b6c8 | 349 | hdrlen = encode_in_pack_object_header(type, entry->size, header); |
64bd76b1 | 350 | |
91ecbeca | 351 | offset = entry->in_pack_offset; |
3449f8c4 | 352 | revidx = find_pack_revindex(p, offset); |
91ecbeca NP |
353 | datalen = revidx[1].offset - offset; |
354 | if (!pack_to_stdout && p->index_version > 1 && | |
64bd76b1 NP |
355 | check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { |
356 | error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1)); | |
357 | unuse_pack(&w_curs); | |
358 | goto no_reuse; | |
359 | } | |
360 | ||
91ecbeca NP |
361 | offset += entry->in_pack_header_size; |
362 | datalen -= entry->in_pack_header_size; | |
64bd76b1 NP |
363 | if (!pack_to_stdout && p->index_version == 1 && |
364 | check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) { | |
365 | error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1)); | |
366 | unuse_pack(&w_curs); | |
367 | goto no_reuse; | |
368 | } | |
369 | ||
2c5ef824 | 370 | if (type == OBJ_OFS_DELTA) { |
aa7e44bf | 371 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; |
17b08f2c DH |
372 | unsigned pos = sizeof(dheader) - 1; |
373 | dheader[pos] = ofs & 127; | |
374 | while (ofs >>= 7) | |
375 | dheader[--pos] = 128 | (--ofs & 127); | |
59dd9ed1 NP |
376 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) { |
377 | unuse_pack(&w_curs); | |
17b08f2c | 378 | return 0; |
59dd9ed1 | 379 | } |
17b08f2c DH |
380 | sha1write(f, header, hdrlen); |
381 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
382 | hdrlen += sizeof(dheader) - pos; | |
64bd76b1 | 383 | reused_delta++; |
2c5ef824 | 384 | } else if (type == OBJ_REF_DELTA) { |
59dd9ed1 NP |
385 | if (limit && hdrlen + 20 + datalen + 20 >= limit) { |
386 | unuse_pack(&w_curs); | |
17b08f2c | 387 | return 0; |
59dd9ed1 | 388 | } |
17b08f2c | 389 | sha1write(f, header, hdrlen); |
aa7e44bf | 390 | sha1write(f, entry->delta->idx.sha1, 20); |
17b08f2c | 391 | hdrlen += 20; |
64bd76b1 | 392 | reused_delta++; |
17b08f2c | 393 | } else { |
59dd9ed1 NP |
394 | if (limit && hdrlen + datalen + 20 >= limit) { |
395 | unuse_pack(&w_curs); | |
17b08f2c | 396 | return 0; |
59dd9ed1 | 397 | } |
17b08f2c DH |
398 | sha1write(f, header, hdrlen); |
399 | } | |
079afb18 | 400 | copy_pack_data(f, p, &w_curs, offset, datalen); |
03e79c88 | 401 | unuse_pack(&w_curs); |
3f9ac8d2 | 402 | reused++; |
a733cb60 | 403 | } |
17b08f2c | 404 | if (usable_delta) |
ab7cd7bb | 405 | written_delta++; |
3f9ac8d2 | 406 | written++; |
78d1e84f | 407 | if (!pack_to_stdout) |
aa7e44bf | 408 | entry->idx.crc32 = crc32_end(f); |
c323ac7d LT |
409 | return hdrlen + datalen; |
410 | } | |
411 | ||
6ed7f25e | 412 | static int write_one(struct sha1file *f, |
9d5ab962 | 413 | struct object_entry *e, |
6ed7f25e | 414 | off_t *offset) |
9d5ab962 | 415 | { |
d7dd0223 NP |
416 | unsigned long size; |
417 | ||
418 | /* offset is non zero if object is written already. */ | |
aa7e44bf | 419 | if (e->idx.offset || e->preferred_base) |
a2430dde | 420 | return -1; |
d7dd0223 | 421 | |
720c9f7b NP |
422 | /* if we are deltified, write out base object first. */ |
423 | if (e->delta && !write_one(f, e->delta, offset)) | |
424 | return 0; | |
d7dd0223 | 425 | |
6ed7f25e NP |
426 | e->idx.offset = *offset; |
427 | size = write_object(f, e, *offset); | |
17b08f2c | 428 | if (!size) { |
aa7e44bf | 429 | e->idx.offset = 0; |
17b08f2c DH |
430 | return 0; |
431 | } | |
79814f42 | 432 | written_list[nr_written++] = &e->idx; |
d7dd0223 NP |
433 | |
434 | /* make sure off_t is sufficiently large not to wrap */ | |
c03c8315 | 435 | if (signed_add_overflows(*offset, size)) |
d7dd0223 | 436 | die("pack too large for current definition of off_t"); |
6ed7f25e NP |
437 | *offset += size; |
438 | return 1; | |
9d5ab962 JH |
439 | } |
440 | ||
1b4bb16b JH |
441 | static int mark_tagged(const char *path, const unsigned char *sha1, int flag, |
442 | void *cb_data) | |
443 | { | |
444 | unsigned char peeled[20]; | |
445 | struct object_entry *entry = locate_object_entry(sha1); | |
446 | ||
447 | if (entry) | |
448 | entry->tagged = 1; | |
449 | if (!peel_ref(path, peeled)) { | |
450 | entry = locate_object_entry(peeled); | |
451 | if (entry) | |
452 | entry->tagged = 1; | |
453 | } | |
454 | return 0; | |
455 | } | |
456 | ||
457 | static void add_to_write_order(struct object_entry **wo, | |
458 | int *endp, | |
459 | struct object_entry *e) | |
460 | { | |
461 | if (e->filled) | |
462 | return; | |
463 | wo[(*endp)++] = e; | |
464 | e->filled = 1; | |
465 | } | |
466 | ||
467 | static void add_descendants_to_write_order(struct object_entry **wo, | |
468 | int *endp, | |
469 | struct object_entry *e) | |
470 | { | |
471 | struct object_entry *child; | |
472 | ||
473 | for (child = e->delta_child; child; child = child->delta_sibling) | |
474 | add_to_write_order(wo, endp, child); | |
475 | for (child = e->delta_child; child; child = child->delta_sibling) | |
476 | add_descendants_to_write_order(wo, endp, child); | |
477 | } | |
478 | ||
479 | static void add_family_to_write_order(struct object_entry **wo, | |
480 | int *endp, | |
481 | struct object_entry *e) | |
482 | { | |
483 | struct object_entry *root; | |
484 | ||
485 | for (root = e; root->delta; root = root->delta) | |
486 | ; /* nothing */ | |
487 | add_to_write_order(wo, endp, root); | |
488 | add_descendants_to_write_order(wo, endp, root); | |
489 | } | |
490 | ||
491 | static struct object_entry **compute_write_order(void) | |
492 | { | |
493 | int i, wo_end; | |
494 | ||
495 | struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo)); | |
496 | ||
497 | for (i = 0; i < nr_objects; i++) { | |
498 | objects[i].tagged = 0; | |
499 | objects[i].filled = 0; | |
500 | objects[i].delta_child = NULL; | |
501 | objects[i].delta_sibling = NULL; | |
502 | } | |
503 | ||
504 | /* | |
505 | * Fully connect delta_child/delta_sibling network. | |
506 | * Make sure delta_sibling is sorted in the original | |
507 | * recency order. | |
508 | */ | |
509 | for (i = nr_objects - 1; 0 <= i; i--) { | |
510 | struct object_entry *e = &objects[i]; | |
511 | if (!e->delta) | |
512 | continue; | |
513 | /* Mark me as the first child */ | |
514 | e->delta_sibling = e->delta->delta_child; | |
515 | e->delta->delta_child = e; | |
516 | } | |
517 | ||
518 | /* | |
519 | * Mark objects that are at the tip of tags. | |
520 | */ | |
521 | for_each_tag_ref(mark_tagged, NULL); | |
522 | ||
523 | /* | |
524 | * Give the commits in the original recency order until | |
525 | * we see a tagged tip. | |
526 | */ | |
527 | for (i = wo_end = 0; i < nr_objects; i++) { | |
528 | if (objects[i].tagged) | |
529 | break; | |
530 | add_to_write_order(wo, &wo_end, &objects[i]); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Then fill all the tagged tips. | |
535 | */ | |
536 | for (; i < nr_objects; i++) { | |
537 | if (objects[i].tagged) | |
538 | add_to_write_order(wo, &wo_end, &objects[i]); | |
539 | } | |
540 | ||
541 | /* | |
542 | * And then all remaining commits and tags. | |
543 | */ | |
544 | for (i = 0; i < nr_objects; i++) { | |
545 | if (objects[i].type != OBJ_COMMIT && | |
546 | objects[i].type != OBJ_TAG) | |
547 | continue; | |
548 | add_to_write_order(wo, &wo_end, &objects[i]); | |
549 | } | |
550 | ||
551 | /* | |
552 | * And then all the trees. | |
553 | */ | |
554 | for (i = 0; i < nr_objects; i++) { | |
555 | if (objects[i].type != OBJ_TREE) | |
556 | continue; | |
557 | add_to_write_order(wo, &wo_end, &objects[i]); | |
558 | } | |
559 | ||
560 | /* | |
561 | * Finally all the rest in really tight order | |
562 | */ | |
563 | for (i = 0; i < nr_objects; i++) | |
564 | add_family_to_write_order(wo, &wo_end, &objects[i]); | |
565 | ||
566 | return wo; | |
567 | } | |
568 | ||
d01fb92f | 569 | static void write_pack_file(void) |
c323ac7d | 570 | { |
ebe27b13 | 571 | uint32_t i = 0, j; |
d22b9290 | 572 | struct sha1file *f; |
6ed7f25e | 573 | off_t offset; |
a733cb60 | 574 | struct pack_header hdr; |
ebe27b13 | 575 | uint32_t nr_remaining = nr_result; |
f746bae8 | 576 | time_t last_mtime = 0; |
1b4bb16b | 577 | struct object_entry **write_order; |
81a216a5 | 578 | |
bcd7954e | 579 | if (progress > pack_to_stdout) |
dc6a0757 | 580 | progress_state = start_progress("Writing objects", nr_result); |
79814f42 | 581 | written_list = xmalloc(nr_objects * sizeof(*written_list)); |
1b4bb16b | 582 | write_order = compute_write_order(); |
183bdb2c | 583 | |
ebe27b13 | 584 | do { |
aa7e44bf | 585 | unsigned char sha1[20]; |
7ba502c4 | 586 | char *pack_tmp_name = NULL; |
aa7e44bf | 587 | |
ebe27b13 | 588 | if (pack_to_stdout) { |
2a128d63 | 589 | f = sha1fd_throughput(1, "<stdout>", progress_state); |
ebe27b13 | 590 | } else { |
7ba502c4 NP |
591 | char tmpname[PATH_MAX]; |
592 | int fd; | |
6e180cdc JH |
593 | fd = odb_mkstemp(tmpname, sizeof(tmpname), |
594 | "pack/tmp_pack_XXXXXX"); | |
ebe27b13 DH |
595 | pack_tmp_name = xstrdup(tmpname); |
596 | f = sha1fd(fd, pack_tmp_name); | |
597 | } | |
598 | ||
599 | hdr.hdr_signature = htonl(PACK_SIGNATURE); | |
600 | hdr.hdr_version = htonl(PACK_VERSION); | |
601 | hdr.hdr_entries = htonl(nr_remaining); | |
602 | sha1write(f, &hdr, sizeof(hdr)); | |
603 | offset = sizeof(hdr); | |
604 | nr_written = 0; | |
720c9f7b | 605 | for (; i < nr_objects; i++) { |
1b4bb16b JH |
606 | struct object_entry *e = write_order[i]; |
607 | if (!write_one(f, e, &offset)) | |
720c9f7b NP |
608 | break; |
609 | display_progress(progress_state, written); | |
610 | } | |
c553ca25 | 611 | |
ebe27b13 DH |
612 | /* |
613 | * Did we write the wrong # entries in the header? | |
614 | * If so, rewrite it like in fast-import | |
615 | */ | |
54352bb2 LT |
616 | if (pack_to_stdout) { |
617 | sha1close(f, sha1, CSUM_CLOSE); | |
618 | } else if (nr_written == nr_remaining) { | |
619 | sha1close(f, sha1, CSUM_FSYNC); | |
ebe27b13 | 620 | } else { |
ac0463ed | 621 | int fd = sha1close(f, sha1, 0); |
abeb40e5 | 622 | fixup_pack_header_footer(fd, sha1, pack_tmp_name, |
ac0463ed | 623 | nr_written, sha1, offset); |
7ba502c4 | 624 | close(fd); |
ebe27b13 DH |
