]>
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" |
c323ac7d | 19 | |
8ecce684 | 20 | #ifdef THREADED_DELTA_SEARCH |
833e3df1 | 21 | #include "thread-utils.h" |
8ecce684 NP |
22 | #include <pthread.h> |
23 | #endif | |
24 | ||
231f240b | 25 | static const char pack_usage[] = "\ |
a97773ce BD |
26 | git-pack-objects [{ -q | --progress | --all-progress }] \n\ |
27 | [--max-pack-size=N] [--local] [--incremental] \n\ | |
28 | [--window=N] [--window-memory=N] [--depth=N] \n\ | |
fa736f72 | 29 | [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\ |
367f4a43 | 30 | [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\ |
ca11b212 NP |
31 | [--stdout | base-name] [--include-tag] \n\ |
32 | [--keep-unreachable | --unpack-unreachable] \n\ | |
f0a24aa5 | 33 | [<ref-list | <object-list]"; |
c323ac7d | 34 | |
c323ac7d | 35 | struct object_entry { |
aa7e44bf | 36 | struct pack_idx_entry idx; |
3f9ac8d2 | 37 | unsigned long size; /* uncompressed size */ |
3f9ac8d2 | 38 | struct packed_git *in_pack; /* already in pack */ |
6777a59f | 39 | off_t in_pack_offset; |
a3fbf4df | 40 | struct object_entry *delta; /* delta base object */ |
82e5a82f | 41 | struct object_entry *delta_child; /* deltified objects who bases me */ |
15b4d577 JH |
42 | struct object_entry *delta_sibling; /* other deltified objects who |
43 | * uses the same base as me | |
44 | */ | |
074b2eea | 45 | void *delta_data; /* cached delta (uncompressed) */ |
a3fbf4df | 46 | unsigned long delta_size; /* delta data size (uncompressed) */ |
ed4a9031 | 47 | unsigned long z_delta_size; /* delta data size (compressed) */ |
5a235b5e | 48 | unsigned int hash; /* name hint hash */ |
a3fbf4df NP |
49 | enum object_type type; |
50 | enum object_type in_pack_type; /* could be delta */ | |
51 | unsigned char in_pack_header_size; | |
52 | unsigned char preferred_base; /* we do not pack this, but is available | |
a74db82e | 53 | * to be used as the base object to delta |
a3fbf4df NP |
54 | * objects against. |
55 | */ | |
a74db82e | 56 | unsigned char no_try_delta; |
c323ac7d LT |
57 | }; |
58 | ||
3f9ac8d2 | 59 | /* |
82e5a82f | 60 | * Objects we are going to pack are collected in objects array (dynamically |
3f9ac8d2 JH |
61 | * expanded). nr_objects & nr_alloc controls this array. They are stored |
62 | * in the order we see -- typically rev-list --objects order that gives us | |
63 | * nice "minimum seek" order. | |
3f9ac8d2 | 64 | */ |
e4d58311 | 65 | static struct object_entry *objects; |
79814f42 | 66 | static struct pack_idx_entry **written_list; |
d01fb92f | 67 | static uint32_t nr_objects, nr_alloc, nr_result, nr_written; |
3f9ac8d2 | 68 | |
96f1e58f | 69 | static int non_empty; |
a7de7130 | 70 | static int reuse_delta = 1, reuse_object = 1; |
e5e9714a | 71 | static int keep_unreachable, unpack_unreachable, include_tag; |
96f1e58f DR |
72 | static int local; |
73 | static int incremental; | |
be6b1914 | 74 | static int allow_ofs_delta; |
d01fb92f | 75 | static const char *base_name; |
024701f1 | 76 | static int progress = 1; |
4812a93a | 77 | static int window = 10; |
2b84b5a8 | 78 | static uint32_t pack_size_limit, pack_size_limit_cfg; |
618e613a | 79 | static int depth = 50; |
367f4a43 | 80 | static int delta_search_threads = 1; |
df6d6101 | 81 | static int pack_to_stdout; |
8d1d8f83 | 82 | static int num_preferred_base; |
dc6a0757 | 83 | static struct progress *progress_state; |
960ccca6 DH |
84 | static int pack_compression_level = Z_DEFAULT_COMPRESSION; |
85 | static int pack_compression_seen; | |
c323ac7d | 86 | |
074b2eea MK |
87 | static unsigned long delta_cache_size = 0; |
88 | static unsigned long max_delta_cache_size = 0; | |
e3dfddb3 | 89 | static unsigned long cache_max_small_delta_size = 1000; |
074b2eea | 90 | |
a97773ce BD |
91 | static unsigned long window_memory_limit = 0; |
92 | ||
3f9ac8d2 JH |
93 | /* |
94 | * The object names in objects array are hashed with this hashtable, | |
e4d58311 | 95 | * to help looking up the entry by object name. |
3f9ac8d2 JH |
96 | * This hashtable is built after all the objects are seen. |
97 | */ | |
96f1e58f DR |
98 | static int *object_ix; |
99 | static int object_ix_hashsz; | |
3f9ac8d2 | 100 | |
3f9ac8d2 JH |
101 | /* |
102 | * stats | |
103 | */ | |
7cadf491 SP |
104 | static uint32_t written, written_delta; |
105 | static uint32_t reused, reused_delta; | |
3f9ac8d2 | 106 | |
780e6e73 | 107 | |
3613f9b4 | 108 | static void *get_delta(struct object_entry *entry) |
c323ac7d | 109 | { |
3613f9b4 NP |
110 | unsigned long size, base_size, delta_size; |
111 | void *buf, *base_buf, *delta_buf; | |
21666f1a | 112 | enum object_type type; |
c323ac7d | 113 | |
3613f9b4 NP |
114 | buf = read_sha1_file(entry->idx.sha1, &type, &size); |
115 | if (!buf) | |
116 | die("unable to read %s", sha1_to_hex(entry->idx.sha1)); | |
117 | base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size); | |
118 | if (!base_buf) | |
aa7e44bf | 119 | die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1)); |
3613f9b4 | 120 | delta_buf = diff_delta(base_buf, base_size, |
dcde55bc | 121 | buf, size, &delta_size, 0); |
3613f9b4 | 122 | if (!delta_buf || delta_size != entry->delta_size) |
a6080a0a | 123 | die("delta size changed"); |
3613f9b4 NP |
124 | free(buf); |
125 | free(base_buf); | |
c323ac7d LT |
126 | return delta_buf; |
127 | } | |
128 | ||
30ebb40a NP |
129 | static unsigned long do_compress(void **pptr, unsigned long size) |
130 | { | |
131 | z_stream stream; | |
132 | void *in, *out; | |
133 | unsigned long maxsize; | |
134 | ||
135 | memset(&stream, 0, sizeof(stream)); | |
136 | deflateInit(&stream, pack_compression_level); | |
137 | maxsize = deflateBound(&stream, size); | |
138 | ||
139 | in = *pptr; | |
140 | out = xmalloc(maxsize); | |
141 | *pptr = out; | |
142 | ||
143 | stream.next_in = in; | |
144 | stream.avail_in = size; | |
145 | stream.next_out = out; | |
146 | stream.avail_out = maxsize; | |
147 | while (deflate(&stream, Z_FINISH) == Z_OK) | |
148 | ; /* nothing */ | |
149 | deflateEnd(&stream); | |
150 | ||
151 | free(in); | |
152 | return stream.total_out; | |
153 | } | |
154 | ||
a733cb60 LT |
155 | /* |
156 | * The per-object header is a pretty dense thing, which is | |
157 | * - first byte: low four bits are "size", then three bits of "type", | |
158 | * and the high bit is "size continues". | |
159 | * - each byte afterwards: low seven bits are size continuation, | |
160 | * with the high bit being "size continues" | |
161 | */ | |
162 | static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr) | |
163 | { | |
01247d87 | 164 | int n = 1; |
a733cb60 LT |
165 | unsigned char c; |
166 | ||
eb32d236 | 167 | if (type < OBJ_COMMIT || type > OBJ_REF_DELTA) |
a733cb60 LT |
168 | die("bad type %d", type); |
169 | ||
01247d87 LT |
170 | c = (type << 4) | (size & 15); |
171 | size >>= 4; | |
172 | while (size) { | |
a733cb60 | 173 | *hdr++ = c | 0x80; |
01247d87 LT |
174 | c = size & 0x7f; |
175 | size >>= 7; | |
176 | n++; | |
a733cb60 LT |
177 | } |
178 | *hdr = c; | |
179 | return n; | |
180 | } | |
181 | ||
780e6e73 NP |
182 | /* |
183 | * we are going to reuse the existing object data as is. make | |
184 | * sure it is not corrupt. | |
185 | */ | |
079afb18 SP |
186 | static int check_pack_inflate(struct packed_git *p, |
187 | struct pack_window **w_curs, | |
6777a59f SP |
188 | off_t offset, |
189 | off_t len, | |
079afb18 SP |
190 | unsigned long expect) |
191 | { | |
192 | z_stream stream; | |
193 | unsigned char fakebuf[4096], *in; | |
194 | int st; | |
195 | ||
196 | memset(&stream, 0, sizeof(stream)); | |
197 | inflateInit(&stream); | |
198 | do { | |
199 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
200 | stream.next_in = in; | |
201 | stream.next_out = fakebuf; | |
202 | stream.avail_out = sizeof(fakebuf); | |
203 | st = inflate(&stream, Z_FINISH); | |
204 | offset += stream.next_in - in; | |
205 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
206 | inflateEnd(&stream); | |
207 | return (st == Z_STREAM_END && | |
208 | stream.total_out == expect && | |
209 | stream.total_in == len) ? 0 : -1; | |
210 | } | |
211 | ||
212 | static void copy_pack_data(struct sha1file *f, | |
213 | struct packed_git *p, | |
214 | struct pack_window **w_curs, | |
6777a59f SP |
215 | off_t offset, |
216 | off_t len) | |
079afb18 SP |
217 | { |
218 | unsigned char *in; | |
219 | unsigned int avail; | |
220 | ||
221 | while (len) { | |
222 | in = use_pack(p, w_curs, offset, &avail); | |
223 | if (avail > len) | |
6777a59f | 224 | avail = (unsigned int)len; |
079afb18 SP |
225 | sha1write(f, in, avail); |
226 | offset += avail; | |
227 | len -= avail; | |
228 | } | |
229 | } | |
230 | ||
d7dd0223 | 231 | static unsigned long write_object(struct sha1file *f, |
17b08f2c DH |
232 | struct object_entry *entry, |
233 | off_t write_offset) | |
c323ac7d | 234 | { |
30ebb40a | 235 | unsigned long size, limit, datalen; |
3f9ac8d2 | 236 | void *buf; |
2c5ef824 | 237 | unsigned char header[10], dheader[10]; |
6777a59f | 238 | unsigned hdrlen; |
2c5ef824 NP |
239 | enum object_type type; |
240 | int usable_delta, to_reuse; | |
c323ac7d | 241 | |
78d1e84f NP |
242 | if (!pack_to_stdout) |
243 | crc32_begin(f); | |
244 | ||
2c5ef824 NP |
245 | type = entry->type; |
246 | ||
247 | /* write limit if limited packsize and not first object */ | |
248 | limit = pack_size_limit && nr_written ? | |
249 | pack_size_limit - write_offset : 0; | |
250 | ||
251 | if (!entry->delta) | |
252 | usable_delta = 0; /* no delta */ | |
253 | else if (!pack_size_limit) | |
254 | usable_delta = 1; /* unlimited packfile */ | |
255 | else if (entry->delta->idx.offset == (off_t)-1) | |
256 | usable_delta = 0; /* base was written to another pack */ | |
257 | else if (entry->delta->idx.offset) | |
258 | usable_delta = 1; /* base already exists in this pack */ | |
259 | else | |
260 | usable_delta = 0; /* base could end up in another pack */ | |
261 | ||
a7de7130 | 262 | if (!reuse_object) |
fa736f72 NP |
263 | to_reuse = 0; /* explicit */ |
264 | else if (!entry->in_pack) | |
ab7cd7bb | 265 | to_reuse = 0; /* can't reuse what we don't have */ |
2c5ef824 | 266 | else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA) |
17b08f2c DH |
267 | /* check_object() decided it for us ... */ |
268 | to_reuse = usable_delta; | |
269 | /* ... but pack split may override that */ | |
2c5ef824 | 270 | else if (type != entry->in_pack_type) |
ab7cd7bb JH |
271 | to_reuse = 0; /* pack has delta which is unusable */ |
272 | else if (entry->delta) | |
273 | to_reuse = 0; /* we want to pack afresh */ | |
274 | else | |
