]>
Commit | Line | Data |
---|---|---|
5d4a6003 | 1 | #include "builtin.h" |
c323ac7d | 2 | #include "cache.h" |
a80d72db | 3 | #include "repository.h" |
b2141fc1 | 4 | #include "config.h" |
a74db82e | 5 | #include "attr.h" |
c323ac7d | 6 | #include "object.h" |
8e440259 PE |
7 | #include "blob.h" |
8 | #include "commit.h" | |
9 | #include "tag.h" | |
10 | #include "tree.h" | |
c323ac7d | 11 | #include "delta.h" |
a733cb60 | 12 | #include "pack.h" |
3449f8c4 | 13 | #include "pack-revindex.h" |
c38138cd | 14 | #include "csum-file.h" |
1b0c7174 | 15 | #include "tree-walk.h" |
b5d97e6b JH |
16 | #include "diff.h" |
17 | #include "revision.h" | |
18 | #include "list-objects.h" | |
9535ce73 JH |
19 | #include "list-objects-filter.h" |
20 | #include "list-objects-filter-options.h" | |
2834bc27 | 21 | #include "pack-objects.h" |
96a02f8f | 22 | #include "progress.h" |
f0a24aa5 | 23 | #include "refs.h" |
cf2ba13a | 24 | #include "streaming.h" |
93749194 | 25 | #include "thread-utils.h" |
6b8fda2d | 26 | #include "pack-bitmap.h" |
28b8a730 | 27 | #include "delta-islands.h" |
abcb8655 | 28 | #include "reachable.h" |
fe299ec5 | 29 | #include "oid-array.h" |
dbbcd44f | 30 | #include "strvec.h" |
ec2dd32c | 31 | #include "list.h" |
0317f455 | 32 | #include "packfile.h" |
a80d72db | 33 | #include "object-store.h" |
ed7e5fc3 | 34 | #include "dir.h" |
6a22d521 | 35 | #include "midx.h" |
ae417807 | 36 | #include "trace2.h" |
120ad2b0 | 37 | #include "shallow.h" |
e00549aa | 38 | #include "promisor-remote.h" |
8ecce684 | 39 | |
43fa44fa | 40 | #define IN_PACK(obj) oe_in_pack(&to_pack, obj) |
ac77d0c3 NTND |
41 | #define SIZE(obj) oe_size(&to_pack, obj) |
42 | #define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) | |
0aca34e8 | 43 | #define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) |
898eba5e NTND |
44 | #define DELTA(obj) oe_delta(&to_pack, obj) |
45 | #define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) | |
46 | #define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) | |
47 | #define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) | |
6a1e32d5 | 48 | #define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) |
0aca34e8 | 49 | #define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) |
898eba5e NTND |
50 | #define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) |
51 | #define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) | |
43fa44fa | 52 | |
99fb6e04 | 53 | static const char *pack_usage[] = { |
b8c1d275 AH |
54 | N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), |
55 | N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), | |
99fb6e04 NTND |
56 | NULL |
57 | }; | |
c323ac7d | 58 | |
3f9ac8d2 | 59 | /* |
2834bc27 VM |
60 | * Objects we are going to pack are collected in the `to_pack` structure. |
61 | * It contains an array (dynamically expanded) of the object data, and a map | |
62 | * that can resolve SHA1s to their position in the array. | |
3f9ac8d2 | 63 | */ |
2834bc27 VM |
64 | static struct packing_data to_pack; |
65 | ||
79814f42 | 66 | static struct pack_idx_entry **written_list; |
5af05043 | 67 | static uint32_t nr_result, nr_written, nr_seen; |
6a1e32d5 | 68 | static struct bitmap_index *bitmap_git; |
28b8a730 | 69 | static uint32_t write_layer; |
3f9ac8d2 | 70 | |
96f1e58f | 71 | static int non_empty; |
a7de7130 | 72 | static int reuse_delta = 1, reuse_object = 1; |
e5e9714a | 73 | static int keep_unreachable, unpack_unreachable, include_tag; |
dddbad72 | 74 | static timestamp_t unpack_unreachable_expiration; |
e26a8c47 | 75 | static int pack_loose_unreachable; |
96f1e58f | 76 | static int local; |
56dfeb62 | 77 | static int have_non_local_packs; |