625 | } |
626 | ||
627 | if (!pack_to_stdout) { | |
f746bae8 | 628 | struct stat st; |
3bb72562 LT |
629 | const char *idx_tmp_name; |
630 | char tmpname[PATH_MAX]; | |
d01fb92f | 631 | |
ebcfb379 JH |
632 | idx_tmp_name = write_idx_file(NULL, written_list, nr_written, |
633 | &pack_idx_opts, sha1); | |
f746bae8 | 634 | |
d01fb92f | 635 | snprintf(tmpname, sizeof(tmpname), "%s-%s.pack", |
aa7e44bf | 636 | base_name, sha1_to_hex(sha1)); |
c74faea1 | 637 | free_pack_by_name(tmpname); |
f80c7ae8 | 638 | if (adjust_shared_perm(pack_tmp_name)) |
d824cbba | 639 | die_errno("unable to make temporary pack file readable"); |
d01fb92f | 640 | if (rename(pack_tmp_name, tmpname)) |
d824cbba | 641 | die_errno("unable to rename temporary pack file"); |
f746bae8 NP |
642 | |
643 | /* | |
644 | * Packs are runtime accessed in their mtime | |
645 | * order since newer packs are more likely to contain | |
646 | * younger objects. So if we are creating multiple | |
647 | * packs then we should modify the mtime of later ones | |
648 | * to preserve this property. | |
649 | */ | |
650 | if (stat(tmpname, &st) < 0) { | |
651 | warning("failed to stat %s: %s", | |
652 | tmpname, strerror(errno)); | |
653 | } else if (!last_mtime) { | |
654 | last_mtime = st.st_mtime; | |
655 | } else { | |
656 | struct utimbuf utb; | |
657 | utb.actime = st.st_atime; | |
658 | utb.modtime = --last_mtime; | |
659 | if (utime(tmpname, &utb) < 0) | |
660 | warning("failed utime() on %s: %s", | |
661 | tmpname, strerror(errno)); | |
662 | } | |
663 | ||
d01fb92f | 664 | snprintf(tmpname, sizeof(tmpname), "%s-%s.idx", |
aa7e44bf | 665 | base_name, sha1_to_hex(sha1)); |
f80c7ae8 | 666 | if (adjust_shared_perm(idx_tmp_name)) |
d824cbba | 667 | die_errno("unable to make temporary index file readable"); |
d01fb92f | 668 | if (rename(idx_tmp_name, tmpname)) |
d824cbba | 669 | die_errno("unable to rename temporary index file"); |
f746bae8 | 670 | |
3bb72562 | 671 | free((void *) idx_tmp_name); |
7ba502c4 | 672 | free(pack_tmp_name); |
aa7e44bf | 673 | puts(sha1_to_hex(sha1)); |
ebe27b13 DH |
674 | } |
675 | ||
676 | /* mark written objects as written to previous pack */ | |
677 | for (j = 0; j < nr_written; j++) { | |
79814f42 | 678 | written_list[j]->offset = (off_t)-1; |
ebe27b13 DH |
679 | } |
680 | nr_remaining -= nr_written; | |
720c9f7b | 681 | } while (nr_remaining && i < nr_objects); |
ebe27b13 DH |
682 | |
683 | free(written_list); | |
1b4bb16b | 684 | free(write_order); |
4d4fcc54 | 685 | stop_progress(&progress_state); |
67c08ce1 | 686 | if (written != nr_result) |
6e1c2344 RJ |
687 | die("wrote %"PRIu32" objects while expecting %"PRIu32, |
688 | written, nr_result); | |
c323ac7d LT |
689 | } |
690 | ||
7a979d99 JH |
691 | static int locate_object_entry_hash(const unsigned char *sha1) |
692 | { | |
693 | int i; | |
694 | unsigned int ui; | |
695 | memcpy(&ui, sha1, sizeof(unsigned int)); | |
696 | i = ui % object_ix_hashsz; | |
697 | while (0 < object_ix[i]) { | |
aa7e44bf | 698 | if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1)) |
7a979d99 JH |
699 | return i; |
700 | if (++i == object_ix_hashsz) | |
701 | i = 0; | |
702 | } | |
703 | return -1 - i; | |
704 | } | |
705 | ||
706 | static struct object_entry *locate_object_entry(const unsigned char *sha1) | |
707 | { | |
708 | int i; | |
709 | ||
710 | if (!object_ix_hashsz) | |
711 | return NULL; | |
712 | ||
713 | i = locate_object_entry_hash(sha1); | |
714 | if (0 <= i) | |
715 | return &objects[object_ix[i]-1]; | |
716 | return NULL; | |
717 | } | |
718 | ||
719 | static void rehash_objects(void) | |
c323ac7d | 720 | { |
7cadf491 | 721 | uint32_t i; |
7a979d99 JH |
722 | struct object_entry *oe; |
723 | ||
724 | object_ix_hashsz = nr_objects * 3; | |
725 | if (object_ix_hashsz < 1024) | |
726 | object_ix_hashsz = 1024; | |
727 | object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz); | |
5379a5c5 | 728 | memset(object_ix, 0, sizeof(int) * object_ix_hashsz); |
7a979d99 | 729 | for (i = 0, oe = objects; i < nr_objects; i++, oe++) { |
aa7e44bf | 730 | int ix = locate_object_entry_hash(oe->idx.sha1); |
7a979d99 JH |
731 | if (0 <= ix) |
732 | continue; | |
733 | ix = -1 - ix; | |
734 | object_ix[ix] = i + 1; | |
735 | } | |
736 | } | |
737 | ||
ce0bd642 | 738 | static unsigned name_hash(const char *name) |
1d6b38cc | 739 | { |
48fb7deb | 740 | unsigned c, hash = 0; |
ce0bd642 | 741 | |
bc32fed5 JH |
742 | if (!name) |
743 | return 0; | |
744 | ||
eeef7135 | 745 | /* |
ce0bd642 LT |
746 | * This effectively just creates a sortable number from the |
747 | * last sixteen non-whitespace characters. Last characters | |
748 | * count "most", so things that end in ".c" sort together. | |
eeef7135 | 749 | */ |
ce0bd642 LT |
750 | while ((c = *name++) != 0) { |
751 | if (isspace(c)) | |
752 | continue; | |
753 | hash = (hash >> 2) + (c << 24); | |
754 | } | |
1d6b38cc JH |
755 | return hash; |
756 | } | |
757 | ||
a74db82e JH |
758 | static void setup_delta_attr_check(struct git_attr_check *check) |
759 | { | |
760 | static struct git_attr *attr_delta; | |
761 | ||
762 | if (!attr_delta) | |
7fb0eaa2 | 763 | attr_delta = git_attr("delta"); |
a74db82e JH |
764 | |
765 | check[0].attr = attr_delta; | |
766 | } | |
767 | ||
768 | static int no_try_delta(const char *path) | |
769 | { | |
770 | struct git_attr_check check[1]; | |
771 | ||
772 | setup_delta_attr_check(check); | |
d932f4eb | 773 | if (git_check_attr(path, ARRAY_SIZE(check), check)) |
a74db82e JH |
774 | return 0; |
775 | if (ATTR_FALSE(check->value)) | |
776 | return 1; | |
777 | return 0; | |
778 | } | |
779 | ||
5c49c116 | 780 | static int add_object_entry(const unsigned char *sha1, enum object_type type, |
bc32fed5 | 781 | const char *name, int exclude) |
c323ac7d | 782 | { |
c323ac7d | 783 | struct object_entry *entry; |
29b734e4 | 784 | struct packed_git *p, *found_pack = NULL; |
6777a59f | 785 | off_t found_offset = 0; |
29b734e4 | 786 | int ix; |
bc32fed5 | 787 | unsigned hash = name_hash(name); |
29b734e4 NP |
788 | |
789 | ix = nr_objects ? locate_object_entry_hash(sha1) : -1; | |
790 | if (ix >= 0) { | |
791 | if (exclude) { | |
792 | entry = objects + object_ix[ix] - 1; | |
81a216a5 NP |
793 | if (!entry->preferred_base) |
794 | nr_result--; | |
29b734e4 NP |
795 | entry->preferred_base = 1; |
796 | } | |
797 | return 0; | |
798 | } | |
c323ac7d | 799 | |
daae0625 BC |
800 | if (!exclude && local && has_loose_object_nonlocal(sha1)) |
801 | return 0; | |
802 | ||
5c49c116 NP |
803 | for (p = packed_git; p; p = p->next) { |
804 | off_t offset = find_pack_entry_one(sha1, p); | |
805 | if (offset) { | |
806 | if (!found_pack) { | |
807 | found_offset = offset; | |
808 | found_pack = p; | |
64560374 | 809 | } |
5c49c116 NP |
810 | if (exclude) |
811 | break; | |
812 | if (incremental) | |
813 | return 0; | |
814 | if (local && !p->pack_local) | |
815 | return 0; | |
e96fb9b8 BC |
816 | if (ignore_packed_keep && p->pack_local && p->pack_keep) |
817 | return 0; | |
64560374 LT |
818 | } |
819 | } | |
eb019375 | 820 | |
29b734e4 NP |
821 | if (nr_objects >= nr_alloc) { |
822 | nr_alloc = (nr_alloc + 1024) * 3 / 2; | |
7cadf491 | 823 | objects = xrealloc(objects, nr_alloc * sizeof(*entry)); |
c323ac7d | 824 | } |
29b734e4 NP |
825 | |
826 | entry = objects + nr_objects++; | |
c323ac7d | 827 | memset(entry, 0, sizeof(*entry)); |
aa7e44bf | 828 | hashcpy(entry->idx.sha1, sha1); |
27225f2e | 829 | entry->hash = hash; |
5c49c116 NP |
830 | if (type) |
831 | entry->type = type; | |
29b734e4 NP |
832 | if (exclude) |
833 | entry->preferred_base = 1; | |
81a216a5 NP |
834 | else |
835 | nr_result++; | |
29b734e4 NP |
836 | if (found_pack) { |
837 | entry->in_pack = found_pack; | |
838 | entry->in_pack_offset = found_offset; | |
839 | } | |
7a979d99 JH |
840 | |
841 | if (object_ix_hashsz * 3 <= nr_objects * 4) | |
842 | rehash_objects(); | |
29b734e4 NP |
843 | else |
844 | object_ix[-1 - ix] = nr_objects; | |
7a979d99 | 845 | |
4d4fcc54 | 846 | display_progress(progress_state, nr_objects); |
29b734e4 | 847 | |
a74db82e JH |
848 | if (name && no_try_delta(name)) |
849 | entry->no_try_delta = 1; | |
850 | ||
29b734e4 | 851 | return 1; |
c323ac7d LT |
852 | } |
853 | ||
5379a5c5 JH |
854 | struct pbase_tree_cache { |
855 | unsigned char sha1[20]; | |
856 | int ref; | |
857 | int temporary; | |
858 | void *tree_data; | |
859 | unsigned long tree_size; | |
860 | }; | |
861 | ||
862 | static struct pbase_tree_cache *(pbase_tree_cache[256]); | |
863 | static int pbase_tree_cache_ix(const unsigned char *sha1) | |
864 | { | |
865 | return sha1[0] % ARRAY_SIZE(pbase_tree_cache); | |
866 | } | |
867 | static int pbase_tree_cache_ix_incr(int ix) | |
868 | { | |
869 | return (ix+1) % ARRAY_SIZE(pbase_tree_cache); | |
870 | } | |
871 | ||
872 | static struct pbase_tree { | |
873 | struct pbase_tree *next; | |
874 | /* This is a phony "cache" entry; we are not | |
875 | * going to evict it nor find it through _get() | |
876 | * mechanism -- this is for the toplevel node that | |
877 | * would almost always change with any commit. | |
878 | */ | |
879 | struct pbase_tree_cache pcache; | |
880 | } *pbase_tree; | |
881 | ||
882 | static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1) | |
883 | { | |
884 | struct pbase_tree_cache *ent, *nent; | |
885 | void *data; | |
886 | unsigned long size; | |
21666f1a | 887 | enum object_type type; |
5379a5c5 JH |
888 | int neigh; |
889 | int my_ix = pbase_tree_cache_ix(sha1); | |
890 | int available_ix = -1; | |
891 | ||
892 | /* pbase-tree-cache acts as a limited hashtable. | |
893 | * your object will be found at your index or within a few | |
894 | * slots after that slot if it is cached. | |
895 | */ | |
896 | for (neigh = 0; neigh < 8; neigh++) { | |
897 | ent = pbase_tree_cache[my_ix]; | |
a89fccd2 | 898 | if (ent && !hashcmp(ent->sha1, sha1)) { |
5379a5c5 JH |
899 | ent->ref++; |
900 | return ent; | |
901 | } | |
902 | else if (((available_ix < 0) && (!ent || !ent->ref)) || | |
903 | ((0 <= available_ix) && | |
904 | (!ent && pbase_tree_cache[available_ix]))) | |