275 | to_reuse = 1; /* we have it in-pack undeltified, | |
276 | * and we do not need to deltify it. | |
277 | */ | |
278 | ||
df6d6101 | 279 | if (!to_reuse) { |
5476a8ad | 280 | if (!usable_delta) { |
2c5ef824 | 281 | buf = read_sha1_file(entry->idx.sha1, &type, &size); |
5476a8ad | 282 | if (!buf) |
aa7e44bf | 283 | die("unable to read %s", sha1_to_hex(entry->idx.sha1)); |
ed4a9031 NP |
284 | /* |
285 | * make sure no cached delta data remains from a | |
286 | * previous attempt before a pack split occured. | |
287 | */ | |
288 | free(entry->delta_data); | |
289 | entry->delta_data = NULL; | |
290 | entry->z_delta_size = 0; | |
5476a8ad | 291 | } else if (entry->delta_data) { |
3f9ac8d2 | 292 | size = entry->delta_size; |
5476a8ad NP |
293 | buf = entry->delta_data; |
294 | entry->delta_data = NULL; | |
2c5ef824 | 295 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
be6b1914 | 296 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
17b08f2c | 297 | } else { |
3613f9b4 | 298 | buf = get_delta(entry); |
5476a8ad | 299 | size = entry->delta_size; |
2c5ef824 | 300 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
5476a8ad | 301 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
3f9ac8d2 | 302 | } |
ed4a9031 NP |
303 | |
304 | if (entry->z_delta_size) | |
305 | datalen = entry->z_delta_size; | |
306 | else | |
307 | datalen = do_compress(&buf, size); | |
8c3c7b2a | 308 | |
3f9ac8d2 JH |
309 | /* |
310 | * The object header is a byte of 'type' followed by zero or | |
be6b1914 | 311 | * more bytes of length. |
3f9ac8d2 | 312 | */ |
2c5ef824 | 313 | hdrlen = encode_header(type, size, header); |
3f9ac8d2 | 314 | |
2c5ef824 | 315 | if (type == OBJ_OFS_DELTA) { |
be6b1914 NP |
316 | /* |
317 | * Deltas with relative base contain an additional | |
318 | * encoding of the relative offset for the delta | |
319 | * base from this object's position in the pack. | |
320 | */ | |
aa7e44bf | 321 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; |
17b08f2c DH |
322 | unsigned pos = sizeof(dheader) - 1; |
323 | dheader[pos] = ofs & 127; | |
be6b1914 | 324 | while (ofs >>= 7) |
17b08f2c DH |
325 | dheader[--pos] = 128 | (--ofs & 127); |
326 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) { | |
17b08f2c DH |
327 | free(buf); |
328 | return 0; | |
329 | } | |
330 | sha1write(f, header, hdrlen); | |
331 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
332 | hdrlen += sizeof(dheader) - pos; | |
2c5ef824 | 333 | } else if (type == OBJ_REF_DELTA) { |
be6b1914 NP |
334 | /* |
335 | * Deltas with a base reference contain | |
336 | * an additional 20 bytes for the base sha1. | |
337 | */ | |
17b08f2c | 338 | if (limit && hdrlen + 20 + datalen + 20 >= limit) { |
17b08f2c DH |
339 | free(buf); |
340 | return 0; | |
341 | } | |
342 | sha1write(f, header, hdrlen); | |
aa7e44bf | 343 | sha1write(f, entry->delta->idx.sha1, 20); |
3f9ac8d2 | 344 | hdrlen += 20; |
17b08f2c DH |
345 | } else { |
346 | if (limit && hdrlen + datalen + 20 >= limit) { | |
17b08f2c DH |
347 | free(buf); |
348 | return 0; | |
349 | } | |
350 | sha1write(f, header, hdrlen); | |
3f9ac8d2 | 351 | } |
30ebb40a | 352 | sha1write(f, buf, datalen); |
3f9ac8d2 | 353 | free(buf); |
c323ac7d | 354 | } |
3f9ac8d2 JH |
355 | else { |
356 | struct packed_git *p = entry->in_pack; | |
03e79c88 | 357 | struct pack_window *w_curs = NULL; |
91ecbeca | 358 | struct revindex_entry *revidx; |
6777a59f | 359 | off_t offset; |
3f9ac8d2 | 360 | |
780e6e73 | 361 | if (entry->delta) { |
2c5ef824 | 362 | type = (allow_ofs_delta && entry->delta->idx.offset) ? |
780e6e73 NP |
363 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
364 | reused_delta++; | |
365 | } | |
2c5ef824 | 366 | hdrlen = encode_header(type, entry->size, header); |
91ecbeca | 367 | offset = entry->in_pack_offset; |
3449f8c4 | 368 | revidx = find_pack_revindex(p, offset); |
91ecbeca NP |
369 | datalen = revidx[1].offset - offset; |
370 | if (!pack_to_stdout && p->index_version > 1 && | |
371 | check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) | |
aa7e44bf | 372 | die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1)); |
91ecbeca NP |
373 | offset += entry->in_pack_header_size; |
374 | datalen -= entry->in_pack_header_size; | |
2c5ef824 | 375 | if (type == OBJ_OFS_DELTA) { |
aa7e44bf | 376 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; |
17b08f2c DH |
377 | unsigned pos = sizeof(dheader) - 1; |
378 | dheader[pos] = ofs & 127; | |
379 | while (ofs >>= 7) | |
380 | dheader[--pos] = 128 | (--ofs & 127); | |
381 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) | |
382 | return 0; | |
383 | sha1write(f, header, hdrlen); | |
384 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
385 | hdrlen += sizeof(dheader) - pos; | |
2c5ef824 | 386 | } else if (type == OBJ_REF_DELTA) { |
17b08f2c DH |
387 | if (limit && hdrlen + 20 + datalen + 20 >= limit) |
388 | return 0; | |
389 | sha1write(f, header, hdrlen); | |
aa7e44bf | 390 | sha1write(f, entry->delta->idx.sha1, 20); |
17b08f2c DH |
391 | hdrlen += 20; |
392 | } else { | |
393 | if (limit && hdrlen + datalen + 20 >= limit) | |
394 | return 0; | |
395 | sha1write(f, header, hdrlen); | |
396 | } | |
397 | ||
91ecbeca NP |
398 | if (!pack_to_stdout && p->index_version == 1 && |
399 | check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) | |
aa7e44bf | 400 | die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1)); |
079afb18 | 401 | copy_pack_data(f, p, &w_curs, offset, datalen); |
03e79c88 | 402 | unuse_pack(&w_curs); |
3f9ac8d2 | 403 | reused++; |
a733cb60 | 404 | } |
17b08f2c | 405 | if (usable_delta) |
ab7cd7bb | 406 | written_delta++; |
3f9ac8d2 | 407 | written++; |
78d1e84f | 408 | if (!pack_to_stdout) |
aa7e44bf | 409 | entry->idx.crc32 = crc32_end(f); |
c323ac7d LT |
410 | return hdrlen + datalen; |
411 | } | |
412 | ||
6777a59f | 413 | static off_t write_one(struct sha1file *f, |
9d5ab962 | 414 | struct object_entry *e, |
6777a59f | 415 | off_t offset) |
9d5ab962 | 416 | { |
d7dd0223 NP |
417 | unsigned long size; |
418 | ||
419 | /* offset is non zero if object is written already. */ | |
aa7e44bf | 420 | if (e->idx.offset || e->preferred_base) |
9d5ab962 | 421 | return offset; |
d7dd0223 NP |
422 | |
423 | /* if we are deltified, write out base object first. */ | |
17b08f2c | 424 | if (e->delta) { |
9d5ab962 | 425 | offset = write_one(f, e->delta, offset); |
17b08f2c DH |
426 | if (!offset) |
427 | return 0; | |
428 | } | |
d7dd0223 | 429 | |
aa7e44bf | 430 | e->idx.offset = offset; |
17b08f2c DH |
431 | size = write_object(f, e, offset); |
432 | if (!size) { | |
aa7e44bf | 433 | e->idx.offset = 0; |
17b08f2c DH |
434 | return 0; |
435 | } | |
79814f42 | 436 | written_list[nr_written++] = &e->idx; |
d7dd0223 NP |
437 | |
438 | /* make sure off_t is sufficiently large not to wrap */ | |
439 | if (offset > offset + size) | |
440 | die("pack too large for current definition of off_t"); | |
441 | return offset + size; | |
9d5ab962 JH |
442 | } |
443 | ||
aa7e44bf | 444 | /* forward declaration for write_pack_file */ |
d01fb92f DH |
445 | static int adjust_perm(const char *path, mode_t mode); |
446 | ||
447 | static void write_pack_file(void) | |
c323ac7d | 448 | { |
ebe27b13 | 449 | uint32_t i = 0, j; |
d22b9290 | 450 | struct sha1file *f; |
ebe27b13 | 451 | off_t offset, offset_one, last_obj_offset = 0; |
a733cb60 | 452 | struct pack_header hdr; |
ebe27b13 | 453 | uint32_t nr_remaining = nr_result; |
f746bae8 | 454 | time_t last_mtime = 0; |
81a216a5 | 455 | |
bcd7954e | 456 | if (progress > pack_to_stdout) |
dc6a0757 | 457 | progress_state = start_progress("Writing objects", nr_result); |
79814f42 | 458 | written_list = xmalloc(nr_objects * sizeof(*written_list)); |
183bdb2c | 459 | |
ebe27b13 | 460 | do { |
aa7e44bf | 461 | unsigned char sha1[20]; |
7ba502c4 | 462 | char *pack_tmp_name = NULL; |
aa7e44bf | 463 | |
ebe27b13 | 464 | if (pack_to_stdout) { |
2a128d63 | 465 | f = sha1fd_throughput(1, "<stdout>", progress_state); |
ebe27b13 | 466 | } else { |
7ba502c4 NP |
467 | char tmpname[PATH_MAX]; |
468 | int fd; | |
469 | snprintf(tmpname, sizeof(tmpname), | |
470 | "%s/tmp_pack_XXXXXX", get_object_directory()); | |
471 | fd = xmkstemp(tmpname); | |
ebe27b13 DH |
472 | pack_tmp_name = xstrdup(tmpname); |
473 | f = sha1fd(fd, pack_tmp_name); | |
474 | } | |
475 | ||
476 | hdr.hdr_signature = htonl(PACK_SIGNATURE); | |
477 | hdr.hdr_version = htonl(PACK_VERSION); | |
478 | hdr.hdr_entries = htonl(nr_remaining); | |
479 | sha1write(f, &hdr, sizeof(hdr)); | |
480 | offset = sizeof(hdr); | |
481 | nr_written = 0; | |
482 | for (; i < nr_objects; i++) { | |
483 | last_obj_offset = offset; | |
484 | offset_one = write_one(f, objects + i, offset); | |
485 | if (!offset_one) | |
486 | break; | |
487 | offset = offset_one; | |
4d4fcc54 | 488 | display_progress(progress_state, written); |
ebe27b13 | 489 | } |
c553ca25 | 490 | |
ebe27b13 DH |
491 | /* |
492 | * Did we write the wrong # entries in the header? | |
493 | * If so, rewrite it like in fast-import | |
494 | */ | |
54352bb2 LT |
495 | if (pack_to_stdout) { |
496 | sha1close(f, sha1, CSUM_CLOSE); | |
497 | } else if (nr_written == nr_remaining) { | |
498 | sha1close(f, sha1, CSUM_FSYNC); | |
ebe27b13 | 499 | } else { |
7ba502c4 NP |
500 | int fd = sha1close(f, NULL, 0); |
501 | fixup_pack_header_footer(fd, sha1, pack_tmp_name, nr_written); | |
4c81b03e | 502 | fsync_or_die(fd, pack_tmp_name); |
7ba502c4 | 503 | close(fd); |
ebe27b13 DH |
504 | } |
505 | ||
506 | if (!pack_to_stdout) { | |
d01fb92f | 507 | mode_t mode = umask(0); |
f746bae8 | 508 | struct stat st; |
7ba502c4 | 509 | char *idx_tmp_name, tmpname[PATH_MAX]; |
d01fb92f DH |
510 | |
511 | umask(mode); | |
512 | mode = 0444 & ~mode; | |
513 | ||
79814f42 NP |
514 | idx_tmp_name = write_idx_file(NULL, written_list, |
515 | nr_written, sha1); | |
f746bae8 | 516 | |
d01fb92f | 517 | snprintf(tmpname, sizeof(tmpname), "%s-%s.pack", |
aa7e44bf | 518 | base_name, sha1_to_hex(sha1)); |
d01fb92f DH |
519 | if (adjust_perm(pack_tmp_name, mode)) |
520 | die("unable to make temporary pack file readable: %s", | |
521 | strerror(errno)); | |
522 | if (rename(pack_tmp_name, tmpname)) | |
523 | die("unable to rename temporary pack file: %s", | |
524 | strerror(errno)); | |
f746bae8 NP |
525 | |
526 | /* | |
527 | * Packs are runtime accessed in their mtime | |
528 | * order since newer packs are more likely to contain | |
529 | * younger objects. So if we are creating multiple | |
530 | * packs then we should modify the mtime of later ones | |
531 | * to preserve this property. | |
532 | */ | |