96f1e58f | 78 | static int incremental; |
ed7e5fc3 NTND |
79 | static int ignore_packed_keep_on_disk; |
80 | static int ignore_packed_keep_in_core; | |
be6b1914 | 81 | static int allow_ofs_delta; |
ebcfb379 | 82 | static struct pack_idx_option pack_idx_opts; |
d01fb92f | 83 | static const char *base_name; |
024701f1 | 84 | static int progress = 1; |
4812a93a | 85 | static int window = 10; |
568508e7 | 86 | static unsigned long pack_size_limit; |
618e613a | 87 | static int depth = 50; |
43cc2b42 | 88 | static int delta_search_threads; |
df6d6101 | 89 | static int pack_to_stdout; |
4f6d26b1 | 90 | static int sparse; |
6a1e32d5 | 91 | static int thin; |
8d1d8f83 | 92 | static int num_preferred_base; |
dc6a0757 | 93 | static struct progress *progress_state; |
c323ac7d | 94 | |
6b8fda2d VM |
95 | static struct packed_git *reuse_packfile; |
96 | static uint32_t reuse_packfile_objects; | |
bb514de3 | 97 | static struct bitmap *reuse_packfile_bitmap; |
6b8fda2d | 98 | |
645c432d KS |
99 | static int use_bitmap_index_default = 1; |
100 | static int use_bitmap_index = -1; | |
e704fc79 | 101 | static int allow_pack_reuse = 1; |
25575015 JK |
102 | static enum { |
103 | WRITE_BITMAP_FALSE = 0, | |
104 | WRITE_BITMAP_QUIET, | |
105 | WRITE_BITMAP_TRUE, | |
106 | } write_bitmap_index; | |
d4316604 | 107 | static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE; |
6b8fda2d | 108 | |
0c16cd49 JT |
109 | static int exclude_promisor_objects; |
110 | ||
28b8a730 JK |
111 | static int use_delta_islands; |
112 | ||
074b2eea | 113 | static unsigned long delta_cache_size = 0; |
9806f5a7 | 114 | static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; |
e3dfddb3 | 115 | static unsigned long cache_max_small_delta_size = 1000; |
074b2eea | 116 | |
a97773ce BD |
117 | static unsigned long window_memory_limit = 0; |
118 | ||
9535ce73 JH |
119 | static struct list_objects_filter_options filter_options; |
120 | ||
dd4b732d JT |
121 | static struct string_list uri_protocols = STRING_LIST_INIT_NODUP; |
122 | ||
9535ce73 | 123 | enum missing_action { |
0c16cd49 JT |
124 | MA_ERROR = 0, /* fail if any missing objects are encountered */ |
125 | MA_ALLOW_ANY, /* silently allow ALL missing objects */ | |
126 | MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */ | |
9535ce73 JH |
127 | }; |
128 | static enum missing_action arg_missing_action; | |
129 | static show_object_fn fn_show_object; | |
130 | ||
dd4b732d JT |
131 | struct configured_exclusion { |
132 | struct oidmap_entry e; | |
133 | char *pack_hash_hex; | |
134 | char *uri; | |
135 | }; | |
136 | static struct oidmap configured_exclusions; | |
137 | ||
138 | static struct oidset excluded_by_config; | |
139 | ||
3f9ac8d2 JH |
140 | /* |
141 | * stats | |
142 | */ | |
7cadf491 SP |
143 | static uint32_t written, written_delta; |
144 | static uint32_t reused, reused_delta; | |
3f9ac8d2 | 145 | |
7cc8f971 VM |
146 | /* |
147 | * Indexed commits | |
148 | */ | |
149 | static struct commit **indexed_commits; | |
150 | static unsigned int indexed_commits_nr; | |
151 | static unsigned int indexed_commits_alloc; | |
152 | ||
153 | static void index_commit_for_bitmap(struct commit *commit) | |
154 | { | |
155 | if (indexed_commits_nr >= indexed_commits_alloc) { | |
156 | indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; | |
2756ca43 | 157 | REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); |
7cc8f971 VM |
158 | } |
159 | ||
160 | indexed_commits[indexed_commits_nr++] = commit; | |
161 | } | |
780e6e73 | 162 | |
3613f9b4 | 163 | static void *get_delta(struct object_entry *entry) |
c323ac7d | 164 | { |
3613f9b4 NP |
165 | unsigned long size, base_size, delta_size; |
166 | void *buf, *base_buf, *delta_buf; | |