905 | available_ix = my_ix; | |
906 | if (!ent) | |
907 | break; | |
908 | my_ix = pbase_tree_cache_ix_incr(my_ix); | |
909 | } | |
910 | ||
911 | /* Did not find one. Either we got a bogus request or | |
912 | * we need to read and perhaps cache. | |
913 | */ | |
21666f1a | 914 | data = read_sha1_file(sha1, &type, &size); |
5379a5c5 JH |
915 | if (!data) |
916 | return NULL; | |
21666f1a | 917 | if (type != OBJ_TREE) { |
5379a5c5 JH |
918 | free(data); |
919 | return NULL; | |
920 | } | |
921 | ||
922 | /* We need to either cache or return a throwaway copy */ | |
923 | ||
924 | if (available_ix < 0) | |
925 | ent = NULL; | |
926 | else { | |
927 | ent = pbase_tree_cache[available_ix]; | |
928 | my_ix = available_ix; | |
929 | } | |
930 | ||
931 | if (!ent) { | |
932 | nent = xmalloc(sizeof(*nent)); | |
933 | nent->temporary = (available_ix < 0); | |
934 | } | |
935 | else { | |
936 | /* evict and reuse */ | |
937 | free(ent->tree_data); | |
938 | nent = ent; | |
939 | } | |
e702496e | 940 | hashcpy(nent->sha1, sha1); |
5379a5c5 JH |
941 | nent->tree_data = data; |
942 | nent->tree_size = size; | |
943 | nent->ref = 1; | |
944 | if (!nent->temporary) | |
945 | pbase_tree_cache[my_ix] = nent; | |
946 | return nent; | |
947 | } | |
948 | ||
949 | static void pbase_tree_put(struct pbase_tree_cache *cache) | |
950 | { | |
951 | if (!cache->temporary) { | |
952 | cache->ref--; | |
953 | return; | |
954 | } | |
955 | free(cache->tree_data); | |
956 | free(cache); | |
957 | } | |
958 | ||
959 | static int name_cmp_len(const char *name) | |
960 | { | |
961 | int i; | |
962 | for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++) | |
963 | ; | |
964 | return i; | |
965 | } | |
966 | ||
967 | static void add_pbase_object(struct tree_desc *tree, | |
5379a5c5 | 968 | const char *name, |
ce0bd642 LT |
969 | int cmplen, |
970 | const char *fullname) | |
3f9ac8d2 | 971 | { |
4c068a98 | 972 | struct name_entry entry; |
8a5a8d6c | 973 | int cmp; |
4c068a98 LT |
974 | |
975 | while (tree_entry(tree,&entry)) { | |
1211be6b LT |
976 | if (S_ISGITLINK(entry.mode)) |
977 | continue; | |
8a5a8d6c NP |
978 | cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 : |
979 | memcmp(name, entry.path, cmplen); | |
980 | if (cmp > 0) | |
7a979d99 | 981 | continue; |
8a5a8d6c NP |
982 | if (cmp < 0) |
983 | return; | |
5379a5c5 | 984 | if (name[cmplen] != '/') { |
5c49c116 | 985 | add_object_entry(entry.sha1, |
4d1012c3 | 986 | object_type(entry.mode), |
bc32fed5 | 987 | fullname, 1); |
5379a5c5 JH |
988 | return; |
989 | } | |
8a5a8d6c | 990 | if (S_ISDIR(entry.mode)) { |
7a979d99 | 991 | struct tree_desc sub; |
5379a5c5 JH |
992 | struct pbase_tree_cache *tree; |
993 | const char *down = name+cmplen+1; | |
994 | int downlen = name_cmp_len(down); | |
995 | ||
4c068a98 | 996 | tree = pbase_tree_get(entry.sha1); |
5379a5c5 JH |
997 | if (!tree) |
998 | return; | |
6fda5e51 | 999 | init_tree_desc(&sub, tree->tree_data, tree->tree_size); |
5379a5c5 | 1000 | |
ce0bd642 | 1001 | add_pbase_object(&sub, down, downlen, fullname); |
5379a5c5 JH |
1002 | pbase_tree_put(tree); |
1003 | } | |
1004 | } | |
1005 | } | |
1d6b38cc | 1006 | |
5379a5c5 JH |
1007 | static unsigned *done_pbase_paths; |
1008 | static int done_pbase_paths_num; | |
1009 | static int done_pbase_paths_alloc; | |
1010 | static int done_pbase_path_pos(unsigned hash) | |
1011 | { | |
1012 | int lo = 0; | |
1013 | int hi = done_pbase_paths_num; | |
1014 | while (lo < hi) { | |
1015 | int mi = (hi + lo) / 2; | |
1016 | if (done_pbase_paths[mi] == hash) | |
1017 | return mi; | |
1018 | if (done_pbase_paths[mi] < hash) | |
1019 | hi = mi; | |
1020 | else | |
1021 | lo = mi + 1; | |
1022 | } | |
1023 | return -lo-1; | |
1024 | } | |
1025 | ||
1026 | static int check_pbase_path(unsigned hash) | |
1027 | { | |
1028 | int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash); | |
1029 | if (0 <= pos) | |
1030 | return 1; | |
1031 | pos = -pos - 1; | |
1032 | if (done_pbase_paths_alloc <= done_pbase_paths_num) { | |
1033 | done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc); | |
1034 | done_pbase_paths = xrealloc(done_pbase_paths, | |
1035 | done_pbase_paths_alloc * | |
1036 | sizeof(unsigned)); | |
1037 | } | |
1038 | done_pbase_paths_num++; | |
1039 | if (pos < done_pbase_paths_num) | |
1040 | memmove(done_pbase_paths + pos + 1, | |
1041 | done_pbase_paths + pos, | |
1042 | (done_pbase_paths_num - pos - 1) * sizeof(unsigned)); | |
1043 | done_pbase_paths[pos] = hash; | |
1044 | return 0; | |
1045 | } | |
1046 | ||
bc32fed5 | 1047 | static void add_preferred_base_object(const char *name) |
5379a5c5 JH |
1048 | { |
1049 | struct pbase_tree *it; | |
8a5a8d6c | 1050 | int cmplen; |
bc32fed5 | 1051 | unsigned hash = name_hash(name); |
5379a5c5 | 1052 | |
8a5a8d6c | 1053 | if (!num_preferred_base || check_pbase_path(hash)) |
5379a5c5 JH |
1054 | return; |
1055 | ||
8a5a8d6c | 1056 | cmplen = name_cmp_len(name); |
5379a5c5 JH |
1057 | for (it = pbase_tree; it; it = it->next) { |
1058 | if (cmplen == 0) { | |
bc32fed5 | 1059 | add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1); |
5379a5c5 JH |
1060 | } |
1061 | else { | |
1062 | struct tree_desc tree; | |
6fda5e51 | 1063 | init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size); |
ce0bd642 | 1064 | add_pbase_object(&tree, name, cmplen, name); |
7a979d99 | 1065 | } |
3f9ac8d2 | 1066 | } |
3f9ac8d2 JH |
1067 | } |
1068 | ||
7a979d99 | 1069 | static void add_preferred_base(unsigned char *sha1) |
3f9ac8d2 | 1070 | { |
5379a5c5 JH |
1071 | struct pbase_tree *it; |
1072 | void *data; | |
1073 | unsigned long size; | |
1074 | unsigned char tree_sha1[20]; | |
1d6b38cc | 1075 | |
8d1d8f83 JH |
1076 | if (window <= num_preferred_base++) |
1077 | return; | |
1078 | ||
5379a5c5 JH |
1079 | data = read_object_with_reference(sha1, tree_type, &size, tree_sha1); |
1080 | if (!data) | |
7a979d99 | 1081 | return; |
5379a5c5 JH |
1082 | |
1083 | for (it = pbase_tree; it; it = it->next) { | |
a89fccd2 | 1084 | if (!hashcmp(it->pcache.sha1, tree_sha1)) { |
5379a5c5 JH |
1085 | free(data); |
1086 | return; | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | it = xcalloc(1, sizeof(*it)); | |
1091 | it->next = pbase_tree; | |
1092 | pbase_tree = it; | |
1093 | ||
e702496e | 1094 | hashcpy(it->pcache.sha1, tree_sha1); |
5379a5c5 JH |
1095 | it->pcache.tree_data = data; |
1096 | it->pcache.tree_size = size; | |
3f9ac8d2 JH |
1097 | } |
1098 | ||
0ef95f72 NP |
1099 | static void cleanup_preferred_base(void) |
1100 | { | |
1101 | struct pbase_tree *it; | |
1102 | unsigned i; | |
1103 | ||
1104 | it = pbase_tree; | |
1105 | pbase_tree = NULL; | |
1106 | while (it) { | |
1107 | struct pbase_tree *this = it; | |
1108 | it = this->next; | |
1109 | free(this->pcache.tree_data); | |
1110 | free(this); | |
1111 | } | |
1112 | ||
1113 | for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) { | |
1114 | if (!pbase_tree_cache[i]) | |
1115 | continue; | |
1116 | free(pbase_tree_cache[i]->tree_data); | |
1117 | free(pbase_tree_cache[i]); | |
1118 | pbase_tree_cache[i] = NULL; | |
1119 | } | |
1120 | ||
1121 | free(done_pbase_paths); | |
1122 | done_pbase_paths = NULL; | |
1123 | done_pbase_paths_num = done_pbase_paths_alloc = 0; | |
1124 | } | |
1125 | ||
c323ac7d LT |
1126 | static void check_object(struct object_entry *entry) |
1127 | { | |
5c49c116 | 1128 | if (entry->in_pack) { |
780e6e73 | 1129 | struct packed_git *p = entry->in_pack; |
03e79c88 | 1130 | struct pack_window *w_curs = NULL; |
5c49c116 NP |
1131 | const unsigned char *base_ref = NULL; |
1132 | struct object_entry *base_entry; | |
1133 | unsigned long used, used_0; | |
ef49a7a0 | 1134 | unsigned long avail; |
5c49c116 NP |
1135 | off_t ofs; |
1136 | unsigned char *buf, c; | |
780e6e73 | 1137 | |
6777a59f | 1138 | buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail); |
ab7cd7bb | 1139 | |
5c49c116 | 1140 | /* |
fa736f72 NP |
1141 | * We want in_pack_type even if we do not reuse delta |
1142 | * since non-delta representations could still be reused. | |
ab7cd7bb | 1143 | */ |
09ded04b | 1144 | used = unpack_object_header_buffer(buf, avail, |
5c49c116 NP |
1145 | &entry->in_pack_type, |
1146 | &entry->size); | |
03d66015 NP |
1147 | if (used == 0) |
1148 | goto give_up; | |
ab7cd7bb | 1149 | |
5c49c116 NP |
1150 | /* |
1151 | * Determine if this is a delta and if so whether we can | |
1152 | * reuse it or not. Otherwise let's find out as cheaply as | |
1153 | * possible what the actual type and size for this object is. | |
3f9ac8d2 | 1154 | */ |
5c49c116 NP |
1155 | switch (entry->in_pack_type) { |
1156 | default: | |
1157 | /* Not a delta hence we've already got all we need. */ | |
1158 | entry->type = entry->in_pack_type; | |
1159 | entry->in_pack_header_size = used; | |
03d66015 NP |
1160 | if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB) |
1161 | goto give_up; | |
5c49c116 NP |
1162 | unuse_pack(&w_curs); |
1163 | return; | |
1164 | case OBJ_REF_DELTA: | |
a7de7130 | 1165 | if (reuse_delta && !entry->preferred_base) |
5c49c116 NP |
1166 | base_ref = use_pack(p, &w_curs, |
1167 | entry->in_pack_offset + used, NULL); | |
1168 | entry->in_pack_header_size = used + 20; | |
1169 | break; | |
1170 | case OBJ_OFS_DELTA: | |
1171 | buf = use_pack(p, &w_curs, | |
1172 | entry->in_pack_offset + used, NULL); | |
1173 | used_0 = 0; | |
1174 | c = buf[used_0++]; | |
1175 | ofs = c & 127; | |
1176 | while (c & 128) { | |
1177 | ofs += 1; | |
03d66015 NP |
1178 | if (!ofs || MSB(ofs, 7)) { |
1179 | error("delta base offset overflow in pack for %s", | |
1180 | sha1_to_hex(entry->idx.sha1)); | |
1181 | goto give_up; | |
1182 | } | |
5c49c116 NP |
1183 | c = buf[used_0++]; |
1184 | ofs = (ofs << 7) + (c & 127); | |
780e6e73 | 1185 | } |
5c49c116 | 1186 | ofs = entry->in_pack_offset - ofs; |
03d66015 NP |
1187 | if (ofs <= 0 || ofs >= entry->in_pack_offset) { |
1188 | error("delta base offset out of bound for %s", | |
1189 | sha1_to_hex(entry->idx.sha1)); | |
1190 | goto give_up; | |
1191 | } | |