533 | if (stat(tmpname, &st) < 0) { | |
534 | warning("failed to stat %s: %s", | |
535 | tmpname, strerror(errno)); | |
536 | } else if (!last_mtime) { | |
537 | last_mtime = st.st_mtime; | |
538 | } else { | |
539 | struct utimbuf utb; | |
540 | utb.actime = st.st_atime; | |
541 | utb.modtime = --last_mtime; | |
542 | if (utime(tmpname, &utb) < 0) | |
543 | warning("failed utime() on %s: %s", | |
544 | tmpname, strerror(errno)); | |
545 | } | |
546 | ||
d01fb92f | 547 | snprintf(tmpname, sizeof(tmpname), "%s-%s.idx", |
aa7e44bf | 548 | base_name, sha1_to_hex(sha1)); |
d01fb92f DH |
549 | if (adjust_perm(idx_tmp_name, mode)) |
550 | die("unable to make temporary index file readable: %s", | |
551 | strerror(errno)); | |
552 | if (rename(idx_tmp_name, tmpname)) | |
553 | die("unable to rename temporary index file: %s", | |
554 | strerror(errno)); | |
f746bae8 | 555 | |
7ba502c4 NP |
556 | free(idx_tmp_name); |
557 | free(pack_tmp_name); | |
aa7e44bf | 558 | puts(sha1_to_hex(sha1)); |
ebe27b13 DH |
559 | } |
560 | ||
561 | /* mark written objects as written to previous pack */ | |
562 | for (j = 0; j < nr_written; j++) { | |
79814f42 | 563 | written_list[j]->offset = (off_t)-1; |
ebe27b13 DH |
564 | } |
565 | nr_remaining -= nr_written; | |
566 | } while (nr_remaining && i < nr_objects); | |
567 | ||
568 | free(written_list); | |
4d4fcc54 | 569 | stop_progress(&progress_state); |
67c08ce1 | 570 | if (written != nr_result) |
6e1c2344 RJ |
571 | die("wrote %"PRIu32" objects while expecting %"PRIu32, |
572 | written, nr_result); | |
01c12a23 DH |
573 | /* |
574 | * We have scanned through [0 ... i). Since we have written | |
575 | * the correct number of objects, the remaining [i ... nr_objects) | |
576 | * items must be either already written (due to out-of-order delta base) | |
577 | * or a preferred base. Count those which are neither and complain if any. | |
578 | */ | |
ebe27b13 DH |
579 | for (j = 0; i < nr_objects; i++) { |
580 | struct object_entry *e = objects + i; | |
aa7e44bf | 581 | j += !e->idx.offset && !e->preferred_base; |
d01fb92f | 582 | } |
ebe27b13 | 583 | if (j) |
6e1c2344 RJ |
584 | die("wrote %"PRIu32" objects as expected but %"PRIu32 |
585 | " unwritten", written, j); | |
c323ac7d LT |
586 | } |
587 | ||
7a979d99 JH |
588 | static int locate_object_entry_hash(const unsigned char *sha1) |
589 | { | |
590 | int i; | |
591 | unsigned int ui; | |
592 | memcpy(&ui, sha1, sizeof(unsigned int)); | |
593 | i = ui % object_ix_hashsz; | |
594 | while (0 < object_ix[i]) { | |
aa7e44bf | 595 | if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1)) |
7a979d99 JH |
596 | return i; |
597 | if (++i == object_ix_hashsz) | |
598 | i = 0; | |
599 | } | |
600 | return -1 - i; | |
601 | } | |
602 | ||
603 | static struct object_entry *locate_object_entry(const unsigned char *sha1) | |
604 | { | |
605 | int i; | |
606 | ||
607 | if (!object_ix_hashsz) | |
608 | return NULL; | |
609 | ||
610 | i = locate_object_entry_hash(sha1); | |
611 | if (0 <= i) | |
612 | return &objects[object_ix[i]-1]; | |
613 | return NULL; | |
614 | } | |
615 | ||
616 | static void rehash_objects(void) | |
c323ac7d | 617 | { |
7cadf491 | 618 | uint32_t i; |
7a979d99 JH |
619 | struct object_entry *oe; |
620 | ||
621 | object_ix_hashsz = nr_objects * 3; | |
622 | if (object_ix_hashsz < 1024) | |
623 | object_ix_hashsz = 1024; | |
624 | object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz); | |
5379a5c5 | 625 | memset(object_ix, 0, sizeof(int) * object_ix_hashsz); |
7a979d99 | 626 | for (i = 0, oe = objects; i < nr_objects; i++, oe++) { |
aa7e44bf | 627 | int ix = locate_object_entry_hash(oe->idx.sha1); |
7a979d99 JH |
628 | if (0 <= ix) |
629 | continue; | |
630 | ix = -1 - ix; | |
631 | object_ix[ix] = i + 1; | |
632 | } | |
633 | } | |
634 | ||
ce0bd642 | 635 | static unsigned name_hash(const char *name) |
1d6b38cc | 636 | { |
ce0bd642 LT |
637 | unsigned char c; |
638 | unsigned hash = 0; | |
639 | ||
bc32fed5 JH |
640 | if (!name) |
641 | return 0; | |
642 | ||
eeef7135 | 643 | /* |
ce0bd642 LT |
644 | * This effectively just creates a sortable number from the |
645 | * last sixteen non-whitespace characters. Last characters | |
646 | * count "most", so things that end in ".c" sort together. | |
eeef7135 | 647 | */ |
ce0bd642 LT |
648 | while ((c = *name++) != 0) { |
649 | if (isspace(c)) | |
650 | continue; | |
651 | hash = (hash >> 2) + (c << 24); | |
652 | } | |
1d6b38cc JH |
653 | return hash; |
654 | } | |
655 | ||
a74db82e JH |
656 | static void setup_delta_attr_check(struct git_attr_check *check) |
657 | { | |
658 | static struct git_attr *attr_delta; | |
659 | ||
660 | if (!attr_delta) | |
661 | attr_delta = git_attr("delta", 5); | |
662 | ||
663 | check[0].attr = attr_delta; | |
664 | } | |
665 | ||
666 | static int no_try_delta(const char *path) | |
667 | { | |
668 | struct git_attr_check check[1]; | |
669 | ||
670 | setup_delta_attr_check(check); | |
671 | if (git_checkattr(path, ARRAY_SIZE(check), check)) | |
672 | return 0; | |
673 | if (ATTR_FALSE(check->value)) | |
674 | return 1; | |
675 | return 0; | |
676 | } | |
677 | ||
5c49c116 | 678 | static int add_object_entry(const unsigned char *sha1, enum object_type type, |
bc32fed5 | 679 | const char *name, int exclude) |
c323ac7d | 680 | { |
c323ac7d | 681 | struct object_entry *entry; |
29b734e4 | 682 | struct packed_git *p, *found_pack = NULL; |
6777a59f | 683 | off_t found_offset = 0; |
29b734e4 | 684 | int ix; |
bc32fed5 | 685 | unsigned hash = name_hash(name); |
29b734e4 NP |
686 | |
687 | ix = nr_objects ? locate_object_entry_hash(sha1) : -1; | |
688 | if (ix >= 0) { | |
689 | if (exclude) { | |
690 | entry = objects + object_ix[ix] - 1; | |
81a216a5 NP |
691 | if (!entry->preferred_base) |
692 | nr_result--; | |
29b734e4 NP |
693 | entry->preferred_base = 1; |
694 | } | |
695 | return 0; | |
696 | } | |
c323ac7d | 697 | |
5c49c116 NP |
698 | for (p = packed_git; p; p = p->next) { |
699 | off_t offset = find_pack_entry_one(sha1, p); | |
700 | if (offset) { | |
701 | if (!found_pack) { | |
702 | found_offset = offset; | |
703 | found_pack = p; | |
64560374 | 704 | } |
5c49c116 NP |
705 | if (exclude) |
706 | break; | |
707 | if (incremental) | |
708 | return 0; | |
709 | if (local && !p->pack_local) | |
710 | return 0; | |
64560374 LT |
711 | } |
712 | } | |
eb019375 | 713 | |
29b734e4 NP |
714 | if (nr_objects >= nr_alloc) { |
715 | nr_alloc = (nr_alloc + 1024) * 3 / 2; | |
7cadf491 | 716 | objects = xrealloc(objects, nr_alloc * sizeof(*entry)); |
c323ac7d | 717 | } |
29b734e4 NP |
718 | |
719 | entry = objects + nr_objects++; | |
c323ac7d | 720 | memset(entry, 0, sizeof(*entry)); |
aa7e44bf | 721 | hashcpy(entry->idx.sha1, sha1); |
27225f2e | 722 | entry->hash = hash; |
5c49c116 NP |
723 | if (type) |
724 | entry->type = type; | |
29b734e4 NP |
725 | if (exclude) |
726 | entry->preferred_base = 1; | |
81a216a5 NP |
727 | else |
728 | nr_result++; | |
29b734e4 NP |
729 | if (found_pack) { |
730 | entry->in_pack = found_pack; | |
731 | entry->in_pack_offset = found_offset; | |
732 | } | |
7a979d99 JH |
733 | |
734 | if (object_ix_hashsz * 3 <= nr_objects * 4) | |
735 | rehash_objects(); | |
29b734e4 NP |
736 | else |
737 | object_ix[-1 - ix] = nr_objects; | |
7a979d99 | 738 | |
4d4fcc54 | 739 | display_progress(progress_state, nr_objects); |
29b734e4 | 740 | |
a74db82e JH |
741 | if (name && no_try_delta(name)) |
742 | entry->no_try_delta = 1; | |
743 | ||
29b734e4 | 744 | return 1; |
c323ac7d LT |
745 | } |
746 | ||
5379a5c5 JH |
747 | struct pbase_tree_cache { |
748 | unsigned char sha1[20]; | |
749 | int ref; | |
750 | int temporary; | |
751 | void *tree_data; | |
752 | unsigned long tree_size; | |
753 | }; | |
754 | ||
755 | static struct pbase_tree_cache *(pbase_tree_cache[256]); | |
756 | static int pbase_tree_cache_ix(const unsigned char *sha1) | |
757 | { | |
758 | return sha1[0] % ARRAY_SIZE(pbase_tree_cache); | |
759 | } | |
760 | static int pbase_tree_cache_ix_incr(int ix) | |
761 | { | |
762 | return (ix+1) % ARRAY_SIZE(pbase_tree_cache); | |
763 | } | |
764 | ||
765 | static struct pbase_tree { | |
766 | struct pbase_tree *next; | |
767 | /* This is a phony "cache" entry; we are not | |
768 | * going to evict it nor find it through _get() | |
769 | * mechanism -- this is for the toplevel node that | |
770 | * would almost always change with any commit. | |
771 | */ | |
772 | struct pbase_tree_cache pcache; | |
773 | } *pbase_tree; | |
774 | ||
775 | static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1) | |
776 | { | |
777 | struct pbase_tree_cache *ent, *nent; | |
778 | void *data; | |
779 | unsigned long size; | |
21666f1a | 780 | enum object_type type; |
5379a5c5 JH |
781 | int neigh; |
782 | int my_ix = pbase_tree_cache_ix(sha1); | |
783 | int available_ix = -1; | |
784 | ||
785 | /* pbase-tree-cache acts as a limited hashtable. | |
786 | * your object will be found at your index or within a few | |
787 | * slots after that slot if it is cached. | |
788 | */ | |
789 | for (neigh = 0; neigh < 8; neigh++) { | |
790 | ent = pbase_tree_cache[my_ix]; | |
a89fccd2 | 791 | if (ent && !hashcmp(ent->sha1, sha1)) { |
5379a5c5 JH |
792 | ent->ref++; |
793 | return ent; | |
794 | } | |
795 | else if (((available_ix < 0) && (!ent || !ent->ref)) || | |
796 | ((0 <= available_ix) && | |
797 | (!ent && pbase_tree_cache[available_ix]))) | |
798 | available_ix = my_ix; | |
799 | if (!ent) | |
800 | break; | |
801 | my_ix = pbase_tree_cache_ix_incr(my_ix); | |
802 | } | |
803 | ||
804 | /* Did not find one. Either we got a bogus request or | |
805 | * we need to read and perhaps cache. | |
806 | */ | |
21666f1a | 807 | data = read_sha1_file(sha1, &type, &size); |
5379a5c5 JH |
808 | if (!data) |
809 | return NULL; | |
21666f1a | 810 | if (type != OBJ_TREE) { |
5379a5c5 JH |
811 | free(data); |
812 | return NULL; | |
813 | } | |
814 | ||
815 | /* We need to either cache or return a throwaway copy */ | |
816 | ||
817 | if (available_ix < 0) | |
818 | ent = NULL; | |
819 | else { | |
820 | ent = pbase_tree_cache[available_ix]; | |
821 | my_ix = available_ix; | |
822 | } | |
823 | ||
824 | if (!ent) { | |
825 | nent = xmalloc(sizeof(*nent)); | |
826 | nent->temporary = (available_ix < 0); | |
827 | } | |
828 | else { | |
829 | /* evict and reuse */ | |
830 | free(ent->tree_data); | |
831 | nent = ent; | |
832 | } | |
e702496e | 833 | hashcpy(nent->sha1, sha1); |
5379a5c5 JH |
834 | nent->tree_data = data; |
835 | nent->tree_size = size; | |
836 | nent->ref = 1; | |
837 | if (!nent->temporary) | |
838 | pbase_tree_cache[my_ix] = nent; | |