a7de7130 | 1192 | if (reuse_delta && !entry->preferred_base) { |
3449f8c4 NP |
1193 | struct revindex_entry *revidx; |
1194 | revidx = find_pack_revindex(p, ofs); | |
08698b1e NP |
1195 | if (!revidx) |
1196 | goto give_up; | |
3449f8c4 NP |
1197 | base_ref = nth_packed_object_sha1(p, revidx->nr); |
1198 | } | |
5c49c116 NP |
1199 | entry->in_pack_header_size = used + used_0; |
1200 | break; | |
780e6e73 | 1201 | } |
780e6e73 | 1202 | |
5c49c116 NP |
1203 | if (base_ref && (base_entry = locate_object_entry(base_ref))) { |
1204 | /* | |
1205 | * If base_ref was set above that means we wish to | |
1206 | * reuse delta data, and we even found that base | |
1207 | * in the list of objects we want to pack. Goodie! | |
1208 | * | |
1209 | * Depth value does not matter - find_deltas() will | |
1210 | * never consider reused delta as the base object to | |
1211 | * deltify other objects against, in order to avoid | |
1212 | * circular deltas. | |
3f9ac8d2 | 1213 | */ |
780e6e73 | 1214 | entry->type = entry->in_pack_type; |
5c49c116 | 1215 | entry->delta = base_entry; |
64bd76b1 | 1216 | entry->delta_size = entry->size; |
15b4d577 JH |
1217 | entry->delta_sibling = base_entry->delta_child; |
1218 | base_entry->delta_child = entry; | |
5c49c116 NP |
1219 | unuse_pack(&w_curs); |
1220 | return; | |
1221 | } | |
ab7cd7bb | 1222 | |
5c49c116 NP |
1223 | if (entry->type) { |
1224 | /* | |
1225 | * This must be a delta and we already know what the | |
1226 | * final object type is. Let's extract the actual | |
1227 | * object size from the delta header. | |
1228 | */ | |
1229 | entry->size = get_size_from_delta(p, &w_curs, | |
1230 | entry->in_pack_offset + entry->in_pack_header_size); | |
03d66015 NP |
1231 | if (entry->size == 0) |
1232 | goto give_up; | |
5c49c116 | 1233 | unuse_pack(&w_curs); |
3f9ac8d2 JH |
1234 | return; |
1235 | } | |
5c49c116 NP |
1236 | |
1237 | /* | |
1238 | * No choice but to fall back to the recursive delta walk | |
1239 | * with sha1_object_info() to find about the object type | |
1240 | * at this point... | |
1241 | */ | |
03d66015 | 1242 | give_up: |
5c49c116 | 1243 | unuse_pack(&w_curs); |
36e4d74a | 1244 | } |
3f9ac8d2 | 1245 | |
aa7e44bf | 1246 | entry->type = sha1_object_info(entry->idx.sha1, &entry->size); |
6d6f9cdd SP |
1247 | /* |
1248 | * The error condition is checked in prepare_pack(). This is | |
1249 | * to permit a missing preferred base object to be ignored | |
1250 | * as a preferred base. Doing so can result in a larger | |
1251 | * pack file, but the transfer will still take place. | |
1252 | */ | |
3f9ac8d2 JH |
1253 | } |
1254 | ||
5c49c116 NP |
1255 | static int pack_offset_sort(const void *_a, const void *_b) |
1256 | { | |
1257 | const struct object_entry *a = *(struct object_entry **)_a; | |
1258 | const struct object_entry *b = *(struct object_entry **)_b; | |
1259 | ||
1260 | /* avoid filesystem trashing with loose objects */ | |
1261 | if (!a->in_pack && !b->in_pack) | |
aa7e44bf | 1262 | return hashcmp(a->idx.sha1, b->idx.sha1); |
5c49c116 NP |
1263 | |
1264 | if (a->in_pack < b->in_pack) | |
1265 | return -1; | |
1266 | if (a->in_pack > b->in_pack) | |
1267 | return 1; | |
1268 | return a->in_pack_offset < b->in_pack_offset ? -1 : | |
1269 | (a->in_pack_offset > b->in_pack_offset); | |
1270 | } | |
1271 | ||
c323ac7d LT |
1272 | static void get_object_details(void) |
1273 | { | |
7cadf491 | 1274 | uint32_t i; |
5c49c116 NP |
1275 | struct object_entry **sorted_by_offset; |
1276 | ||
1277 | sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *)); | |
1278 | for (i = 0; i < nr_objects; i++) | |
1279 | sorted_by_offset[i] = objects + i; | |
1280 | qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort); | |
c323ac7d | 1281 | |
15366280 JH |
1282 | for (i = 0; i < nr_objects; i++) { |
1283 | struct object_entry *entry = sorted_by_offset[i]; | |
1284 | check_object(entry); | |
1285 | if (big_file_threshold <= entry->size) | |
1286 | entry->no_try_delta = 1; | |
1287 | } | |
3449f8c4 | 1288 | |
5c49c116 | 1289 | free(sorted_by_offset); |
c323ac7d LT |
1290 | } |
1291 | ||
b904166c NP |
1292 | /* |
1293 | * We search for deltas in a list sorted by type, by filename hash, and then | |
1294 | * by size, so that we see progressively smaller and smaller files. | |
1295 | * That's because we prefer deltas to be from the bigger file | |
1296 | * to the smaller -- deletes are potentially cheaper, but perhaps | |
1297 | * more importantly, the bigger file is likely the more recent | |
1298 | * one. The deepest deltas are therefore the oldest objects which are | |
1299 | * less susceptible to be accessed often. | |
1300 | */ | |
9668cf59 | 1301 | static int type_size_sort(const void *_a, const void *_b) |
c323ac7d | 1302 | { |
9668cf59 NP |
1303 | const struct object_entry *a = *(struct object_entry **)_a; |
1304 | const struct object_entry *b = *(struct object_entry **)_b; | |
1305 | ||
c323ac7d | 1306 | if (a->type > b->type) |
27225f2e | 1307 | return -1; |
b904166c | 1308 | if (a->type < b->type) |
27225f2e | 1309 | return 1; |
b904166c | 1310 | if (a->hash > b->hash) |
7a979d99 | 1311 | return -1; |
b904166c | 1312 | if (a->hash < b->hash) |
7a979d99 | 1313 | return 1; |
b904166c | 1314 | if (a->preferred_base > b->preferred_base) |
c323ac7d | 1315 | return -1; |
b904166c NP |
1316 | if (a->preferred_base < b->preferred_base) |
1317 | return 1; | |
c323ac7d | 1318 | if (a->size > b->size) |
b904166c NP |
1319 | return -1; |
1320 | if (a->size < b->size) | |
c323ac7d | 1321 | return 1; |
b904166c | 1322 | return a < b ? -1 : (a > b); /* newest first */ |
c323ac7d LT |
1323 | } |
1324 | ||
1325 | struct unpacked { | |
1326 | struct object_entry *entry; | |
1327 | void *data; | |
f6c7081a | 1328 | struct delta_index *index; |
5a235b5e | 1329 | unsigned depth; |
c323ac7d LT |
1330 | }; |
1331 | ||
d250626c NP |
1332 | static int delta_cacheable(unsigned long src_size, unsigned long trg_size, |
1333 | unsigned long delta_size) | |
074b2eea MK |
1334 | { |
1335 | if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size) | |
1336 | return 0; | |
1337 | ||
e3dfddb3 MK |
1338 | if (delta_size < cache_max_small_delta_size) |
1339 | return 1; | |
1340 | ||
074b2eea MK |
1341 | /* cache delta, if objects are large enough compared to delta size */ |
1342 | if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10)) | |
1343 | return 1; | |
1344 | ||
1345 | return 0; | |
1346 | } | |
1347 | ||
7eb151d6 | 1348 | #ifndef NO_PTHREADS |
8ecce684 | 1349 | |
44626dc7 | 1350 | static pthread_mutex_t read_mutex; |
8ecce684 NP |
1351 | #define read_lock() pthread_mutex_lock(&read_mutex) |
1352 | #define read_unlock() pthread_mutex_unlock(&read_mutex) | |
1353 | ||
44626dc7 | 1354 | static pthread_mutex_t cache_mutex; |
3c701839 NP |
1355 | #define cache_lock() pthread_mutex_lock(&cache_mutex) |
1356 | #define cache_unlock() pthread_mutex_unlock(&cache_mutex) | |
1357 | ||
44626dc7 | 1358 | static pthread_mutex_t progress_mutex; |
8ecce684 NP |
1359 | #define progress_lock() pthread_mutex_lock(&progress_mutex) |
1360 | #define progress_unlock() pthread_mutex_unlock(&progress_mutex) | |
1361 | ||
1362 | #else | |
1363 | ||
e1ef8673 JH |
1364 | #define read_lock() (void)0 |
1365 | #define read_unlock() (void)0 | |
1366 | #define cache_lock() (void)0 | |
1367 | #define cache_unlock() (void)0 | |
1368 | #define progress_lock() (void)0 | |
1369 | #define progress_unlock() (void)0 | |
8ecce684 NP |
1370 | |
1371 | #endif | |
1372 | ||
f6c7081a | 1373 | static int try_delta(struct unpacked *trg, struct unpacked *src, |
ef0316fc | 1374 | unsigned max_depth, unsigned long *mem_usage) |
c323ac7d | 1375 | { |
f6c7081a NP |
1376 | struct object_entry *trg_entry = trg->entry; |
1377 | struct object_entry *src_entry = src->entry; | |
560b25a8 | 1378 | unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz; |
c83f032e | 1379 | unsigned ref_depth; |
21666f1a | 1380 | enum object_type type; |
c323ac7d LT |
1381 | void *delta_buf; |
1382 | ||
1383 | /* Don't bother doing diffs between different types */ | |
f6c7081a | 1384 | if (trg_entry->type != src_entry->type) |
c323ac7d LT |
1385 | return -1; |
1386 | ||
51d1e83f LT |
1387 | /* |
1388 | * We do not bother to try a delta that we discarded | |
8dbbd14e | 1389 | * on an earlier try, but only when reusing delta data. |
51d1e83f | 1390 | */ |
a7de7130 | 1391 | if (reuse_delta && trg_entry->in_pack && |
e9195b58 JH |
1392 | trg_entry->in_pack == src_entry->in_pack && |
1393 | trg_entry->in_pack_type != OBJ_REF_DELTA && | |
1394 | trg_entry->in_pack_type != OBJ_OFS_DELTA) | |
51d1e83f LT |
1395 | return 0; |
1396 | ||
898b14ce | 1397 | /* Let's not bust the allowed depth. */ |
5a235b5e | 1398 | if (src->depth >= max_depth) |
d116a45a | 1399 | return 0; |
c323ac7d | 1400 | |
c3b06a69 | 1401 | /* Now some size filtering heuristics. */ |
560b25a8 | 1402 | trg_size = trg_entry->size; |
c83f032e NP |
1403 | if (!trg_entry->delta) { |
1404 | max_size = trg_size/2 - 20; | |
1405 | ref_depth = 1; | |
1406 | } else { | |
1407 | max_size = trg_entry->delta_size; | |
5a235b5e | 1408 | ref_depth = trg->depth; |
c83f032e | 1409 | } |
720fe22d | 1410 | max_size = (uint64_t)max_size * (max_depth - src->depth) / |
c83f032e | 1411 | (max_depth - ref_depth + 1); |
c3b06a69 NP |
1412 | if (max_size == 0) |
1413 | return 0; | |
f6c7081a | 1414 | src_size = src_entry->size; |
560b25a8 | 1415 | sizediff = src_size < trg_size ? trg_size - src_size : 0; |
27225f2e | 1416 | if (sizediff >= max_size) |
f527cb8c | 1417 | return 0; |
a1dab41a BD |
1418 | if (trg_size < src_size / 32) |
1419 | return 0; | |
f6c7081a | 1420 | |
560b25a8 NP |
1421 | /* Load data if not already done */ |
1422 | if (!trg->data) { | |
8ecce684 | 1423 | read_lock(); |
aa7e44bf | 1424 | trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz); |
8ecce684 | 1425 | read_unlock(); |
2e3404c3 JH |
1426 | if (!trg->data) |
1427 | die("object %s cannot be read", | |
1428 | sha1_to_hex(trg_entry->idx.sha1)); | |
560b25a8 NP |
1429 | if (sz != trg_size) |
1430 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1431 | sha1_to_hex(trg_entry->idx.sha1), sz, trg_size); |
ef0316fc | 1432 | *mem_usage += sz; |
560b25a8 NP |
1433 | } |
1434 | if (!src->data) { | |
8ecce684 | 1435 | read_lock(); |
aa7e44bf | 1436 | src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz); |
8ecce684 | 1437 | read_unlock(); |
71064a95 NP |
1438 | if (!src->data) { |
1439 | if (src_entry->preferred_base) { | |
1440 | static int warned = 0; | |
1441 | if (!warned++) | |
1442 | warning("object %s cannot be read", | |
1443 | sha1_to_hex(src_entry->idx.sha1)); | |
1444 | /* | |
1445 | * Those objects are not included in the | |
1446 | * resulting pack. Be resilient and ignore | |
1447 | * them if they can't be read, in case the | |
1448 | * pack could be created nevertheless. | |
1449 | */ | |
1450 | return 0; | |
1451 | } | |
2e3404c3 JH |
1452 | die("object %s cannot be read", |
1453 | sha1_to_hex(src_entry->idx.sha1)); | |
71064a95 | 1454 | } |
560b25a8 NP |
1455 | if (sz != src_size) |
1456 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1457 | sha1_to_hex(src_entry->idx.sha1), sz, src_size); |
ef0316fc | 1458 | *mem_usage += sz; |
560b25a8 NP |
1459 | } |
1460 | if (!src->index) { | |
1461 | src->index = create_delta_index(src->data, src_size); | |
a588d88a MK |
1462 | if (!src->index) { |
1463 | static int warned = 0; | |
1464 | if (!warned++) | |
1465 | warning("suboptimal pack - out of memory"); | |
1466 | return 0; | |
1467 | } | |
ef0316fc | 1468 | *mem_usage += sizeof_delta_index(src->index); |
560b25a8 NP |
1469 | } |
1470 | ||
1471 | delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size); | |
c323ac7d | 1472 | if (!delta_buf) |
75c42d8c | 1473 | return 0; |
f6c7081a | 1474 | |
9e2d57a0 | 1475 | if (trg_entry->delta) { |
848d732c BD |
1476 | /* Prefer only shallower same-sized deltas. */ |
1477 | if (delta_size == trg_entry->delta_size && | |
5a235b5e | 1478 | src->depth + 1 >= trg->depth) { |
848d732c BD |
1479 | free(delta_buf); |
1480 | return 0; | |
1481 | } | |
074b2eea | 1482 | } |
9e2d57a0 | 1483 | |
3c701839 NP |
1484 | /* |
1485 | * Handle memory allocation outside of the cache | |
1486 | * accounting lock. Compiler will optimize the strangeness | |
7eb151d6 | 1487 | * away when NO_PTHREADS is defined. |
3c701839 | 1488 | */ |
8e0f7003 | 1489 | free(trg_entry->delta_data); |
3c701839 | 1490 | cache_lock(); |
9e2d57a0 NP |
1491 | if (trg_entry->delta_data) { |
1492 | delta_cache_size -= trg_entry->delta_size; | |
9e2d57a0 NP |
1493 | trg_entry->delta_data = NULL; |
1494 | } | |
d250626c | 1495 | if (delta_cacheable(src_size, trg_size, delta_size)) { |
b7a28f78 | 1496 | delta_cache_size += delta_size; |
3c701839 NP |
1497 | cache_unlock(); |
1498 | trg_entry->delta_data = xrealloc(delta_buf, delta_size); | |
1499 | } else { | |
1500 | cache_unlock(); | |
074b2eea | 1501 | free(delta_buf); |
3c701839 NP |
1502 | } |
1503 | ||
b7a28f78 NP |
1504 | trg_entry->delta = src_entry; |
1505 | trg_entry->delta_size = delta_size; | |
1506 | trg->depth = src->depth + 1; | |
1507 | ||
f6c7081a | 1508 | return 1; |
c323ac7d LT |
1509 | } |
1510 | ||
898b14ce | 1511 | static unsigned int check_delta_limit(struct object_entry *me, unsigned int n) |
b2504a0d | 1512 | { |
898b14ce NP |
1513 | struct object_entry *child = me->delta_child; |
1514 | unsigned int m = n; | |
1515 | while (child) { | |
1516 | unsigned int c = check_delta_limit(child, n + 1); | |
1517 | if (m < c) | |
1518 | m = c; | |
1519 | child = child->delta_sibling; | |
1520 | } | |
1521 | return m; | |
b2504a0d NP |
1522 | } |
1523 | ||
75ad235c | 1524 | static unsigned long free_unpacked(struct unpacked *n) |
a97773ce | 1525 | { |
ef0316fc | 1526 | unsigned long freed_mem = sizeof_delta_index(n->index); |
a97773ce BD |
1527 | free_delta_index(n->index); |
1528 | n->index = NULL; | |
1529 | if (n->data) { | |
ef0316fc | 1530 | freed_mem += n->entry->size; |
a97773ce BD |
1531 | free(n->data); |
1532 | n->data = NULL; | |
a97773ce BD |
1533 | } |
1534 | n->entry = NULL; | |
7d7baa5e | 1535 | n->depth = 0; |
ef0316fc | 1536 | return freed_mem; |
a97773ce BD |
1537 | } |
1538 | ||
384b32c0 | 1539 | static void find_deltas(struct object_entry **list, unsigned *list_size, |
e334977d | 1540 | int window, int depth, unsigned *processed) |
c323ac7d | 1541 | { |
384b32c0 | 1542 | uint32_t i, idx = 0, count = 0; |
7cadf491 | 1543 | struct unpacked *array; |
ef0316fc | 1544 | unsigned long mem_usage = 0; |
c323ac7d | 1545 | |
19d4b416 | 1546 | array = xcalloc(window, sizeof(struct unpacked)); |
21fcd1bd | 1547 | |
384b32c0 | 1548 | for (;;) { |
421b488a | 1549 | struct object_entry *entry; |
c323ac7d | 1550 | struct unpacked *n = array + idx; |
ef0316fc | 1551 | int j, max_depth, best_base = -1; |
c323ac7d | 1552 | |
384b32c0 NP |
1553 | progress_lock(); |
1554 | if (!*list_size) { | |
1555 | progress_unlock(); | |
1556 | break; | |
1557 | } | |
421b488a | 1558 | entry = *list++; |
384b32c0 NP |
1559 | (*list_size)--; |
1560 | if (!entry->preferred_base) { | |
1561 | (*processed)++; | |
1562 | display_progress(progress_state, *processed); | |
1563 | } | |
1564 | progress_unlock(); | |
1565 | ||
ef0316fc | 1566 | mem_usage -= free_unpacked(n); |
c323ac7d | 1567 | n->entry = entry; |
ab7cd7bb | 1568 | |
a97773ce | 1569 | while (window_memory_limit && |
ef0316fc | 1570 | mem_usage > window_memory_limit && |
a97773ce BD |
1571 | count > 1) { |
1572 | uint32_t tail = (idx + window - count) % window; | |
75ad235c | 1573 | mem_usage -= free_unpacked(array + tail); |
a97773ce BD |
1574 | count--; |
1575 | } | |
1576 | ||
75d39853 NP |
1577 | /* We do not compute delta to *create* objects we are not |
1578 | * going to pack. | |
1579 | */ | |
1580 | if (entry->preferred_base) | |
1581 | goto next; | |
1582 | ||
898b14ce NP |
1583 | /* |
1584 | * If the current object is at pack edge, take the depth the | |
1585 | * objects that depend on the current object into account | |
1586 | * otherwise they would become too deep. | |
1587 | */ | |
1588 | max_depth = depth; | |
1589 | if (entry->delta_child) { | |
1590 | max_depth -= check_delta_limit(entry, 0); | |
1591 | if (max_depth <= 0) | |
1592 | goto next; | |
1593 | } | |
1594 | ||
78817c15 LT |
1595 | j = window; |
1596 | while (--j > 0) { | |
77639870 | 1597 | int ret; |
7cadf491 | 1598 | uint32_t other_idx = idx + j; |
c323ac7d | 1599 | struct unpacked *m; |
78817c15 LT |
1600 | if (other_idx >= window) |
1601 | other_idx -= window; | |
c323ac7d LT |
1602 | m = array + other_idx; |
1603 | if (!m->entry) | |
1604 | break; | |
ef0316fc | 1605 | ret = try_delta(n, m, max_depth, &mem_usage); |
77639870 | 1606 | if (ret < 0) |
c323ac7d | 1607 | break; |
77639870 JH |
1608 | else if (ret > 0) |
1609 | best_base = other_idx; | |
c323ac7d | 1610 | } |
898b14ce | 1611 | |
ed4a9031 NP |
1612 | /* |
1613 | * If we decided to cache the delta data, then it is best | |
1614 | * to compress it right away. First because we have to do | |
1615 | * it anyway, and doing it here while we're threaded will | |
1616 | * save a lot of time in the non threaded write phase, | |
1617 | * as well as allow for caching more deltas within | |
1618 | * the same cache size limit. | |
1619 | * ... | |
1620 | * But only if not writing to stdout, since in that case | |
1621 | * the network is most likely throttling writes anyway, | |
1622 | * and therefore it is best to go to the write phase ASAP | |
1623 | * instead, as we can afford spending more time compressing | |
1624 | * between writes at that moment. | |
1625 | */ | |
1626 | if (entry->delta_data && !pack_to_stdout) { | |
1627 | entry->z_delta_size = do_compress(&entry->delta_data, | |
1628 | entry->delta_size); | |
1629 | cache_lock(); | |
1630 | delta_cache_size -= entry->delta_size; | |
1631 | delta_cache_size += entry->z_delta_size; | |
1632 | cache_unlock(); | |
1633 | } | |
1634 | ||
70ca1a3f JH |
1635 | /* if we made n a delta, and if n is already at max |
1636 | * depth, leaving it in the window is pointless. we | |
1637 | * should evict it first. | |
70ca1a3f | 1638 | */ |
70baf5d4 | 1639 | if (entry->delta && max_depth <= n->depth) |
70ca1a3f | 1640 | continue; |
ff45715c | 1641 | |
77639870 JH |
1642 | /* |
1643 | * Move the best delta base up in the window, after the | |
1644 | * currently deltified object, to keep it longer. It will | |
1645 | * be the first base object to be attempted next. | |
1646 | */ | |
1647 | if (entry->delta) { | |
1648 | struct unpacked swap = array[best_base]; | |
1649 | int dist = (window + idx - best_base) % window; | |
1650 | int dst = best_base; | |
1651 | while (dist--) { | |
1652 | int src = (dst + 1) % window; | |
1653 | array[dst] = array[src]; | |
1654 | dst = src; | |
1655 | } | |
1656 | array[dst] = swap; | |
1657 | } | |
1658 | ||
898b14ce | 1659 | next: |
521a4f4c | 1660 | idx++; |
a97773ce BD |
1661 | if (count + 1 < window) |
1662 | count++; | |
521a4f4c LT |
1663 | if (idx >= window) |
1664 | idx = 0; | |
384b32c0 | 1665 | } |
adee7bdf | 1666 | |
f6c7081a | 1667 | for (i = 0; i < window; ++i) { |
ff45715c | 1668 | free_delta_index(array[i].index); |
adee7bdf | 1669 | free(array[i].data); |
f6c7081a | 1670 | } |
adee7bdf | 1671 | free(array); |
c323ac7d LT |
1672 | } |
1673 | ||
7eb151d6 | 1674 | #ifndef NO_PTHREADS |
8ecce684 | 1675 | |
a9a74636 NP |
1676 | static void try_to_free_from_threads(size_t size) |
1677 | { | |
1678 | read_lock(); | |
1679 | release_pack_memory(size, -1); | |
1680 | read_unlock(); | |
1681 | } | |
1682 | ||
bc9b2175 | 1683 | static try_to_free_t old_try_to_free_routine; |
851c34b0 | 1684 | |
50f22ada JS |
1685 | /* |
1686 | * The main thread waits on the condition that (at least) one of the workers | |
1687 | * has stopped working (which is indicated in the .working member of | |
1688 | * struct thread_params). | |
1689 | * When a work thread has completed its work, it sets .working to 0 and | |
1690 | * signals the main thread and waits on the condition that .data_ready | |
1691 | * becomes 1. | |
1692 | */ | |
1693 | ||
8ecce684 NP |