839 | return nent; | |
840 | } | |
841 | ||
842 | static void pbase_tree_put(struct pbase_tree_cache *cache) | |
843 | { | |
844 | if (!cache->temporary) { | |
845 | cache->ref--; | |
846 | return; | |
847 | } | |
848 | free(cache->tree_data); | |
849 | free(cache); | |
850 | } | |
851 | ||
852 | static int name_cmp_len(const char *name) | |
853 | { | |
854 | int i; | |
855 | for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++) | |
856 | ; | |
857 | return i; | |
858 | } | |
859 | ||
860 | static void add_pbase_object(struct tree_desc *tree, | |
5379a5c5 | 861 | const char *name, |
ce0bd642 LT |
862 | int cmplen, |
863 | const char *fullname) | |
3f9ac8d2 | 864 | { |
4c068a98 | 865 | struct name_entry entry; |
8a5a8d6c | 866 | int cmp; |
4c068a98 LT |
867 | |
868 | while (tree_entry(tree,&entry)) { | |
1211be6b LT |
869 | if (S_ISGITLINK(entry.mode)) |
870 | continue; | |
8a5a8d6c NP |
871 | cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 : |
872 | memcmp(name, entry.path, cmplen); | |
873 | if (cmp > 0) | |
7a979d99 | 874 | continue; |
8a5a8d6c NP |
875 | if (cmp < 0) |
876 | return; | |
5379a5c5 | 877 | if (name[cmplen] != '/') { |
5c49c116 | 878 | add_object_entry(entry.sha1, |
4d1012c3 | 879 | object_type(entry.mode), |
bc32fed5 | 880 | fullname, 1); |
5379a5c5 JH |
881 | return; |
882 | } | |
8a5a8d6c | 883 | if (S_ISDIR(entry.mode)) { |
7a979d99 | 884 | struct tree_desc sub; |
5379a5c5 JH |
885 | struct pbase_tree_cache *tree; |
886 | const char *down = name+cmplen+1; | |
887 | int downlen = name_cmp_len(down); | |
888 | ||
4c068a98 | 889 | tree = pbase_tree_get(entry.sha1); |
5379a5c5 JH |
890 | if (!tree) |
891 | return; | |
6fda5e51 | 892 | init_tree_desc(&sub, tree->tree_data, tree->tree_size); |
5379a5c5 | 893 | |
ce0bd642 | 894 | add_pbase_object(&sub, down, downlen, fullname); |
5379a5c5 JH |
895 | pbase_tree_put(tree); |
896 | } | |
897 | } | |
898 | } | |
1d6b38cc | 899 | |
5379a5c5 JH |
900 | static unsigned *done_pbase_paths; |
901 | static int done_pbase_paths_num; | |
902 | static int done_pbase_paths_alloc; | |
903 | static int done_pbase_path_pos(unsigned hash) | |
904 | { | |
905 | int lo = 0; | |
906 | int hi = done_pbase_paths_num; | |
907 | while (lo < hi) { | |
908 | int mi = (hi + lo) / 2; | |
909 | if (done_pbase_paths[mi] == hash) | |
910 | return mi; | |
911 | if (done_pbase_paths[mi] < hash) | |
912 | hi = mi; | |
913 | else | |
914 | lo = mi + 1; | |
915 | } | |
916 | return -lo-1; | |
917 | } | |
918 | ||
919 | static int check_pbase_path(unsigned hash) | |
920 | { | |
921 | int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash); | |
922 | if (0 <= pos) | |
923 | return 1; | |
924 | pos = -pos - 1; | |
925 | if (done_pbase_paths_alloc <= done_pbase_paths_num) { | |
926 | done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc); | |
927 | done_pbase_paths = xrealloc(done_pbase_paths, | |
928 | done_pbase_paths_alloc * | |
929 | sizeof(unsigned)); | |
930 | } | |
931 | done_pbase_paths_num++; | |
932 | if (pos < done_pbase_paths_num) | |
933 | memmove(done_pbase_paths + pos + 1, | |
934 | done_pbase_paths + pos, | |
935 | (done_pbase_paths_num - pos - 1) * sizeof(unsigned)); | |
936 | done_pbase_paths[pos] = hash; | |
937 | return 0; | |
938 | } | |
939 | ||
bc32fed5 | 940 | static void add_preferred_base_object(const char *name) |
5379a5c5 JH |
941 | { |
942 | struct pbase_tree *it; | |
8a5a8d6c | 943 | int cmplen; |
bc32fed5 | 944 | unsigned hash = name_hash(name); |
5379a5c5 | 945 | |
8a5a8d6c | 946 | if (!num_preferred_base || check_pbase_path(hash)) |
5379a5c5 JH |
947 | return; |
948 | ||
8a5a8d6c | 949 | cmplen = name_cmp_len(name); |
5379a5c5 JH |
950 | for (it = pbase_tree; it; it = it->next) { |
951 | if (cmplen == 0) { | |
bc32fed5 | 952 | add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1); |
5379a5c5 JH |
953 | } |
954 | else { | |
955 | struct tree_desc tree; | |
6fda5e51 | 956 | init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size); |
ce0bd642 | 957 | add_pbase_object(&tree, name, cmplen, name); |
7a979d99 | 958 | } |
3f9ac8d2 | 959 | } |
3f9ac8d2 JH |
960 | } |
961 | ||
7a979d99 | 962 | static void add_preferred_base(unsigned char *sha1) |
3f9ac8d2 | 963 | { |
5379a5c5 JH |
964 | struct pbase_tree *it; |
965 | void *data; | |
966 | unsigned long size; | |
967 | unsigned char tree_sha1[20]; | |
1d6b38cc | 968 | |
8d1d8f83 JH |
969 | if (window <= num_preferred_base++) |
970 | return; | |
971 | ||
5379a5c5 JH |
972 | data = read_object_with_reference(sha1, tree_type, &size, tree_sha1); |
973 | if (!data) | |
7a979d99 | 974 | return; |
5379a5c5 JH |
975 | |
976 | for (it = pbase_tree; it; it = it->next) { | |
a89fccd2 | 977 | if (!hashcmp(it->pcache.sha1, tree_sha1)) { |
5379a5c5 JH |
978 | free(data); |
979 | return; | |
980 | } | |
981 | } | |
982 | ||
983 | it = xcalloc(1, sizeof(*it)); | |
984 | it->next = pbase_tree; | |
985 | pbase_tree = it; | |
986 | ||
e702496e | 987 | hashcpy(it->pcache.sha1, tree_sha1); |
5379a5c5 JH |
988 | it->pcache.tree_data = data; |
989 | it->pcache.tree_size = size; | |
3f9ac8d2 JH |
990 | } |
991 | ||
c323ac7d LT |
992 | static void check_object(struct object_entry *entry) |
993 | { | |
5c49c116 | 994 | if (entry->in_pack) { |
780e6e73 | 995 | struct packed_git *p = entry->in_pack; |
03e79c88 | 996 | struct pack_window *w_curs = NULL; |
5c49c116 NP |
997 | const unsigned char *base_ref = NULL; |
998 | struct object_entry *base_entry; | |
999 | unsigned long used, used_0; | |
6777a59f | 1000 | unsigned int avail; |
5c49c116 NP |
1001 | off_t ofs; |
1002 | unsigned char *buf, c; | |
780e6e73 | 1003 | |
6777a59f | 1004 | buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail); |
ab7cd7bb | 1005 | |
5c49c116 | 1006 | /* |
fa736f72 NP |
1007 | * We want in_pack_type even if we do not reuse delta |
1008 | * since non-delta representations could still be reused. | |
ab7cd7bb | 1009 | */ |
6777a59f | 1010 | used = unpack_object_header_gently(buf, avail, |
5c49c116 NP |
1011 | &entry->in_pack_type, |
1012 | &entry->size); | |
ab7cd7bb | 1013 | |
5c49c116 NP |
1014 | /* |
1015 | * Determine if this is a delta and if so whether we can | |
1016 | * reuse it or not. Otherwise let's find out as cheaply as | |
1017 | * possible what the actual type and size for this object is. | |
3f9ac8d2 | 1018 | */ |
5c49c116 NP |
1019 | switch (entry->in_pack_type) { |
1020 | default: | |
1021 | /* Not a delta hence we've already got all we need. */ | |
1022 | entry->type = entry->in_pack_type; | |
1023 | entry->in_pack_header_size = used; | |
1024 | unuse_pack(&w_curs); | |
1025 | return; | |
1026 | case OBJ_REF_DELTA: | |
a7de7130 | 1027 | if (reuse_delta && !entry->preferred_base) |
5c49c116 NP |
1028 | base_ref = use_pack(p, &w_curs, |
1029 | entry->in_pack_offset + used, NULL); | |
1030 | entry->in_pack_header_size = used + 20; | |
1031 | break; | |
1032 | case OBJ_OFS_DELTA: | |
1033 | buf = use_pack(p, &w_curs, | |
1034 | entry->in_pack_offset + used, NULL); | |
1035 | used_0 = 0; | |
1036 | c = buf[used_0++]; | |
1037 | ofs = c & 127; | |
1038 | while (c & 128) { | |
1039 | ofs += 1; | |
1040 | if (!ofs || MSB(ofs, 7)) | |
1041 | die("delta base offset overflow in pack for %s", | |
aa7e44bf | 1042 | sha1_to_hex(entry->idx.sha1)); |
5c49c116 NP |
1043 | c = buf[used_0++]; |
1044 | ofs = (ofs << 7) + (c & 127); | |
780e6e73 | 1045 | } |
5c49c116 NP |
1046 | if (ofs >= entry->in_pack_offset) |
1047 | die("delta base offset out of bound for %s", | |
aa7e44bf | 1048 | sha1_to_hex(entry->idx.sha1)); |
5c49c116 | 1049 | ofs = entry->in_pack_offset - ofs; |
a7de7130 | 1050 | if (reuse_delta && !entry->preferred_base) { |
3449f8c4 NP |
1051 | struct revindex_entry *revidx; |
1052 | revidx = find_pack_revindex(p, ofs); | |
1053 | base_ref = nth_packed_object_sha1(p, revidx->nr); | |
1054 | } | |
5c49c116 NP |
1055 | entry->in_pack_header_size = used + used_0; |
1056 | break; | |
780e6e73 | 1057 | } |
780e6e73 | 1058 | |
5c49c116 NP |
1059 | if (base_ref && (base_entry = locate_object_entry(base_ref))) { |
1060 | /* | |
1061 | * If base_ref was set above that means we wish to | |
1062 | * reuse delta data, and we even found that base | |
1063 | * in the list of objects we want to pack. Goodie! | |
1064 | * | |
1065 | * Depth value does not matter - find_deltas() will | |
1066 | * never consider reused delta as the base object to | |
1067 | * deltify other objects against, in order to avoid | |
1068 | * circular deltas. | |
3f9ac8d2 | 1069 | */ |
780e6e73 | 1070 | entry->type = entry->in_pack_type; |
5c49c116 | 1071 | entry->delta = base_entry; |
15b4d577 JH |
1072 | entry->delta_sibling = base_entry->delta_child; |
1073 | base_entry->delta_child = entry; | |
5c49c116 NP |
1074 | unuse_pack(&w_curs); |
1075 | return; | |
1076 | } | |
ab7cd7bb | 1077 | |
5c49c116 NP |
1078 | if (entry->type) { |
1079 | /* | |
1080 | * This must be a delta and we already know what the | |
1081 | * final object type is. Let's extract the actual | |
1082 | * object size from the delta header. | |
1083 | */ | |
1084 | entry->size = get_size_from_delta(p, &w_curs, | |
1085 | entry->in_pack_offset + entry->in_pack_header_size); | |
1086 | unuse_pack(&w_curs); | |
3f9ac8d2 JH |
1087 | return; |
1088 | } | |
5c49c116 NP |
1089 | |
1090 | /* | |
1091 | * No choice but to fall back to the recursive delta walk | |
1092 | * with sha1_object_info() to find about the object type | |
1093 | * at this point... | |
1094 | */ | |
1095 | unuse_pack(&w_curs); | |
36e4d74a | 1096 | } |
3f9ac8d2 | 1097 | |
aa7e44bf | 1098 | entry->type = sha1_object_info(entry->idx.sha1, &entry->size); |
21666f1a | 1099 | if (entry->type < 0) |
36e4d74a | 1100 | die("unable to get type of object %s", |
aa7e44bf | 1101 | sha1_to_hex(entry->idx.sha1)); |
3f9ac8d2 JH |
1102 | } |
1103 | ||
5c49c116 NP |
1104 | static int pack_offset_sort(const void *_a, const void *_b) |
1105 | { | |
1106 | const struct object_entry *a = *(struct object_entry **)_a; | |
1107 | const struct object_entry *b = *(struct object_entry **)_b; | |
1108 | ||
1109 | /* avoid filesystem trashing with loose objects */ | |
1110 | if (!a->in_pack && !b->in_pack) | |
aa7e44bf | 1111 | return hashcmp(a->idx.sha1, b->idx.sha1); |
5c49c116 NP |
1112 | |
1113 | if (a->in_pack < b->in_pack) | |
1114 | return -1; | |
1115 | if (a->in_pack > b->in_pack) | |
1116 | return 1; | |
1117 | return a->in_pack_offset < b->in_pack_offset ? -1 : | |
1118 | (a->in_pack_offset > b->in_pack_offset); | |
1119 | } | |
1120 | ||
c323ac7d LT |
1121 | static void get_object_details(void) |