1694 | struct thread_params { |
1695 | pthread_t thread; | |
1696 | struct object_entry **list; | |
1697 | unsigned list_size; | |
384b32c0 | 1698 | unsigned remaining; |
8ecce684 NP |
1699 | int window; |
1700 | int depth; | |
50f22ada JS |
1701 | int working; |
1702 | int data_ready; | |
1703 | pthread_mutex_t mutex; | |
1704 | pthread_cond_t cond; | |
8ecce684 NP |
1705 | unsigned *processed; |
1706 | }; | |
1707 | ||
44626dc7 AH |
1708 | static pthread_cond_t progress_cond; |
1709 | ||
1710 | /* | |
1711 | * Mutex and conditional variable can't be statically-initialized on Windows. | |
1712 | */ | |
1713 | static void init_threaded_search(void) | |
1714 | { | |
93749194 | 1715 | init_recursive_mutex(&read_mutex); |
44626dc7 AH |
1716 | pthread_mutex_init(&cache_mutex, NULL); |
1717 | pthread_mutex_init(&progress_mutex, NULL); | |
1718 | pthread_cond_init(&progress_cond, NULL); | |
851c34b0 | 1719 | old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads); |
44626dc7 AH |
1720 | } |
1721 | ||
1722 | static void cleanup_threaded_search(void) | |
1723 | { | |
851c34b0 | 1724 | set_try_to_free_routine(old_try_to_free_routine); |
44626dc7 AH |
1725 | pthread_cond_destroy(&progress_cond); |
1726 | pthread_mutex_destroy(&read_mutex); | |
1727 | pthread_mutex_destroy(&cache_mutex); | |
1728 | pthread_mutex_destroy(&progress_mutex); | |
1729 | } | |
c2a33679 | 1730 | |
8ecce684 NP |
1731 | static void *threaded_find_deltas(void *arg) |
1732 | { | |
c2a33679 NP |
1733 | struct thread_params *me = arg; |
1734 | ||
50f22ada | 1735 | while (me->remaining) { |
384b32c0 | 1736 | find_deltas(me->list, &me->remaining, |
c2a33679 | 1737 | me->window, me->depth, me->processed); |
50f22ada JS |
1738 | |
1739 | progress_lock(); | |
1740 | me->working = 0; | |
1741 | pthread_cond_signal(&progress_cond); | |
1742 | progress_unlock(); | |
1743 | ||
1744 | /* | |
1745 | * We must not set ->data_ready before we wait on the | |
1746 | * condition because the main thread may have set it to 1 | |
1747 | * before we get here. In order to be sure that new | |
1748 | * work is available if we see 1 in ->data_ready, it | |
1749 | * was initialized to 0 before this thread was spawned | |
1750 | * and we reset it to 0 right away. | |
1751 | */ | |
1752 | pthread_mutex_lock(&me->mutex); | |
1753 | while (!me->data_ready) | |
1754 | pthread_cond_wait(&me->cond, &me->mutex); | |
1755 | me->data_ready = 0; | |
1756 | pthread_mutex_unlock(&me->mutex); | |
c2a33679 | 1757 | } |
50f22ada JS |
1758 | /* leave ->working 1 so that this doesn't get more work assigned */ |
1759 | return NULL; | |
8ecce684 NP |
1760 | } |
1761 | ||
8ecce684 NP |
1762 | static void ll_find_deltas(struct object_entry **list, unsigned list_size, |
1763 | int window, int depth, unsigned *processed) | |
1764 | { | |
dcda3614 | 1765 | struct thread_params *p; |
384b32c0 | 1766 | int i, ret, active_threads = 0; |
c2a33679 | 1767 | |
44626dc7 AH |
1768 | init_threaded_search(); |
1769 | ||
ef055571 NP |
1770 | if (!delta_search_threads) /* --threads=0 means autodetect */ |
1771 | delta_search_threads = online_cpus(); | |
367f4a43 | 1772 | if (delta_search_threads <= 1) { |
384b32c0 | 1773 | find_deltas(list, &list_size, window, depth, processed); |
44626dc7 | 1774 | cleanup_threaded_search(); |
367f4a43 NP |
1775 | return; |
1776 | } | |
43cc2b42 | 1777 | if (progress > pack_to_stdout) |
b6c29915 | 1778 | fprintf(stderr, "Delta compression using up to %d threads.\n", |
43cc2b42 | 1779 | delta_search_threads); |
dcda3614 | 1780 | p = xcalloc(delta_search_threads, sizeof(*p)); |
367f4a43 | 1781 | |
50f22ada | 1782 | /* Partition the work amongst work threads. */ |
367f4a43 | 1783 | for (i = 0; i < delta_search_threads; i++) { |
50f22ada JS |
1784 | unsigned sub_size = list_size / (delta_search_threads - i); |
1785 | ||
bf874896 NP |
1786 | /* don't use too small segments or no deltas will be found */ |
1787 | if (sub_size < 2*window && i+1 < delta_search_threads) | |
1788 | sub_size = 0; | |
1789 | ||
8ecce684 NP |
1790 | p[i].window = window; |
1791 | p[i].depth = depth; | |
1792 | p[i].processed = processed; | |
50f22ada JS |
1793 | p[i].working = 1; |
1794 | p[i].data_ready = 0; | |
c2a33679 | 1795 | |
59921b4b | 1796 | /* try to split chunks on "path" boundaries */ |
6fc74703 NP |
1797 | while (sub_size && sub_size < list_size && |
1798 | list[sub_size]->hash && | |
384b32c0 NP |
1799 | list[sub_size]->hash == list[sub_size-1]->hash) |
1800 | sub_size++; | |
1801 | ||
50f22ada JS |
1802 | p[i].list = list; |
1803 | p[i].list_size = sub_size; | |
1804 | p[i].remaining = sub_size; | |
59921b4b | 1805 | |
384b32c0 NP |
1806 | list += sub_size; |
1807 | list_size -= sub_size; | |
1808 | } | |
1809 | ||
50f22ada JS |
1810 | /* Start work threads. */ |
1811 | for (i = 0; i < delta_search_threads; i++) { | |
1812 | if (!p[i].list_size) | |
1813 | continue; | |
68e6a4f8 JS |
1814 | pthread_mutex_init(&p[i].mutex, NULL); |
1815 | pthread_cond_init(&p[i].cond, NULL); | |
50f22ada JS |
1816 | ret = pthread_create(&p[i].thread, NULL, |
1817 | threaded_find_deltas, &p[i]); | |
1818 | if (ret) | |
1819 | die("unable to create thread: %s", strerror(ret)); | |
1820 | active_threads++; | |
1821 | } | |
1822 | ||
384b32c0 NP |
1823 | /* |
1824 | * Now let's wait for work completion. Each time a thread is done | |
1825 | * with its work, we steal half of the remaining work from the | |
1826 | * thread with the largest number of unprocessed objects and give | |
1827 | * it to that newly idle thread. This ensure good load balancing | |
1828 | * until the remaining object list segments are simply too short | |
1829 | * to be worth splitting anymore. | |
1830 | */ | |
50f22ada JS |
1831 | while (active_threads) { |
1832 | struct thread_params *target = NULL; | |
384b32c0 NP |
1833 | struct thread_params *victim = NULL; |
1834 | unsigned sub_size = 0; | |
384b32c0 NP |
1835 | |
1836 | progress_lock(); | |
50f22ada JS |
1837 | for (;;) { |
1838 | for (i = 0; !target && i < delta_search_threads; i++) | |
1839 | if (!p[i].working) | |
1840 | target = &p[i]; | |
1841 | if (target) | |
1842 | break; | |
1843 | pthread_cond_wait(&progress_cond, &progress_mutex); | |
1844 | } | |
1845 | ||
384b32c0 NP |
1846 | for (i = 0; i < delta_search_threads; i++) |
1847 | if (p[i].remaining > 2*window && | |
1848 | (!victim || victim->remaining < p[i].remaining)) | |
1849 | victim = &p[i]; | |
1850 | if (victim) { | |
1851 | sub_size = victim->remaining / 2; | |
1852 | list = victim->list + victim->list_size - sub_size; | |
1853 | while (sub_size && list[0]->hash && | |
1854 | list[0]->hash == list[-1]->hash) { | |
1855 | list++; | |
1856 | sub_size--; | |
1857 | } | |
eb9688ff NP |
1858 | if (!sub_size) { |
1859 | /* | |
1860 | * It is possible for some "paths" to have | |
1861 | * so many objects that no hash boundary | |
1862 | * might be found. Let's just steal the | |
1863 | * exact half in that case. | |
1864 | */ | |
1865 | sub_size = victim->remaining / 2; | |
1866 | list -= sub_size; | |
1867 | } | |
384b32c0 NP |
1868 | target->list = list; |
1869 | victim->list_size -= sub_size; | |
1870 | victim->remaining -= sub_size; | |
1871 | } | |
384b32c0 NP |
1872 | target->list_size = sub_size; |
1873 | target->remaining = sub_size; | |
50f22ada JS |
1874 | target->working = 1; |
1875 | progress_unlock(); | |
1876 | ||
1877 | pthread_mutex_lock(&target->mutex); | |
1878 | target->data_ready = 1; | |
1879 | pthread_cond_signal(&target->cond); | |
1880 | pthread_mutex_unlock(&target->mutex); | |
c2a33679 | 1881 | |
384b32c0 | 1882 | if (!sub_size) { |
b81d9af7 | 1883 | pthread_join(target->thread, NULL); |
50f22ada JS |
1884 | pthread_cond_destroy(&target->cond); |
1885 | pthread_mutex_destroy(&target->mutex); | |
384b32c0 | 1886 | active_threads--; |
c2a33679 | 1887 | } |
50f22ada | 1888 | } |
44626dc7 | 1889 | cleanup_threaded_search(); |
dcda3614 | 1890 | free(p); |
8ecce684 NP |
1891 | } |
1892 | ||
1893 | #else | |
384b32c0 | 1894 | #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p) |
8ecce684 NP |
1895 | #endif |
1896 | ||
f0a24aa5 SP |
1897 | static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data) |
1898 | { | |
1899 | unsigned char peeled[20]; | |
1900 | ||
1901 | if (!prefixcmp(path, "refs/tags/") && /* is a tag? */ | |
1902 | !peel_ref(path, peeled) && /* peelable? */ | |
1903 | !is_null_sha1(peeled) && /* annotated tag? */ | |
1904 | locate_object_entry(peeled)) /* object packed? */ | |
1905 | add_object_entry(sha1, OBJ_TAG, NULL, 0); | |
1906 | return 0; | |
1907 | } | |
1908 | ||
f3123c4a JH |
1909 | static void prepare_pack(int window, int depth) |
1910 | { | |
9668cf59 | 1911 | struct object_entry **delta_list; |
6e1c2344 RJ |
1912 | uint32_t i, nr_deltas; |
1913 | unsigned n; | |
9668cf59 | 1914 | |
3f9ac8d2 | 1915 | get_object_details(); |
9668cf59 | 1916 | |
0e8189e2 NP |
1917 | /* |
1918 | * If we're locally repacking then we need to be doubly careful | |
1919 | * from now on in order to make sure no stealth corruption gets | |
1920 | * propagated to the new pack. Clients receiving streamed packs | |
1921 | * should validate everything they get anyway so no need to incur | |
1922 | * the additional cost here in that case. | |
1923 | */ | |
1924 | if (!pack_to_stdout) | |
1925 | do_check_packed_object_crc = 1; | |
1926 | ||
75d39853 | 1927 | if (!nr_objects || !window || !depth) |
9668cf59 NP |
1928 | return; |
1929 | ||
1930 | delta_list = xmalloc(nr_objects * sizeof(*delta_list)); | |
75d39853 NP |
1931 | nr_deltas = n = 0; |
1932 | ||
1933 | for (i = 0; i < nr_objects; i++) { | |
1934 | struct object_entry *entry = objects + i; | |
1935 | ||
1936 | if (entry->delta) | |
1937 | /* This happens if we decided to reuse existing | |
a7de7130 | 1938 | * delta from a pack. "reuse_delta &&" is implied. |
75d39853 NP |
1939 | */ |
1940 | continue; | |
1941 | ||
1942 | if (entry->size < 50) | |
1943 | continue; | |
1944 | ||
1945 | if (entry->no_try_delta) | |
1946 | continue; | |
1947 | ||
6d6f9cdd | 1948 | if (!entry->preferred_base) { |
75d39853 | 1949 | nr_deltas++; |
6d6f9cdd SP |
1950 | if (entry->type < 0) |
1951 | die("unable to get type of object %s", | |
1952 | sha1_to_hex(entry->idx.sha1)); | |
eede9f42 NP |
1953 | } else { |
1954 | if (entry->type < 0) { | |
1955 | /* | |
1956 | * This object is not found, but we | |
1957 | * don't have to include it anyway. | |
1958 | */ | |
1959 | continue; | |
1960 | } | |
6d6f9cdd | 1961 | } |
75d39853 NP |
1962 | |
1963 | delta_list[n++] = entry; | |
1964 | } | |
1965 | ||
2f8b8947 | 1966 | if (nr_deltas && n > 1) { |
e334977d NP |
1967 | unsigned nr_done = 0; |
1968 | if (progress) | |
dc6a0757 NP |
1969 | progress_state = start_progress("Compressing objects", |
1970 | nr_deltas); | |
75d39853 | 1971 | qsort(delta_list, n, sizeof(*delta_list), type_size_sort); |
8ecce684 | 1972 | ll_find_deltas(delta_list, n, window+1, depth, &nr_done); |
4d4fcc54 | 1973 | stop_progress(&progress_state); |
e334977d NP |
1974 | if (nr_done != nr_deltas) |
1975 | die("inconsistency with delta count"); | |
75d39853 | 1976 | } |
9668cf59 | 1977 | free(delta_list); |
f3123c4a JH |
1978 | } |
1979 | ||
ef90d6d4 | 1980 | static int git_pack_config(const char *k, const char *v, void *cb) |
4812a93a | 1981 | { |
eeefa7c9 | 1982 | if (!strcmp(k, "pack.window")) { |
4812a93a JK |
1983 | window = git_config_int(k, v); |
1984 | return 0; | |
1985 | } | |
a97773ce BD |
1986 | if (!strcmp(k, "pack.windowmemory")) { |
1987 | window_memory_limit = git_config_ulong(k, v); | |
1988 | return 0; | |
1989 | } | |
1990 | if (!strcmp(k, "pack.depth")) { | |
842aaf93 TT |
1991 | depth = git_config_int(k, v); |
1992 | return 0; | |
1993 | } | |
960ccca6 DH |
1994 | if (!strcmp(k, "pack.compression")) { |
1995 | int level = git_config_int(k, v); | |
1996 | if (level == -1) | |
1997 | level = Z_DEFAULT_COMPRESSION; | |
1998 | else if (level < 0 || level > Z_BEST_COMPRESSION) | |
1999 | die("bad pack compression level %d", level); | |
2000 | pack_compression_level = level; | |
2001 | pack_compression_seen = 1; | |
2002 | return 0; | |
2003 | } | |
074b2eea MK |
2004 | if (!strcmp(k, "pack.deltacachesize")) { |
2005 | max_delta_cache_size = git_config_int(k, v); | |
2006 | return 0; | |
2007 | } | |
e3dfddb3 MK |
2008 | if (!strcmp(k, "pack.deltacachelimit")) { |
2009 | cache_max_small_delta_size = git_config_int(k, v); | |
2010 | return 0; | |
2011 | } | |
693b86ff NP |
2012 | if (!strcmp(k, "pack.threads")) { |
2013 | delta_search_threads = git_config_int(k, v); | |
833e3df1 | 2014 | if (delta_search_threads < 0) |
693b86ff NP |
2015 | die("invalid number of threads specified (%d)", |
2016 | delta_search_threads); | |
7eb151d6 | 2017 | #ifdef NO_PTHREADS |
833e3df1 | 2018 | if (delta_search_threads != 1) |
693b86ff NP |
2019 | warning("no threads support, ignoring %s", k); |
2020 | #endif | |
2021 | return 0; | |
2022 | } | |
4d00bda2 | 2023 | if (!strcmp(k, "pack.indexversion")) { |
ebcfb379 JH |
2024 | pack_idx_opts.version = git_config_int(k, v); |
2025 | if (pack_idx_opts.version > 2) | |
6e1c2344 | 2026 | die("bad pack.indexversion=%"PRIu32, |
ebcfb379 | 2027 | pack_idx_opts.version); |
4d00bda2 NP |
2028 | return 0; |
2029 | } | |
2b84b5a8 JS |
2030 | if (!strcmp(k, "pack.packsizelimit")) { |
2031 | pack_size_limit_cfg = git_config_ulong(k, v); | |
2032 | return 0; | |
2033 | } | |
ef90d6d4 | 2034 | return git_default_config(k, v, cb); |
4812a93a JK |
2035 | } |
2036 | ||
b5d97e6b | 2037 | static void read_object_list_from_stdin(void) |
c323ac7d | 2038 | { |
b5d97e6b JH |
2039 | char line[40 + 1 + PATH_MAX + 2]; |
2040 | unsigned char sha1[20]; | |
b2504a0d | 2041 | |
da93d12b | 2042 | for (;;) { |
da93d12b LT |
2043 | if (!fgets(line, sizeof(line), stdin)) { |
2044 | if (feof(stdin)) | |
2045 | break; | |
2046 | if (!ferror(stdin)) | |
2047 | die("fgets returned NULL, not EOF, not error!"); | |
687dd75c | 2048 | if (errno != EINTR) |
d824cbba | 2049 | die_errno("fgets"); |
687dd75c JH |
2050 | clearerr(stdin); |
2051 | continue; | |
da93d12b | 2052 | } |
7a979d99 JH |
2053 | if (line[0] == '-') { |
2054 | if (get_sha1_hex(line+1, sha1)) | |
2055 | die("expected edge sha1, got garbage:\n %s", | |
b5d97e6b | 2056 | line); |
8d1d8f83 | 2057 | add_preferred_base(sha1); |
7a979d99 | 2058 | continue; |
21fcd1bd | 2059 | } |
c323ac7d | 2060 | if (get_sha1_hex(line, sha1)) |
ef07618f | 2061 | die("expected sha1, got garbage:\n %s", line); |
b5d97e6b | 2062 | |
bc32fed5 JH |
2063 | add_preferred_base_object(line+41); |
2064 | add_object_entry(sha1, 0, line+41, 0); | |
c323ac7d | 2065 | } |
b5d97e6b JH |
2066 | } |
2067 | ||
08cdfb13 JH |
2068 | #define OBJECT_ADDED (1u<<20) |
2069 | ||
11c211fa | 2070 | static void show_commit(struct commit *commit, void *data) |
b5d97e6b | 2071 | { |
bc32fed5 | 2072 | add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0); |
08cdfb13 | 2073 | commit->object.flags |= OBJECT_ADDED; |
b5d97e6b JH |
2074 | } |
2075 | ||
cf2ab916 | 2076 | static void show_object(struct object *obj, const struct name_path *path, const char *last) |
b5d97e6b | 2077 | { |
cf2ab916 LT |
2078 | char *name = path_name(path, last); |
2079 | ||
8d2dfc49 LT |
2080 | add_preferred_base_object(name); |
2081 | add_object_entry(obj->sha1, obj->type, name, 0); | |
2082 | obj->flags |= OBJECT_ADDED; | |
2083 | ||
2084 | /* | |
2085 | * We will have generated the hash from the name, | |
2086 | * but not saved a pointer to it - we can free it | |
2087 | */ | |
2088 | free((char *)name); | |
b5d97e6b JH |
2089 | } |
2090 | ||
8d1d8f83 JH |
2091 | static void show_edge(struct commit *commit) |
2092 | { | |
2093 | add_preferred_base(commit->object.sha1); | |
2094 | } | |
2095 | ||
08cdfb13 JH |
2096 | struct in_pack_object { |
2097 | off_t offset; | |
2098 | struct object *object; | |
2099 | }; | |
2100 | ||
2101 | struct in_pack { | |
2102 | int alloc; | |
2103 | int nr; | |
2104 | struct in_pack_object *array; | |
2105 | }; | |
2106 | ||
2107 | static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack) | |
2108 | { | |
2109 | in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p); | |
2110 | in_pack->array[in_pack->nr].object = object; | |
2111 | in_pack->nr++; | |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * Compare the objects in the offset order, in order to emulate the | |
f18d244a | 2116 | * "git rev-list --objects" output that produced the pack originally. |
08cdfb13 JH |
2117 | */ |
2118 | static int ofscmp(const void *a_, const void *b_) | |
2119 | { | |
2120 | struct in_pack_object *a = (struct in_pack_object *)a_; | |
2121 | struct in_pack_object *b = (struct in_pack_object *)b_; | |
2122 | ||
2123 | if (a->offset < b->offset) | |
2124 | return -1; | |
2125 | else if (a->offset > b->offset) | |
2126 | return 1; | |
2127 | else | |
2128 | return hashcmp(a->object->sha1, b->object->sha1); | |
2129 | } | |
2130 | ||
2131 | static void add_objects_in_unpacked_packs(struct rev_info *revs) | |
2132 | { | |
2133 | struct packed_git *p; | |
2134 | struct in_pack in_pack; | |
2135 | uint32_t i; | |
2136 | ||
2137 | memset(&in_pack, 0, sizeof(in_pack)); | |
2138 | ||
2139 | for (p = packed_git; p; p = p->next) { | |
2140 | const unsigned char *sha1; | |
2141 | struct object *o; | |
2142 | ||
79bc4c71 | 2143 | if (!p->pack_local || p->pack_keep) |
08cdfb13 JH |
2144 | continue; |
2145 | if (open_pack_index(p)) | |
2146 | die("cannot open pack index"); | |
2147 | ||
2148 | ALLOC_GROW(in_pack.array, | |
2149 | in_pack.nr + p->num_objects, | |
2150 | in_pack.alloc); | |
2151 | ||
2152 | for (i = 0; i < p->num_objects; i++) { | |
2153 | sha1 = nth_packed_object_sha1(p, i); | |
2154 | o = lookup_unknown_object(sha1); | |
2155 | if (!(o->flags & OBJECT_ADDED)) | |
2156 | mark_in_pack_object(o, p, &in_pack); | |
2157 | o->flags |= OBJECT_ADDED; | |
2158 | } | |
2159 | } | |
2160 | ||
2161 | if (in_pack.nr) { | |
2162 | qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]), | |
2163 | ofscmp); | |
2164 | for (i = 0; i < in_pack.nr; i++) { | |
2165 | struct object *o = in_pack.array[i].object; | |
2166 | add_object_entry(o->sha1, o->type, "", 0); | |
2167 | } | |
2168 | } | |
2169 | free(in_pack.array); | |
2170 | } | |
2171 | ||
094085e3 BC |
2172 | static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1) |
2173 | { | |
2174 | static struct packed_git *last_found = (void *)1; | |
2175 | struct packed_git *p; | |
2176 | ||
2177 | p = (last_found != (void *)1) ? last_found : packed_git; | |
2178 | ||
2179 | while (p) { | |
2180 | if ((!p->pack_local || p->pack_keep) && | |
2181 | find_pack_entry_one(sha1, p)) { | |
2182 | last_found = p; | |
2183 | return 1; | |
2184 | } | |
2185 | if (p == last_found) | |
2186 | p = packed_git; | |
2187 | else | |
2188 | p = p->next; | |
2189 | if (p == last_found) | |
2190 | p = p->next; | |
2191 | } | |
2192 | return 0; | |
2193 | } | |
2194 | ||
ca11b212 NP |
2195 | static void loosen_unused_packed_objects(struct rev_info *revs) |
2196 | { | |
2197 | struct packed_git *p; | |
2198 | uint32_t i; | |
2199 | const unsigned char *sha1; | |
2200 | ||
2201 | for (p = packed_git; p; p = p->next) { | |
79bc4c71 | 2202 | if (!p->pack_local || p->pack_keep) |
ca11b212 NP |
2203 | continue; |
2204 | ||
2205 | if (open_pack_index(p)) | |
2206 | die("cannot open pack index"); | |
2207 | ||
2208 | for (i = 0; i < p->num_objects; i++) { | |
2209 | sha1 = nth_packed_object_sha1(p, i); | |
094085e3 BC |
2210 | if (!locate_object_entry(sha1) && |
2211 | !has_sha1_pack_kept_or_nonlocal(sha1)) | |
ca11b212 NP |
2212 | if (force_object_loose(sha1, p->mtime)) |
2213 | die("unable to force loose object"); | |
2214 | } | |
2215 | } | |
2216 | } | |
2217 | ||
8d1d8f83 | 2218 | static void get_object_list(int ac, const char **av) |
b5d97e6b JH |
2219 | { |
2220 | struct rev_info revs; | |
2221 | char line[1000]; | |
b5d97e6b JH |
2222 | int flags = 0; |
2223 | ||
b5d97e6b JH |
2224 | init_revisions(&revs, NULL); |
2225 | save_commit_buffer = 0; | |
b5d97e6b JH |
2226 | setup_revisions(ac, av, &revs, NULL); |
2227 | ||
b5d97e6b JH |