1122 | { | |
7cadf491 | 1123 | uint32_t i; |
5c49c116 NP |
1124 | struct object_entry **sorted_by_offset; |
1125 | ||
1126 | sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *)); | |
1127 | for (i = 0; i < nr_objects; i++) | |
1128 | sorted_by_offset[i] = objects + i; | |
1129 | qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort); | |
c323ac7d | 1130 | |
5c49c116 NP |
1131 | for (i = 0; i < nr_objects; i++) |
1132 | check_object(sorted_by_offset[i]); | |
3449f8c4 | 1133 | |
5c49c116 | 1134 | free(sorted_by_offset); |
c323ac7d LT |
1135 | } |
1136 | ||
b904166c NP |
1137 | /* |
1138 | * We search for deltas in a list sorted by type, by filename hash, and then | |
1139 | * by size, so that we see progressively smaller and smaller files. | |
1140 | * That's because we prefer deltas to be from the bigger file | |
1141 | * to the smaller -- deletes are potentially cheaper, but perhaps | |
1142 | * more importantly, the bigger file is likely the more recent | |
1143 | * one. The deepest deltas are therefore the oldest objects which are | |
1144 | * less susceptible to be accessed often. | |
1145 | */ | |
9668cf59 | 1146 | static int type_size_sort(const void *_a, const void *_b) |
c323ac7d | 1147 | { |
9668cf59 NP |
1148 | const struct object_entry *a = *(struct object_entry **)_a; |
1149 | const struct object_entry *b = *(struct object_entry **)_b; | |
1150 | ||
c323ac7d | 1151 | if (a->type > b->type) |
27225f2e | 1152 | return -1; |
b904166c | 1153 | if (a->type < b->type) |
27225f2e | 1154 | return 1; |
b904166c | 1155 | if (a->hash > b->hash) |
7a979d99 | 1156 | return -1; |
b904166c | 1157 | if (a->hash < b->hash) |
7a979d99 | 1158 | return 1; |
b904166c | 1159 | if (a->preferred_base > b->preferred_base) |
c323ac7d | 1160 | return -1; |
b904166c NP |
1161 | if (a->preferred_base < b->preferred_base) |
1162 | return 1; | |
c323ac7d | 1163 | if (a->size > b->size) |
b904166c NP |
1164 | return -1; |
1165 | if (a->size < b->size) | |
c323ac7d | 1166 | return 1; |
b904166c | 1167 | return a < b ? -1 : (a > b); /* newest first */ |
c323ac7d LT |
1168 | } |
1169 | ||
1170 | struct unpacked { | |
1171 | struct object_entry *entry; | |
1172 | void *data; | |
f6c7081a | 1173 | struct delta_index *index; |
5a235b5e | 1174 | unsigned depth; |
c323ac7d LT |
1175 | }; |
1176 | ||
d250626c NP |
1177 | static int delta_cacheable(unsigned long src_size, unsigned long trg_size, |
1178 | unsigned long delta_size) | |
074b2eea MK |
1179 | { |
1180 | if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size) | |
1181 | return 0; | |
1182 | ||
e3dfddb3 MK |
1183 | if (delta_size < cache_max_small_delta_size) |
1184 | return 1; | |
1185 | ||
074b2eea MK |
1186 | /* cache delta, if objects are large enough compared to delta size */ |
1187 | if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10)) | |
1188 | return 1; | |
1189 | ||
1190 | return 0; | |
1191 | } | |
1192 | ||
8ecce684 NP |
1193 | #ifdef THREADED_DELTA_SEARCH |
1194 | ||
1195 | static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER; | |
1196 | #define read_lock() pthread_mutex_lock(&read_mutex) | |
1197 | #define read_unlock() pthread_mutex_unlock(&read_mutex) | |
1198 | ||
3c701839 NP |
1199 | static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER; |
1200 | #define cache_lock() pthread_mutex_lock(&cache_mutex) | |
1201 | #define cache_unlock() pthread_mutex_unlock(&cache_mutex) | |
1202 | ||
8ecce684 NP |
1203 | static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER; |
1204 | #define progress_lock() pthread_mutex_lock(&progress_mutex) | |
1205 | #define progress_unlock() pthread_mutex_unlock(&progress_mutex) | |
1206 | ||
1207 | #else | |
1208 | ||
e1ef8673 JH |
1209 | #define read_lock() (void)0 |
1210 | #define read_unlock() (void)0 | |
1211 | #define cache_lock() (void)0 | |
1212 | #define cache_unlock() (void)0 | |
1213 | #define progress_lock() (void)0 | |
1214 | #define progress_unlock() (void)0 | |
8ecce684 NP |
1215 | |
1216 | #endif | |
1217 | ||
f6c7081a | 1218 | static int try_delta(struct unpacked *trg, struct unpacked *src, |
ef0316fc | 1219 | unsigned max_depth, unsigned long *mem_usage) |
c323ac7d | 1220 | { |
f6c7081a NP |
1221 | struct object_entry *trg_entry = trg->entry; |
1222 | struct object_entry *src_entry = src->entry; | |
560b25a8 | 1223 | unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz; |
c83f032e | 1224 | unsigned ref_depth; |
21666f1a | 1225 | enum object_type type; |
c323ac7d LT |
1226 | void *delta_buf; |
1227 | ||
1228 | /* Don't bother doing diffs between different types */ | |
f6c7081a | 1229 | if (trg_entry->type != src_entry->type) |
c323ac7d LT |
1230 | return -1; |
1231 | ||
51d1e83f LT |
1232 | /* |
1233 | * We do not bother to try a delta that we discarded | |
8dbbd14e | 1234 | * on an earlier try, but only when reusing delta data. |
51d1e83f | 1235 | */ |
a7de7130 | 1236 | if (reuse_delta && trg_entry->in_pack && |
e9195b58 JH |
1237 | trg_entry->in_pack == src_entry->in_pack && |
1238 | trg_entry->in_pack_type != OBJ_REF_DELTA && | |
1239 | trg_entry->in_pack_type != OBJ_OFS_DELTA) | |
51d1e83f LT |
1240 | return 0; |
1241 | ||
898b14ce | 1242 | /* Let's not bust the allowed depth. */ |
5a235b5e | 1243 | if (src->depth >= max_depth) |
d116a45a | 1244 | return 0; |
c323ac7d | 1245 | |
c3b06a69 | 1246 | /* Now some size filtering heuristics. */ |
560b25a8 | 1247 | trg_size = trg_entry->size; |
c83f032e NP |
1248 | if (!trg_entry->delta) { |
1249 | max_size = trg_size/2 - 20; | |
1250 | ref_depth = 1; | |
1251 | } else { | |
1252 | max_size = trg_entry->delta_size; | |
5a235b5e | 1253 | ref_depth = trg->depth; |
c83f032e | 1254 | } |
5a235b5e | 1255 | max_size = max_size * (max_depth - src->depth) / |
c83f032e | 1256 | (max_depth - ref_depth + 1); |
c3b06a69 NP |
1257 | if (max_size == 0) |
1258 | return 0; | |
f6c7081a | 1259 | src_size = src_entry->size; |
560b25a8 | 1260 | sizediff = src_size < trg_size ? trg_size - src_size : 0; |
27225f2e | 1261 | if (sizediff >= max_size) |
f527cb8c | 1262 | return 0; |
a1dab41a BD |
1263 | if (trg_size < src_size / 32) |
1264 | return 0; | |
f6c7081a | 1265 | |
560b25a8 NP |
1266 | /* Load data if not already done */ |
1267 | if (!trg->data) { | |
8ecce684 | 1268 | read_lock(); |
aa7e44bf | 1269 | trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz); |
8ecce684 | 1270 | read_unlock(); |
2e3404c3 JH |
1271 | if (!trg->data) |
1272 | die("object %s cannot be read", | |
1273 | sha1_to_hex(trg_entry->idx.sha1)); | |
560b25a8 NP |
1274 | if (sz != trg_size) |
1275 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1276 | sha1_to_hex(trg_entry->idx.sha1), sz, trg_size); |
ef0316fc | 1277 | *mem_usage += sz; |
560b25a8 NP |
1278 | } |
1279 | if (!src->data) { | |
8ecce684 | 1280 | read_lock(); |
aa7e44bf | 1281 | src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz); |
8ecce684 | 1282 | read_unlock(); |
2e3404c3 JH |
1283 | if (!src->data) |
1284 | die("object %s cannot be read", | |
1285 | sha1_to_hex(src_entry->idx.sha1)); | |
560b25a8 NP |
1286 | if (sz != src_size) |
1287 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1288 | sha1_to_hex(src_entry->idx.sha1), sz, src_size); |
ef0316fc | 1289 | *mem_usage += sz; |
560b25a8 NP |
1290 | } |
1291 | if (!src->index) { | |
1292 | src->index = create_delta_index(src->data, src_size); | |
a588d88a MK |
1293 | if (!src->index) { |
1294 | static int warned = 0; | |
1295 | if (!warned++) | |
1296 | warning("suboptimal pack - out of memory"); | |
1297 | return 0; | |
1298 | } | |
ef0316fc | 1299 | *mem_usage += sizeof_delta_index(src->index); |
560b25a8 NP |
1300 | } |
1301 | ||
1302 | delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size); | |
c323ac7d | 1303 | if (!delta_buf) |
75c42d8c | 1304 | return 0; |
f6c7081a | 1305 | |
9e2d57a0 | 1306 | if (trg_entry->delta) { |
848d732c BD |
1307 | /* Prefer only shallower same-sized deltas. */ |
1308 | if (delta_size == trg_entry->delta_size && | |
5a235b5e | 1309 | src->depth + 1 >= trg->depth) { |
848d732c BD |
1310 | free(delta_buf); |
1311 | return 0; | |
1312 | } | |
074b2eea | 1313 | } |
9e2d57a0 | 1314 | |
3c701839 NP |
1315 | /* |
1316 | * Handle memory allocation outside of the cache | |
1317 | * accounting lock. Compiler will optimize the strangeness | |
1318 | * away when THREADED_DELTA_SEARCH is not defined. | |
1319 | */ | |
8e0f7003 | 1320 | free(trg_entry->delta_data); |
3c701839 | 1321 | cache_lock(); |
9e2d57a0 NP |
1322 | if (trg_entry->delta_data) { |
1323 | delta_cache_size -= trg_entry->delta_size; | |
9e2d57a0 NP |
1324 | trg_entry->delta_data = NULL; |
1325 | } | |
d250626c | 1326 | if (delta_cacheable(src_size, trg_size, delta_size)) { |
b7a28f78 | 1327 | delta_cache_size += delta_size; |
3c701839 NP |
1328 | cache_unlock(); |
1329 | trg_entry->delta_data = xrealloc(delta_buf, delta_size); | |
1330 | } else { | |
1331 | cache_unlock(); | |
074b2eea | 1332 | free(delta_buf); |
3c701839 NP |
1333 | } |
1334 | ||
b7a28f78 NP |
1335 | trg_entry->delta = src_entry; |
1336 | trg_entry->delta_size = delta_size; | |
1337 | trg->depth = src->depth + 1; | |
1338 | ||
f6c7081a | 1339 | return 1; |
c323ac7d LT |
1340 | } |
1341 | ||
898b14ce | 1342 | static unsigned int check_delta_limit(struct object_entry *me, unsigned int n) |
b2504a0d | 1343 | { |
898b14ce NP |
1344 | struct object_entry *child = me->delta_child; |
1345 | unsigned int m = n; | |
1346 | while (child) { | |
1347 | unsigned int c = check_delta_limit(child, n + 1); | |
1348 | if (m < c) | |
1349 | m = c; | |
1350 | child = child->delta_sibling; | |
1351 | } | |
1352 | return m; | |
b2504a0d NP |
1353 | } |
1354 | ||
75ad235c | 1355 | static unsigned long free_unpacked(struct unpacked *n) |
a97773ce | 1356 | { |
ef0316fc | 1357 | unsigned long freed_mem = sizeof_delta_index(n->index); |
a97773ce BD |
1358 | free_delta_index(n->index); |
1359 | n->index = NULL; | |
1360 | if (n->data) { | |
ef0316fc | 1361 | freed_mem += n->entry->size; |
a97773ce BD |
1362 | free(n->data); |
1363 | n->data = NULL; | |
a97773ce BD |
1364 | } |
1365 | n->entry = NULL; | |
7d7baa5e | 1366 | n->depth = 0; |
ef0316fc | 1367 | return freed_mem; |
a97773ce BD |
1368 | } |
1369 | ||
384b32c0 | 1370 | static void find_deltas(struct object_entry **list, unsigned *list_size, |
e334977d | 1371 | int window, int depth, unsigned *processed) |
c323ac7d | 1372 | { |
384b32c0 | 1373 | uint32_t i, idx = 0, count = 0; |
c323ac7d | 1374 | unsigned int array_size = window * sizeof(struct unpacked); |
7cadf491 | 1375 | struct unpacked *array; |