2228 | while (fgets(line, sizeof(line), stdin) != NULL) { |
2229 | int len = strlen(line); | |
872c930d | 2230 | if (len && line[len - 1] == '\n') |
b5d97e6b JH |
2231 | line[--len] = 0; |
2232 | if (!len) | |
2233 | break; | |
2234 | if (*line == '-') { | |
2235 | if (!strcmp(line, "--not")) { | |
2236 | flags ^= UNINTERESTING; | |
2237 | continue; | |
2238 | } | |
2239 | die("not a rev '%s'", line); | |
2240 | } | |
2241 | if (handle_revision_arg(line, &revs, flags, 1)) | |
2242 | die("bad revision '%s'", line); | |
2243 | } | |
2244 | ||
3d51e1b5 MK |
2245 | if (prepare_revision_walk(&revs)) |
2246 | die("revision walk setup failed"); | |
8d1d8f83 | 2247 | mark_edges_uninteresting(revs.commits, &revs, show_edge); |
11c211fa | 2248 | traverse_commit_list(&revs, show_commit, show_object, NULL); |
08cdfb13 JH |
2249 | |
2250 | if (keep_unreachable) | |
2251 | add_objects_in_unpacked_packs(&revs); | |
ca11b212 NP |
2252 | if (unpack_unreachable) |
2253 | loosen_unused_packed_objects(&revs); | |
b5d97e6b JH |
2254 | } |
2255 | ||
2256 | int cmd_pack_objects(int argc, const char **argv, const char *prefix) | |
2257 | { | |
b5d97e6b | 2258 | int use_internal_rev_list = 0; |
8d1d8f83 | 2259 | int thin = 0; |
4f366275 | 2260 | int all_progress_implied = 0; |
7cadf491 | 2261 | uint32_t i; |
ffa84ffb RD |
2262 | const char **rp_av; |
2263 | int rp_ac_alloc = 64; | |
8d1d8f83 JH |
2264 | int rp_ac; |
2265 | ||
dae556bd CC |
2266 | read_replace_refs = 0; |
2267 | ||
ffa84ffb RD |
2268 | rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av)); |
2269 | ||
8d1d8f83 JH |
2270 | rp_av[0] = "pack-objects"; |
2271 | rp_av[1] = "--objects"; /* --thin will make it --objects-edge */ | |
2272 | rp_ac = 2; | |
b5d97e6b | 2273 | |
ebcfb379 | 2274 | reset_pack_idx_option(&pack_idx_opts); |
ef90d6d4 | 2275 | git_config(git_pack_config, NULL); |
960ccca6 DH |
2276 | if (!pack_compression_seen && core_compression_seen) |
2277 | pack_compression_level = core_compression_level; | |
b5d97e6b JH |
2278 | |
2279 | progress = isatty(2); | |
2280 | for (i = 1; i < argc; i++) { | |
2281 | const char *arg = argv[i]; | |
2282 | ||
2283 | if (*arg != '-') | |
2284 | break; | |
2285 | ||
2286 | if (!strcmp("--non-empty", arg)) { | |
2287 | non_empty = 1; | |
2288 | continue; | |
2289 | } | |
2290 | if (!strcmp("--local", arg)) { | |
2291 | local = 1; | |
2292 | continue; | |
2293 | } | |
b5d97e6b JH |
2294 | if (!strcmp("--incremental", arg)) { |
2295 | incremental = 1; | |
2296 | continue; | |
2297 | } | |
e96fb9b8 BC |
2298 | if (!strcmp("--honor-pack-keep", arg)) { |
2299 | ignore_packed_keep = 1; | |
2300 | continue; | |
2301 | } | |
960ccca6 DH |
2302 | if (!prefixcmp(arg, "--compression=")) { |
2303 | char *end; | |
2304 | int level = strtoul(arg+14, &end, 0); | |
2305 | if (!arg[14] || *end) | |
2306 | usage(pack_usage); | |
2307 | if (level == -1) | |
2308 | level = Z_DEFAULT_COMPRESSION; | |
2309 | else if (level < 0 || level > Z_BEST_COMPRESSION) | |
2310 | die("bad pack compression level %d", level); | |
2311 | pack_compression_level = level; | |
2312 | continue; | |
2313 | } | |
6b94b1a0 | 2314 | if (!prefixcmp(arg, "--max-pack-size=")) { |
2b84b5a8 | 2315 | pack_size_limit_cfg = 0; |
07cf0f24 | 2316 | if (!git_parse_ulong(arg+16, &pack_size_limit)) |
6b94b1a0 DH |
2317 | usage(pack_usage); |
2318 | continue; | |
2319 | } | |
599065a3 | 2320 | if (!prefixcmp(arg, "--window=")) { |
b5d97e6b JH |
2321 | char *end; |
2322 | window = strtoul(arg+9, &end, 0); | |
2323 | if (!arg[9] || *end) | |
2324 | usage(pack_usage); | |
2325 | continue; | |
2326 | } | |
a97773ce BD |
2327 | if (!prefixcmp(arg, "--window-memory=")) { |
2328 | if (!git_parse_ulong(arg+16, &window_memory_limit)) | |
2329 | usage(pack_usage); | |
2330 | continue; | |
2331 | } | |
367f4a43 NP |
2332 | if (!prefixcmp(arg, "--threads=")) { |
2333 | char *end; | |
2334 | delta_search_threads = strtoul(arg+10, &end, 0); | |
833e3df1 | 2335 | if (!arg[10] || *end || delta_search_threads < 0) |
367f4a43 | 2336 | usage(pack_usage); |
7eb151d6 | 2337 | #ifdef NO_PTHREADS |
833e3df1 | 2338 | if (delta_search_threads != 1) |
367f4a43 NP |
2339 | warning("no threads support, " |
2340 | "ignoring %s", arg); | |
2341 | #endif | |
2342 | continue; | |
2343 | } | |
599065a3 | 2344 | if (!prefixcmp(arg, "--depth=")) { |
b5d97e6b JH |
2345 | char *end; |
2346 | depth = strtoul(arg+8, &end, 0); | |
2347 | if (!arg[8] || *end) | |
2348 | usage(pack_usage); | |
2349 | continue; | |
2350 | } | |
2351 | if (!strcmp("--progress", arg)) { | |
2352 | progress = 1; | |
2353 | continue; | |
2354 | } | |
231f240b NP |
2355 | if (!strcmp("--all-progress", arg)) { |
2356 | progress = 2; | |
2357 | continue; | |
2358 | } | |
4f366275 NP |
2359 | if (!strcmp("--all-progress-implied", arg)) { |
2360 | all_progress_implied = 1; | |
2361 | continue; | |
2362 | } | |
b5d97e6b JH |
2363 | if (!strcmp("-q", arg)) { |
2364 | progress = 0; | |
2365 | continue; | |
2366 | } | |
2367 | if (!strcmp("--no-reuse-delta", arg)) { | |
a7de7130 | 2368 | reuse_delta = 0; |
b5d97e6b JH |
2369 | continue; |
2370 | } | |
fa736f72 | 2371 | if (!strcmp("--no-reuse-object", arg)) { |
a7de7130 | 2372 | reuse_object = reuse_delta = 0; |
fa736f72 NP |
2373 | continue; |
2374 | } | |
be6b1914 | 2375 | if (!strcmp("--delta-base-offset", arg)) { |
780e6e73 | 2376 | allow_ofs_delta = 1; |
be6b1914 NP |
2377 | continue; |
2378 | } | |
b5d97e6b JH |
2379 | if (!strcmp("--stdout", arg)) { |
2380 | pack_to_stdout = 1; | |
2381 | continue; | |
2382 | } | |
2383 | if (!strcmp("--revs", arg)) { | |
2384 | use_internal_rev_list = 1; | |
2385 | continue; | |
2386 | } | |
08cdfb13 JH |
2387 | if (!strcmp("--keep-unreachable", arg)) { |
2388 | keep_unreachable = 1; | |
2389 | continue; | |
2390 | } | |
ca11b212 NP |
2391 | if (!strcmp("--unpack-unreachable", arg)) { |
2392 | unpack_unreachable = 1; | |
2393 | continue; | |
2394 | } | |
f0a24aa5 SP |
2395 | if (!strcmp("--include-tag", arg)) { |
2396 | include_tag = 1; | |
2397 | continue; | |
2398 | } | |
8d1d8f83 | 2399 | if (!strcmp("--unpacked", arg) || |
63049292 | 2400 | !strcmp("--reflog", arg) || |
8d1d8f83 JH |
2401 | !strcmp("--all", arg)) { |
2402 | use_internal_rev_list = 1; | |
ffa84ffb RD |
2403 | if (rp_ac >= rp_ac_alloc - 1) { |
2404 | rp_ac_alloc = alloc_nr(rp_ac_alloc); | |
2405 | rp_av = xrealloc(rp_av, | |
2406 | rp_ac_alloc * sizeof(*rp_av)); | |
2407 | } | |
8d1d8f83 | 2408 | rp_av[rp_ac++] = arg; |
b5d97e6b JH |
2409 | continue; |
2410 | } | |
8d1d8f83 JH |
2411 | if (!strcmp("--thin", arg)) { |
2412 | use_internal_rev_list = 1; | |
2413 | thin = 1; | |
2414 | rp_av[1] = "--objects-edge"; | |
b5d97e6b | 2415 | continue; |
4ba7d711 NP |
2416 | } |
2417 | if (!prefixcmp(arg, "--index-version=")) { | |
2418 | char *c; | |
ebcfb379 JH |
2419 | pack_idx_opts.version = strtoul(arg + 16, &c, 10); |
2420 | if (pack_idx_opts.version > 2) | |
4ba7d711 NP |
2421 | die("bad %s", arg); |
2422 | if (*c == ',') | |
ebcfb379 JH |
2423 | pack_idx_opts.off32_limit = strtoul(c+1, &c, 0); |
2424 | if (*c || pack_idx_opts.off32_limit & 0x80000000) | |
4ba7d711 NP |
2425 | die("bad %s", arg); |
2426 | continue; | |
b5d97e6b | 2427 | } |
7f3140cd JS |
2428 | if (!strcmp(arg, "--keep-true-parents")) { |
2429 | grafts_replace_parents = 0; | |
2430 | continue; | |
2431 | } | |
b5d97e6b JH |
2432 | usage(pack_usage); |
2433 | } | |
2434 | ||
2435 | /* Traditionally "pack-objects [options] base extra" failed; | |
2436 | * we would however want to take refs parameter that would | |
2437 | * have been given to upstream rev-list ourselves, which means | |
2438 | * we somehow want to say what the base name is. So the | |
2439 | * syntax would be: | |
2440 | * | |
2441 | * pack-objects [options] base <refs...> | |
2442 | * | |
2443 | * in other words, we would treat the first non-option as the | |
2444 | * base_name and send everything else to the internal revision | |
2445 | * walker. | |
2446 | */ | |
2447 | ||
2448 | if (!pack_to_stdout) | |
2449 | base_name = argv[i++]; | |
2450 | ||
2451 | if (pack_to_stdout != !base_name) | |
2452 | usage(pack_usage); | |
2453 | ||
2b84b5a8 JS |
2454 | if (!pack_to_stdout && !pack_size_limit) |
2455 | pack_size_limit = pack_size_limit_cfg; | |
01c12a23 DH |
2456 | if (pack_to_stdout && pack_size_limit) |
2457 | die("--max-pack-size cannot be used to build a pack for transfer."); | |
07cf0f24 NP |
2458 | if (pack_size_limit && pack_size_limit < 1024*1024) { |
2459 | warning("minimum pack size limit is 1 MiB"); | |
2460 | pack_size_limit = 1024*1024; | |
2461 | } | |
01c12a23 | 2462 | |
8d1d8f83 JH |
2463 | if (!pack_to_stdout && thin) |
2464 | die("--thin cannot be used to build an indexable pack."); | |
b5d97e6b | 2465 | |
ca11b212 NP |
2466 | if (keep_unreachable && unpack_unreachable) |
2467 | die("--keep-unreachable and --unpack-unreachable are incompatible."); | |
2468 | ||
4f366275 NP |
2469 | if (progress && all_progress_implied) |
2470 | progress = 2; | |
2471 | ||
b5d97e6b JH |
2472 | prepare_packed_git(); |
2473 | ||
13aaf148 | 2474 | if (progress) |
dc6a0757 | 2475 | progress_state = start_progress("Counting objects", 0); |
b5d97e6b JH |
2476 | if (!use_internal_rev_list) |
2477 | read_object_list_from_stdin(); | |
8d1d8f83 JH |
2478 | else { |
2479 | rp_av[rp_ac] = NULL; | |
2480 | get_object_list(rp_ac, rp_av); | |
2481 | } | |
0ef95f72 | 2482 | cleanup_preferred_base(); |
f0a24aa5 SP |
2483 | if (include_tag && nr_result) |
2484 | for_each_ref(add_ref_tag, NULL); | |
4d4fcc54 | 2485 | stop_progress(&progress_state); |
96a02f8f | 2486 | |
f0b0af1b | 2487 | if (non_empty && !nr_result) |
1c4a2912 | 2488 | return 0; |
f7ae6a93 NP |
2489 | if (nr_result) |
2490 | prepare_pack(window, depth); | |
d01fb92f | 2491 | write_pack_file(); |
ab7cd7bb | 2492 | if (progress) |
6e1c2344 RJ |
2493 | fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32")," |
2494 | " reused %"PRIu32" (delta %"PRIu32")\n", | |
67c08ce1 | 2495 | written, written_delta, reused, reused_delta); |
c323ac7d LT |
2496 | return 0; |
2497 | } |