ef0316fc | 1376 | unsigned long mem_usage = 0; |
c323ac7d | 1377 | |
7cadf491 | 1378 | array = xmalloc(array_size); |
c323ac7d | 1379 | memset(array, 0, array_size); |
21fcd1bd | 1380 | |
384b32c0 NP |
1381 | for (;;) { |
1382 | struct object_entry *entry = *list++; | |
c323ac7d | 1383 | struct unpacked *n = array + idx; |
ef0316fc | 1384 | int j, max_depth, best_base = -1; |
c323ac7d | 1385 | |
384b32c0 NP |
1386 | progress_lock(); |
1387 | if (!*list_size) { | |
1388 | progress_unlock(); | |
1389 | break; | |
1390 | } | |
1391 | (*list_size)--; | |
1392 | if (!entry->preferred_base) { | |
1393 | (*processed)++; | |
1394 | display_progress(progress_state, *processed); | |
1395 | } | |
1396 | progress_unlock(); | |
1397 | ||
ef0316fc | 1398 | mem_usage -= free_unpacked(n); |
c323ac7d | 1399 | n->entry = entry; |
ab7cd7bb | 1400 | |
a97773ce | 1401 | while (window_memory_limit && |
ef0316fc | 1402 | mem_usage > window_memory_limit && |
a97773ce BD |
1403 | count > 1) { |
1404 | uint32_t tail = (idx + window - count) % window; | |
75ad235c | 1405 | mem_usage -= free_unpacked(array + tail); |
a97773ce BD |
1406 | count--; |
1407 | } | |
1408 | ||
75d39853 NP |
1409 | /* We do not compute delta to *create* objects we are not |
1410 | * going to pack. | |
1411 | */ | |
1412 | if (entry->preferred_base) | |
1413 | goto next; | |
1414 | ||
898b14ce NP |
1415 | /* |
1416 | * If the current object is at pack edge, take the depth the | |
1417 | * objects that depend on the current object into account | |
1418 | * otherwise they would become too deep. | |
1419 | */ | |
1420 | max_depth = depth; | |
1421 | if (entry->delta_child) { | |
1422 | max_depth -= check_delta_limit(entry, 0); | |
1423 | if (max_depth <= 0) | |
1424 | goto next; | |
1425 | } | |
1426 | ||
78817c15 LT |
1427 | j = window; |
1428 | while (--j > 0) { | |
77639870 | 1429 | int ret; |
7cadf491 | 1430 | uint32_t other_idx = idx + j; |
c323ac7d | 1431 | struct unpacked *m; |
78817c15 LT |
1432 | if (other_idx >= window) |
1433 | other_idx -= window; | |
c323ac7d LT |
1434 | m = array + other_idx; |
1435 | if (!m->entry) | |
1436 | break; | |
ef0316fc | 1437 | ret = try_delta(n, m, max_depth, &mem_usage); |
77639870 | 1438 | if (ret < 0) |
c323ac7d | 1439 | break; |
77639870 JH |
1440 | else if (ret > 0) |
1441 | best_base = other_idx; | |
c323ac7d | 1442 | } |
898b14ce | 1443 | |
ed4a9031 NP |
1444 | /* |
1445 | * If we decided to cache the delta data, then it is best | |
1446 | * to compress it right away. First because we have to do | |
1447 | * it anyway, and doing it here while we're threaded will | |
1448 | * save a lot of time in the non threaded write phase, | |
1449 | * as well as allow for caching more deltas within | |
1450 | * the same cache size limit. | |
1451 | * ... | |
1452 | * But only if not writing to stdout, since in that case | |
1453 | * the network is most likely throttling writes anyway, | |
1454 | * and therefore it is best to go to the write phase ASAP | |
1455 | * instead, as we can afford spending more time compressing | |
1456 | * between writes at that moment. | |
1457 | */ | |
1458 | if (entry->delta_data && !pack_to_stdout) { | |
1459 | entry->z_delta_size = do_compress(&entry->delta_data, | |
1460 | entry->delta_size); | |
1461 | cache_lock(); | |
1462 | delta_cache_size -= entry->delta_size; | |
1463 | delta_cache_size += entry->z_delta_size; | |
1464 | cache_unlock(); | |
1465 | } | |
1466 | ||
70ca1a3f JH |
1467 | /* if we made n a delta, and if n is already at max |
1468 | * depth, leaving it in the window is pointless. we | |
1469 | * should evict it first. | |
70ca1a3f | 1470 | */ |
70baf5d4 | 1471 | if (entry->delta && max_depth <= n->depth) |
70ca1a3f | 1472 | continue; |
ff45715c | 1473 | |
77639870 JH |
1474 | /* |
1475 | * Move the best delta base up in the window, after the | |
1476 | * currently deltified object, to keep it longer. It will | |
1477 | * be the first base object to be attempted next. | |
1478 | */ | |
1479 | if (entry->delta) { | |
1480 | struct unpacked swap = array[best_base]; | |
1481 | int dist = (window + idx - best_base) % window; | |
1482 | int dst = best_base; | |
1483 | while (dist--) { | |
1484 | int src = (dst + 1) % window; | |
1485 | array[dst] = array[src]; | |
1486 | dst = src; | |
1487 | } | |
1488 | array[dst] = swap; | |
1489 | } | |
1490 | ||
898b14ce | 1491 | next: |
521a4f4c | 1492 | idx++; |
a97773ce BD |
1493 | if (count + 1 < window) |
1494 | count++; | |
521a4f4c LT |
1495 | if (idx >= window) |
1496 | idx = 0; | |
384b32c0 | 1497 | } |
adee7bdf | 1498 | |
f6c7081a | 1499 | for (i = 0; i < window; ++i) { |
ff45715c | 1500 | free_delta_index(array[i].index); |
adee7bdf | 1501 | free(array[i].data); |
f6c7081a | 1502 | } |
adee7bdf | 1503 | free(array); |
c323ac7d LT |
1504 | } |
1505 | ||
8ecce684 NP |
1506 | #ifdef THREADED_DELTA_SEARCH |
1507 | ||
50f22ada JS |
1508 | /* |
1509 | * The main thread waits on the condition that (at least) one of the workers | |
1510 | * has stopped working (which is indicated in the .working member of | |
1511 | * struct thread_params). | |
1512 | * When a work thread has completed its work, it sets .working to 0 and | |
1513 | * signals the main thread and waits on the condition that .data_ready | |
1514 | * becomes 1. | |
1515 | */ | |
1516 | ||
8ecce684 NP |
1517 | struct thread_params { |
1518 | pthread_t thread; | |
1519 | struct object_entry **list; | |
1520 | unsigned list_size; | |
384b32c0 | 1521 | unsigned remaining; |
8ecce684 NP |
1522 | int window; |
1523 | int depth; | |
50f22ada JS |
1524 | int working; |
1525 | int data_ready; | |
1526 | pthread_mutex_t mutex; | |
1527 | pthread_cond_t cond; | |
8ecce684 NP |
1528 | unsigned *processed; |
1529 | }; | |
1530 | ||
50f22ada | 1531 | static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER; |
c2a33679 | 1532 | |
8ecce684 NP |
1533 | static void *threaded_find_deltas(void *arg) |
1534 | { | |
c2a33679 NP |
1535 | struct thread_params *me = arg; |
1536 | ||
50f22ada | 1537 | while (me->remaining) { |
384b32c0 | 1538 | find_deltas(me->list, &me->remaining, |
c2a33679 | 1539 | me->window, me->depth, me->processed); |
50f22ada JS |
1540 | |
1541 | progress_lock(); | |
1542 | me->working = 0; | |
1543 | pthread_cond_signal(&progress_cond); | |
1544 | progress_unlock(); | |
1545 | ||
1546 | /* | |
1547 | * We must not set ->data_ready before we wait on the | |
1548 | * condition because the main thread may have set it to 1 | |
1549 | * before we get here. In order to be sure that new | |
1550 | * work is available if we see 1 in ->data_ready, it | |
1551 | * was initialized to 0 before this thread was spawned | |
1552 | * and we reset it to 0 right away. | |
1553 | */ | |
1554 | pthread_mutex_lock(&me->mutex); | |
1555 | while (!me->data_ready) | |
1556 | pthread_cond_wait(&me->cond, &me->mutex); | |
1557 | me->data_ready = 0; | |
1558 | pthread_mutex_unlock(&me->mutex); | |
c2a33679 | 1559 | } |
50f22ada JS |
1560 | /* leave ->working 1 so that this doesn't get more work assigned */ |
1561 | return NULL; | |
8ecce684 NP |
1562 | } |
1563 | ||
8ecce684 NP |
1564 | static void ll_find_deltas(struct object_entry **list, unsigned list_size, |
1565 | int window, int depth, unsigned *processed) | |
1566 | { | |
50f22ada | 1567 | struct thread_params p[delta_search_threads]; |
384b32c0 | 1568 | int i, ret, active_threads = 0; |
c2a33679 | 1569 | |
367f4a43 | 1570 | if (delta_search_threads <= 1) { |
384b32c0 | 1571 | find_deltas(list, &list_size, window, depth, processed); |
367f4a43 NP |
1572 | return; |
1573 | } | |
1574 | ||
50f22ada | 1575 | /* Partition the work amongst work threads. */ |
367f4a43 | 1576 | for (i = 0; i < delta_search_threads; i++) { |
50f22ada JS |
1577 | unsigned sub_size = list_size / (delta_search_threads - i); |
1578 | ||
8ecce684 NP |
1579 | p[i].window = window; |
1580 | p[i].depth = depth; | |
1581 | p[i].processed = processed; | |
50f22ada JS |
1582 | p[i].working = 1; |
1583 | p[i].data_ready = 0; | |
c2a33679 | 1584 | |
59921b4b | 1585 | /* try to split chunks on "path" boundaries */ |
6fc74703 NP |
1586 | while (sub_size && sub_size < list_size && |
1587 | list[sub_size]->hash && | |
384b32c0 NP |
1588 | list[sub_size]->hash == list[sub_size-1]->hash) |
1589 | sub_size++; | |
1590 | ||
50f22ada JS |
1591 | p[i].list = list; |
1592 | p[i].list_size = sub_size; | |
1593 | p[i].remaining = sub_size; | |
59921b4b | 1594 | |
384b32c0 NP |
1595 | list += sub_size; |
1596 | list_size -= sub_size; | |
1597 | } | |
1598 | ||
50f22ada JS |
1599 | /* Start work threads. */ |
1600 | for (i = 0; i < delta_search_threads; i++) { | |
1601 | if (!p[i].list_size) | |
1602 | continue; | |
68e6a4f8 JS |
1603 | pthread_mutex_init(&p[i].mutex, NULL); |
1604 | pthread_cond_init(&p[i].cond, NULL); | |
50f22ada JS |
1605 | ret = pthread_create(&p[i].thread, NULL, |
1606 | threaded_find_deltas, &p[i]); | |
1607 | if (ret) | |
1608 | die("unable to create thread: %s", strerror(ret)); | |
1609 | active_threads++; | |
1610 | } | |
1611 | ||
384b32c0 NP |
1612 | /* |
1613 | * Now let's wait for work completion. Each time a thread is done | |
1614 | * with its work, we steal half of the remaining work from the | |
1615 | * thread with the largest number of unprocessed objects and give | |
1616 | * it to that newly idle thread. This ensure good load balancing | |
1617 | * until the remaining object list segments are simply too short | |
1618 | * to be worth splitting anymore. | |
1619 | */ | |
50f22ada JS |
1620 | while (active_threads) { |
1621 | struct thread_params *target = NULL; | |
384b32c0 NP |
1622 | struct thread_params *victim = NULL; |
1623 | unsigned sub_size = 0; | |
384b32c0 NP |
1624 | |
1625 | progress_lock(); | |
50f22ada JS |
1626 | for (;;) { |
1627 | for (i = 0; !target && i < delta_search_threads; i++) | |
1628 | if (!p[i].working) | |
1629 | target = &p[i]; | |
1630 | if (target) | |
1631 | break; | |
1632 | pthread_cond_wait(&progress_cond, &progress_mutex); | |
1633 | } | |
1634 | ||
384b32c0 NP |
1635 | for (i = 0; i < delta_search_threads; i++) |
1636 | if (p[i].remaining > 2*window && | |
1637 | (!victim || victim->remaining < p[i].remaining)) | |
1638 | victim = &p[i]; | |
1639 | if (victim) { | |
1640 | sub_size = victim->remaining / 2; | |
1641 | list = victim->list + victim->list_size - sub_size; | |
1642 | while (sub_size && list[0]->hash && | |
1643 | list[0]->hash == list[-1]->hash) { | |
1644 | list++; | |
1645 | sub_size--; | |
1646 | } | |
eb9688ff NP |
1647 | if (!sub_size) { |
1648 | /* | |
1649 | * It is possible for some "paths" to have | |
1650 | * so many objects that no hash boundary | |
1651 | * might be found. Let's just steal the | |
1652 | * exact half in that case. | |
1653 | */ | |
1654 | sub_size = victim->remaining / 2; | |
1655 | list -= sub_size; | |
1656 | } | |
384b32c0 NP |
1657 | target->list = list; |
1658 | victim->list_size -= sub_size; | |
1659 | victim->remaining -= sub_size; | |
1660 | } | |
384b32c0 NP |
1661 | target->list_size = sub_size; |
1662 | target->remaining = sub_size; | |
50f22ada JS |
1663 | target->working = 1; |
1664 | progress_unlock(); | |
1665 | ||
1666 | pthread_mutex_lock(&target->mutex); | |
1667 | target->data_ready = 1; | |
1668 | pthread_cond_signal(&target->cond); | |
1669 | pthread_mutex_unlock(&target->mutex); | |
c2a33679 | 1670 | |
384b32c0 | 1671 | if (!sub_size) { |
b81d9af7 | 1672 | pthread_join(target->thread, NULL); |
50f22ada JS |
1673 | pthread_cond_destroy(&target->cond); |
1674 | pthread_mutex_destroy(&target->mutex); | |
384b32c0 | 1675 | active_threads--; |
c2a33679 | 1676 | } |
50f22ada | 1677 | } |
8ecce684 NP |
1678 | } |
1679 | ||
1680 | #else | |
384b32c0 | 1681 | #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p) |
8ecce684 NP |
1682 | #endif |
1683 | ||
f0a24aa5 SP |
1684 | static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data) |
1685 | { | |
1686 | unsigned char peeled[20]; | |
1687 | ||
1688 | if (!prefixcmp(path, "refs/tags/") && /* is a tag? */ | |
1689 | !peel_ref(path, peeled) && /* peelable? */ | |
1690 | !is_null_sha1(peeled) && /* annotated tag? */ | |
1691 | locate_object_entry(peeled)) /* object packed? */ | |
1692 | add_object_entry(sha1, OBJ_TAG, NULL, 0); | |
1693 | return 0; | |
1694 | } | |
1695 | ||
f3123c4a JH |
1696 | static void prepare_pack(int window, int depth) |
1697 | { | |
9668cf59 | 1698 | struct object_entry **delta_list; |
6e1c2344 RJ |
1699 | uint32_t i, nr_deltas; |
1700 | unsigned n; | |
9668cf59 | 1701 | |
3f9ac8d2 | 1702 | get_object_details(); |
9668cf59 | 1703 | |
75d39853 | 1704 | if (!nr_objects || !window || !depth) |
9668cf59 NP |
1705 | return; |
1706 | ||
1707 | delta_list = xmalloc(nr_objects * sizeof(*delta_list)); | |
75d39853 NP |
1708 | nr_deltas = n = 0; |
1709 | ||
1710 | for (i = 0; i < nr_objects; i++) { | |
1711 | struct object_entry *entry = objects + i; | |
1712 | ||
1713 | if (entry->delta) | |
1714 | /* This happens if we decided to reuse existing | |
a7de7130 | 1715 | * delta from a pack. "reuse_delta &&" is implied. |
75d39853 NP |
1716 | */ |
1717 | continue; | |
1718 | ||
1719 | if (entry->size < 50) | |
1720 | continue; | |
1721 | ||
1722 | if (entry->no_try_delta) | |
1723 | continue; | |
1724 | ||
1725 | if (!entry->preferred_base) | |
1726 | nr_deltas++; | |
1727 | ||
1728 | delta_list[n++] = entry; | |
1729 | } | |
1730 | ||
2f8b8947 | 1731 | if (nr_deltas && n > 1) { |
e334977d NP |
1732 | unsigned nr_done = 0; |
1733 | if (progress) | |
dc6a0757 NP |
1734 | progress_state = start_progress("Compressing objects", |
1735 | nr_deltas); | |
75d39853 | 1736 | qsort(delta_list, n, sizeof(*delta_list), type_size_sort); |
8ecce684 | 1737 | ll_find_deltas(delta_list, n, window+1, depth, &nr_done); |
4d4fcc54 | 1738 | stop_progress(&progress_state); |
e334977d NP |
1739 | if (nr_done != nr_deltas) |
1740 | die("inconsistency with delta count"); | |
75d39853 | 1741 | } |
9668cf59 | 1742 | free(delta_list); |
f3123c4a JH |
1743 | } |
1744 | ||
ef90d6d4 | 1745 | static int git_pack_config(const char *k, const char *v, void *cb) |
4812a93a JK |
1746 | { |
1747 | if(!strcmp(k, "pack.window")) { | |
1748 | window = git_config_int(k, v); | |
1749 | return 0; | |
1750 | } | |
a97773ce BD |
1751 | if (!strcmp(k, "pack.windowmemory")) { |
1752 | window_memory_limit = git_config_ulong(k, v); | |
1753 | return 0; | |
1754 | } | |
1755 | if (!strcmp(k, "pack.depth")) { | |
842aaf93 TT |
1756 | depth = git_config_int(k, v); |
1757 | return 0; | |
1758 | } | |
960ccca6 DH |
1759 | if (!strcmp(k, "pack.compression")) { |
1760 | int level = git_config_int(k, v); | |
1761 | if (level == -1) | |
1762 | level = Z_DEFAULT_COMPRESSION; | |
1763 | else if (level < 0 || level > Z_BEST_COMPRESSION) | |
1764 | die("bad pack compression level %d", level); | |
1765 | pack_compression_level = level; | |
1766 | pack_compression_seen = 1; | |
1767 | return 0; | |
1768 | } | |
074b2eea MK |
1769 | if (!strcmp(k, "pack.deltacachesize")) { |
1770 | max_delta_cache_size = git_config_int(k, v); | |
1771 | return 0; | |
1772 | } | |
e3dfddb3 MK |
1773 | if (!strcmp(k, "pack.deltacachelimit")) { |
1774 | cache_max_small_delta_size = git_config_int(k, v); | |
1775 | return 0; | |
1776 | } | |
693b86ff NP |
1777 | if (!strcmp(k, "pack.threads")) { |
1778 | delta_search_threads = git_config_int(k, v); | |
833e3df1 | 1779 | if (delta_search_threads < 0) |
693b86ff NP |
1780 | die("invalid number of threads specified (%d)", |
1781 | delta_search_threads); | |
1782 | #ifndef THREADED_DELTA_SEARCH | |
833e3df1 | 1783 | if (delta_search_threads != 1) |
693b86ff NP |
1784 | warning("no threads support, ignoring %s", k); |
1785 | #endif | |
1786 | return 0; | |
1787 | } | |
4d00bda2 NP |
1788 | if (!strcmp(k, "pack.indexversion")) { |
1789 | pack_idx_default_version = git_config_int(k, v); | |
1790 | if (pack_idx_default_version > 2) | |
6e1c2344 RJ |
1791 | die("bad pack.indexversion=%"PRIu32, |
1792 | pack_idx_default_version); | |
4d00bda2 NP |
1793 | return 0; |
1794 | } | |
2b84b5a8 JS |
1795 | if (!strcmp(k, "pack.packsizelimit")) { |
1796 | pack_size_limit_cfg = git_config_ulong(k, v); | |
1797 | return 0; | |
1798 | } | |
ef90d6d4 | 1799 | return git_default_config(k, v, cb); |
4812a93a JK |
1800 | } |
1801 | ||
b5d97e6b | 1802 | static void read_object_list_from_stdin(void) |
c323ac7d | 1803 | { |
b5d97e6b JH |
1804 | char line[40 + 1 + PATH_MAX + 2]; |
1805 | unsigned char sha1[20]; | |
b2504a0d | 1806 | |
da93d12b | 1807 | for (;;) { |
da93d12b LT |
1808 | if (!fgets(line, sizeof(line), stdin)) { |
1809 | if (feof(stdin)) | |
1810 | break; | |
1811 | if (!ferror(stdin)) | |
1812 | die("fgets returned NULL, not EOF, not error!"); | |
687dd75c JH |
1813 | if (errno != EINTR) |
1814 | die("fgets: %s", strerror(errno)); | |
1815 | clearerr(stdin); | |
1816 | continue; | |
da93d12b | 1817 | } |
7a979d99 JH |
1818 | if (line[0] == '-') { |
1819 | if (get_sha1_hex(line+1, sha1)) | |
1820 | die("expected edge sha1, got garbage:\n %s", | |
b5d97e6b | 1821 | line); |
8d1d8f83 | 1822 | add_preferred_base(sha1); |
7a979d99 | 1823 | continue; |
21fcd1bd | 1824 | } |
c323ac7d | 1825 | if (get_sha1_hex(line, sha1)) |
ef07618f | 1826 | die("expected sha1, got garbage:\n %s", line); |
b5d97e6b | 1827 | |
bc32fed5 JH |
1828 | add_preferred_base_object(line+41); |
1829 | add_object_entry(sha1, 0, line+41, 0); | |
c323ac7d | 1830 | } |
b5d97e6b JH |
1831 | } |
1832 | ||
08cdfb13 JH |
1833 | #define OBJECT_ADDED (1u<<20) |
1834 | ||
b5d97e6b JH |
1835 | static void show_commit(struct commit *commit) |
1836 | { | |
bc32fed5 | 1837 | add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0); |
08cdfb13 | 1838 | commit->object.flags |= OBJECT_ADDED; |
b5d97e6b JH |
1839 | } |
1840 | ||
1841 | static void show_object(struct object_array_entry *p) | |
1842 | { | |
bc32fed5 JH |
1843 | add_preferred_base_object(p->name); |
1844 | add_object_entry(p->item->sha1, p->item->type, p->name, 0); | |
08cdfb13 | 1845 | p->item->flags |= OBJECT_ADDED; |
b5d97e6b JH |
1846 | } |
1847 | ||
8d1d8f83 JH |
1848 | static void show_edge(struct commit *commit) |
1849 | { | |
1850 | add_preferred_base(commit->object.sha1); | |
1851 | } | |
1852 | ||
08cdfb13 JH |
1853 | struct in_pack_object { |
1854 | off_t offset; | |
1855 | struct object *object; | |
1856 | }; | |
1857 | ||
1858 | struct in_pack { | |
1859 | int alloc; | |
1860 | int nr; | |
1861 | struct in_pack_object *array; | |
1862 | }; | |
1863 | ||
1864 | static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack) | |
1865 | { | |
1866 | in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p); | |
1867 | in_pack->array[in_pack->nr].object = object; | |
1868 | in_pack->nr++; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * Compare the objects in the offset order, in order to emulate the | |
1873 | * "git-rev-list --objects" output that produced the pack originally. | |
1874 | */ | |
1875 | static int ofscmp(const void *a_, const void *b_) | |
1876 | { | |
1877 | struct in_pack_object *a = (struct in_pack_object *)a_; | |
1878 | struct in_pack_object *b = (struct in_pack_object *)b_; | |
1879 | ||
1880 | if (a->offset < b->offset) | |
1881 | return -1; | |
1882 | else if (a->offset > b->offset) | |
1883 | return 1; | |
1884 | else | |
1885 | return hashcmp(a->object->sha1, b->object->sha1); | |
1886 | } | |
1887 | ||
1888 | static void add_objects_in_unpacked_packs(struct rev_info *revs) | |
1889 | { | |
1890 | struct packed_git *p; | |
1891 | struct in_pack in_pack; | |
1892 | uint32_t i; | |
1893 | ||
1894 | memset(&in_pack, 0, sizeof(in_pack)); | |
1895 | ||
1896 | for (p = packed_git; p; p = p->next) { | |
1897 | const unsigned char *sha1; | |
1898 | struct object *o; | |
1899 | ||
1900 | for (i = 0; i < revs->num_ignore_packed; i++) { | |
1901 | if (matches_pack_name(p, revs->ignore_packed[i])) | |
1902 | break; | |
1903 | } | |
1904 | if (revs->num_ignore_packed <= i) | |
1905 | continue; | |
1906 | if (open_pack_index(p)) | |
1907 | die("cannot open pack index"); | |
1908 | ||
1909 | ALLOC_GROW(in_pack.array, | |
1910 | in_pack.nr + p->num_objects, | |
1911 | in_pack.alloc); | |
1912 | ||
1913 | for (i = 0; i < p->num_objects; i++) { | |
1914 | sha1 = nth_packed_object_sha1(p, i); | |
1915 | o = lookup_unknown_object(sha1); | |
1916 | if (!(o->flags & OBJECT_ADDED)) | |
1917 | mark_in_pack_object(o, p, &in_pack); | |
1918 | o->flags |= OBJECT_ADDED; | |
1919 | } | |
1920 | } | |
1921 | ||
1922 | if (in_pack.nr) { | |
1923 | qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]), | |
1924 | ofscmp); | |
1925 | for (i = 0; i < in_pack.nr; i++) { | |
1926 | struct object *o = in_pack.array[i].object; | |
1927 | add_object_entry(o->sha1, o->type, "", 0); | |
1928 | } | |
1929 | } | |
1930 | free(in_pack.array); | |
1931 | } | |
1932 | ||
ca11b212 NP |
1933 | static void loosen_unused_packed_objects(struct rev_info *revs) |
1934 | { | |
1935 | struct packed_git *p; | |
1936 | uint32_t i; | |
1937 | const unsigned char *sha1; | |
1938 | ||
1939 | for (p = packed_git; p; p = p->next) { | |
1940 | for (i = 0; i < revs->num_ignore_packed; i++) { | |
1941 | if (matches_pack_name(p, revs->ignore_packed[i])) | |
1942 | break; | |
1943 | } | |
1944 | if (revs->num_ignore_packed <= i) | |
1945 | continue; | |
1946 | ||
1947 | if (open_pack_index(p)) | |
1948 | die("cannot open pack index"); | |
1949 | ||
1950 | for (i = 0; i < p->num_objects; i++) { | |
1951 | sha1 = nth_packed_object_sha1(p, i); | |
1952 | if (!locate_object_entry(sha1)) | |
1953 | if (force_object_loose(sha1, p->mtime)) | |
1954 | die("unable to force loose object"); | |
1955 | } | |
1956 | } | |
1957 | } | |
1958 | ||
8d1d8f83 | 1959 | static void get_object_list(int ac, const char **av) |
b5d97e6b JH |
1960 | { |
1961 | struct rev_info revs; | |
1962 | char line[1000]; | |
b5d97e6b JH |
1963 | int flags = 0; |
1964 | ||
b5d97e6b JH |
1965 | init_revisions(&revs, NULL); |
1966 | save_commit_buffer = 0; | |
b5d97e6b JH |
1967 | setup_revisions(ac, av, &revs, NULL); |
1968 | ||
b5d97e6b JH |
1969 | while (fgets(line, sizeof(line), stdin) != NULL) { |
1970 | int len = strlen(line); | |
872c930d | 1971 | if (len && line[len - 1] == '\n') |
b5d97e6b JH |
1972 | line[--len] = 0; |
1973 | if (!len) | |
1974 | break; | |
1975 | if (*line == '-') { | |
1976 | if (!strcmp(line, "--not")) { | |
1977 | flags ^= UNINTERESTING; | |
1978 | continue; | |
1979 | } | |
1980 | die("not a rev '%s'", line); | |
1981 | } | |
1982 | if (handle_revision_arg(line, &revs, flags, 1)) | |
1983 | die("bad revision '%s'", line); | |
1984 | } | |
1985 | ||
3d51e1b5 MK |
1986 | if (prepare_revision_walk(&revs)) |
1987 | die("revision walk setup failed"); | |
8d1d8f83 | 1988 | mark_edges_uninteresting(revs.commits, &revs, show_edge); |
b5d97e6b | 1989 | traverse_commit_list(&revs, show_commit, show_object); |
08cdfb13 JH |
1990 | |
1991 | if (keep_unreachable) | |
1992 | add_objects_in_unpacked_packs(&revs); | |
ca11b212 NP |
1993 | if (unpack_unreachable) |
1994 | loosen_unused_packed_objects(&revs); | |
b5d97e6b JH |
1995 | } |
1996 | ||
b6b32ccb JH |
1997 | static int adjust_perm(const char *path, mode_t mode) |
1998 | { | |
1999 | if (chmod(path, mode)) | |
2000 | return -1; | |
2001 | return adjust_shared_perm(path); | |
2002 | } | |
2003 | ||
b5d97e6b JH |
2004 | int cmd_pack_objects(int argc, const char **argv, const char *prefix) |
2005 | { | |
b5d97e6b | 2006 | int use_internal_rev_list = 0; |
8d1d8f83 | 2007 | int thin = 0; |
7cadf491 | 2008 | uint32_t i; |
ffa84ffb RD |
2009 | const char **rp_av; |
2010 | int rp_ac_alloc = 64; | |
8d1d8f83 JH |
2011 | int rp_ac; |
2012 | ||
ffa84ffb RD |
2013 | rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av)); |
2014 | ||
8d1d8f83 JH |
2015 | rp_av[0] = "pack-objects"; |
2016 | rp_av[1] = "--objects"; /* --thin will make it --objects-edge */ | |
2017 | rp_ac = 2; | |
b5d97e6b | 2018 | |
ef90d6d4 | 2019 | git_config(git_pack_config, NULL); |
960ccca6 DH |
2020 | if (!pack_compression_seen && core_compression_seen) |
2021 | pack_compression_level = core_compression_level; | |
b5d97e6b JH |
2022 | |
2023 | progress = isatty(2); | |
2024 | for (i = 1; i < argc; i++) { | |
2025 | const char *arg = argv[i]; | |
2026 | ||
2027 | if (*arg != '-') | |
2028 | break; | |
2029 | ||
2030 | if (!strcmp("--non-empty", arg)) { | |
2031 | non_empty = 1; | |
2032 | continue; | |
2033 | } | |
2034 | if (!strcmp("--local", arg)) { | |
2035 | local = 1; | |
2036 | continue; | |
2037 | } | |
b5d97e6b JH |
2038 | if (!strcmp("--incremental", arg)) { |
2039 | incremental = 1; | |
2040 | continue; | |
2041 | } | |
960ccca6 DH |
2042 | if (!prefixcmp(arg, "--compression=")) { |
2043 | char *end; | |
2044 | int level = strtoul(arg+14, &end, 0); | |
2045 | if (!arg[14] || *end) | |
2046 | usage(pack_usage); | |
2047 | if (level == -1) | |
2048 | level = Z_DEFAULT_COMPRESSION; | |
2049 | else if (level < 0 || level > Z_BEST_COMPRESSION) | |
2050 | die("bad pack compression level %d", level); | |
2051 | pack_compression_level = level; | |
2052 | continue; | |
2053 | } | |
6b94b1a0 DH |
2054 | if (!prefixcmp(arg, "--max-pack-size=")) { |
2055 | char *end; | |
2b84b5a8 | 2056 | pack_size_limit_cfg = 0; |
6b94b1a0 DH |
2057 | pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024; |
2058 | if (!arg[16] || *end) | |
2059 | usage(pack_usage); | |
2060 | continue; | |
2061 | } | |
599065a3 | 2062 | if (!prefixcmp(arg, "--window=")) { |
b5d97e6b JH |
2063 | char *end; |
2064 | window = strtoul(arg+9, &end, 0); | |
2065 | if (!arg[9] || *end) | |
2066 | usage(pack_usage); | |
2067 | continue; | |
2068 | } | |
a97773ce BD |
2069 | if (!prefixcmp(arg, "--window-memory=")) { |
2070 | if (!git_parse_ulong(arg+16, &window_memory_limit)) | |
2071 | usage(pack_usage); | |
2072 | continue; | |
2073 | } | |
367f4a43 NP |
2074 | if (!prefixcmp(arg, "--threads=")) { |
2075 | char *end; | |
2076 | delta_search_threads = strtoul(arg+10, &end, 0); | |
833e3df1 | 2077 | if (!arg[10] || *end || delta_search_threads < 0) |
367f4a43 NP |
2078 | usage(pack_usage); |
2079 | #ifndef THREADED_DELTA_SEARCH | |
833e3df1 | 2080 | if (delta_search_threads != 1) |
367f4a43 NP |
2081 | warning("no threads support, " |
2082 | "ignoring %s", arg); | |
2083 | #endif | |
2084 | continue; | |
2085 | } | |
599065a3 | 2086 | if (!prefixcmp(arg, "--depth=")) { |
b5d97e6b JH |
2087 | char *end; |
2088 | depth = strtoul(arg+8, &end, 0); | |
2089 | if (!arg[8] || *end) | |
2090 | usage(pack_usage); | |
2091 | continue; | |
2092 | } | |
2093 | if (!strcmp("--progress", arg)) { | |
2094 | progress = 1; | |
2095 | continue; | |
2096 | } | |
231f240b NP |
2097 | if (!strcmp("--all-progress", arg)) { |
2098 | progress = 2; | |
2099 | continue; | |
2100 | } | |
b5d97e6b JH |
2101 | if (!strcmp("-q", arg)) { |
2102 | progress = 0; | |
2103 | continue; | |
2104 | } | |
2105 | if (!strcmp("--no-reuse-delta", arg)) { | |
a7de7130 | 2106 | reuse_delta = 0; |
b5d97e6b JH |
2107 | continue; |
2108 | } | |
fa736f72 | 2109 | if (!strcmp("--no-reuse-object", arg)) { |
a7de7130 | 2110 | reuse_object = reuse_delta = 0; |
fa736f72 NP |
2111 | continue; |
2112 | } | |
be6b1914 | 2113 | if (!strcmp("--delta-base-offset", arg)) { |
780e6e73 | 2114 | allow_ofs_delta = 1; |
be6b1914 NP |
2115 | continue; |
2116 | } | |
b5d97e6b JH |
2117 | if (!strcmp("--stdout", arg)) { |
2118 | pack_to_stdout = 1; | |
2119 | continue; | |
2120 | } | |
2121 | if (!strcmp("--revs", arg)) { | |
2122 | use_internal_rev_list = 1; | |
2123 | continue; | |
2124 | } | |
08cdfb13 JH |
2125 | if (!strcmp("--keep-unreachable", arg)) { |
2126 | keep_unreachable = 1; | |
2127 | continue; | |
2128 | } | |
ca11b212 NP |
2129 | if (!strcmp("--unpack-unreachable", arg)) { |
2130 | unpack_unreachable = 1; | |
2131 | continue; | |
2132 | } | |
f0a24aa5 SP |
2133 | if (!strcmp("--include-tag", arg)) { |
2134 | include_tag = 1; | |
2135 | continue; | |
2136 | } | |
8d1d8f83 | 2137 | if (!strcmp("--unpacked", arg) || |
599065a3 | 2138 | !prefixcmp(arg, "--unpacked=") || |
63049292 | 2139 | !strcmp("--reflog", arg) || |
8d1d8f83 JH |
2140 | !strcmp("--all", arg)) { |
2141 | use_internal_rev_list = 1; | |
ffa84ffb RD |
2142 | if (rp_ac >= rp_ac_alloc - 1) { |
2143 | rp_ac_alloc = alloc_nr(rp_ac_alloc); | |
2144 | rp_av = xrealloc(rp_av, | |
2145 | rp_ac_alloc * sizeof(*rp_av)); | |
2146 | } | |
8d1d8f83 | 2147 | rp_av[rp_ac++] = arg; |
b5d97e6b JH |
2148 | continue; |
2149 | } | |
8d1d8f83 JH |
2150 | if (!strcmp("--thin", arg)) { |
2151 | use_internal_rev_list = 1; | |
2152 | thin = 1; | |
2153 | rp_av[1] = "--objects-edge"; | |
b5d97e6b | 2154 | continue; |
4ba7d711 NP |
2155 | } |
2156 | if (!prefixcmp(arg, "--index-version=")) { | |
2157 | char *c; | |
aa7e44bf GB |
2158 | pack_idx_default_version = strtoul(arg + 16, &c, 10); |
2159 | if (pack_idx_default_version > 2) | |
4ba7d711 NP |
2160 | die("bad %s", arg); |
2161 | if (*c == ',') | |
aa7e44bf GB |
2162 | pack_idx_off32_limit = strtoul(c+1, &c, 0); |
2163 | if (*c || pack_idx_off32_limit & 0x80000000) | |
4ba7d711 NP |
2164 | die("bad %s", arg); |
2165 | continue; | |
b5d97e6b JH |
2166 | } |
2167 | usage(pack_usage); | |
2168 | } | |
2169 | ||
2170 | /* Traditionally "pack-objects [options] base extra" failed; | |
2171 | * we would however want to take refs parameter that would | |
2172 | * have been given to upstream rev-list ourselves, which means | |
2173 | * we somehow want to say what the base name is. So the | |
2174 | * syntax would be: | |
2175 | * | |
2176 | * pack-objects [options] base <refs...> | |
2177 | * | |
2178 | * in other words, we would treat the first non-option as the | |
2179 | * base_name and send everything else to the internal revision | |
2180 | * walker. | |
2181 | */ | |
2182 | ||
2183 | if (!pack_to_stdout) | |
2184 | base_name = argv[i++]; | |
2185 | ||
2186 | if (pack_to_stdout != !base_name) | |
2187 | usage(pack_usage); | |
2188 | ||
2b84b5a8 JS |
2189 | if (!pack_to_stdout && !pack_size_limit) |
2190 | pack_size_limit = pack_size_limit_cfg; | |
2191 | ||
01c12a23 DH |
2192 | if (pack_to_stdout && pack_size_limit) |
2193 | die("--max-pack-size cannot be used to build a pack for transfer."); | |
2194 | ||
8d1d8f83 JH |
2195 | if (!pack_to_stdout && thin) |
2196 | die("--thin cannot be used to build an indexable pack."); | |
b5d97e6b | 2197 | |
ca11b212 NP |
2198 | if (keep_unreachable && unpack_unreachable) |
2199 | die("--keep-unreachable and --unpack-unreachable are incompatible."); | |
2200 | ||
833e3df1 AE |
2201 | #ifdef THREADED_DELTA_SEARCH |
2202 | if (!delta_search_threads) /* --threads=0 means autodetect */ | |
2203 | delta_search_threads = online_cpus(); | |
2204 | #endif | |
2205 | ||
b5d97e6b JH |
2206 | prepare_packed_git(); |
2207 | ||
13aaf148 | 2208 | if (progress) |
dc6a0757 | 2209 | progress_state = start_progress("Counting objects", 0); |
b5d97e6b JH |
2210 | if (!use_internal_rev_list) |
2211 | read_object_list_from_stdin(); | |
8d1d8f83 JH |
2212 | else { |
2213 | rp_av[rp_ac] = NULL; | |
2214 | get_object_list(rp_ac, rp_av); | |
2215 | } | |
f0a24aa5 SP |
2216 | if (include_tag && nr_result) |
2217 | for_each_ref(add_ref_tag, NULL); | |
4d4fcc54 | 2218 | stop_progress(&progress_state); |
96a02f8f | 2219 | |
f0b0af1b | 2220 | if (non_empty && !nr_result) |
1c4a2912 | 2221 | return 0; |
f7ae6a93 NP |
2222 | if (nr_result) |
2223 | prepare_pack(window, depth); | |
d01fb92f | 2224 | write_pack_file(); |
ab7cd7bb | 2225 | if (progress) |
6e1c2344 RJ |
2226 | fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32")," |
2227 | " reused %"PRIu32" (delta %"PRIu32")\n", | |
67c08ce1 | 2228 | written, written_delta, reused, reused_delta); |
c323ac7d LT |
2229 | return 0; |
2230 | } |