21666f1a | 167 | enum object_type type; |
c323ac7d | 168 | |
b4f5aca4 | 169 | buf = read_object_file(&entry->idx.oid, &type, &size); |
3613f9b4 | 170 | if (!buf) |
f616db6a | 171 | die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); |
898eba5e NTND |
172 | base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, |
173 | &base_size); | |
3613f9b4 | 174 | if (!base_buf) |
e6a492b7 | 175 | die("unable to read %s", |
898eba5e | 176 | oid_to_hex(&DELTA(entry)->idx.oid)); |
3613f9b4 | 177 | delta_buf = diff_delta(base_buf, base_size, |
dcde55bc | 178 | buf, size, &delta_size, 0); |
1a07e59c | 179 | /* |
15beaaa3 | 180 | * We successfully computed this delta once but dropped it for |
1a07e59c NTND |
181 | * memory reasons. Something is very wrong if this time we |
182 | * recompute and create a different delta. | |
183 | */ | |
0aca34e8 | 184 | if (!delta_buf || delta_size != DELTA_SIZE(entry)) |
1a07e59c | 185 | BUG("delta size changed"); |
3613f9b4 NP |
186 | free(buf); |
187 | free(base_buf); | |
c323ac7d LT |
188 | return delta_buf; |
189 | } | |
190 | ||
30ebb40a NP |
191 | static unsigned long do_compress(void **pptr, unsigned long size) |
192 | { | |
ef49a7a0 | 193 | git_zstream stream; |
30ebb40a NP |
194 | void *in, *out; |
195 | unsigned long maxsize; | |
196 | ||
55bb5c91 | 197 | git_deflate_init(&stream, pack_compression_level); |
225a6f10 | 198 | maxsize = git_deflate_bound(&stream, size); |
30ebb40a NP |
199 | |
200 | in = *pptr; | |
201 | out = xmalloc(maxsize); | |
202 | *pptr = out; | |
203 | ||
204 | stream.next_in = in; | |
205 | stream.avail_in = size; | |
206 | stream.next_out = out; | |
207 | stream.avail_out = maxsize; | |
55bb5c91 | 208 | while (git_deflate(&stream, Z_FINISH) == Z_OK) |
30ebb40a | 209 | ; /* nothing */ |
55bb5c91 | 210 | git_deflate_end(&stream); |
30ebb40a NP |
211 | |
212 | free(in); | |
213 | return stream.total_out; | |
214 | } | |
215 | ||
98a3beab | 216 | static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, |
188960b4 | 217 | const struct object_id *oid) |
cf2ba13a NTND |
218 | { |
219 | git_zstream stream; | |
220 | unsigned char ibuf[1024 * 16]; | |
221 | unsigned char obuf[1024 * 16]; | |
222 | unsigned long olen = 0; | |
223 | ||
cf2ba13a NTND |
224 | git_deflate_init(&stream, pack_compression_level); |
225 | ||
226 | for (;;) { | |
227 | ssize_t readlen; | |
228 | int zret = Z_OK; | |
229 | readlen = read_istream(st, ibuf, sizeof(ibuf)); | |
230 | if (readlen == -1) | |
188960b4 | 231 | die(_("unable to read %s"), oid_to_hex(oid)); |
cf2ba13a NTND |
232 | |
233 | stream.next_in = ibuf; | |
234 | stream.avail_in = readlen; | |
235 | while ((stream.avail_in || readlen == 0) && | |
236 | (zret == Z_OK || zret == Z_BUF_ERROR)) { | |
237 | stream.next_out = obuf; | |
238 | stream.avail_out = sizeof(obuf); | |
239 | zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); | |
98a3beab | 240 | hashwrite(f, obuf, stream.next_out - obuf); |
cf2ba13a NTND |
241 | olen += stream.next_out - obuf; |
242 | } | |
243 | if (stream.avail_in) | |
244 | die(_("deflate error (%d)"), zret); | |
245 | if (readlen == 0) { | |
246 | if (zret != Z_STREAM_END) | |
247 | die(_("deflate error (%d)"), zret); | |
248 | break; | |
249 | } | |
250 | } | |
251 | git_deflate_end(&stream); | |
252 | return olen; | |
253 | } | |
254 | ||
780e6e73 NP |
255 | /* |
256 | * we are going to reuse the existing object data as is. make | |
257 | * sure it is not corrupt. | |
258 | */ | |
079afb18 SP |
259 | static int check_pack_inflate(struct packed_git *p, |
260 | struct pack_window **w_curs, | |
6777a59f SP |
261 | off_t offset, |
262 | off_t len, | |
079afb18 SP |
263 | unsigned long expect) |
264 | { | |
ef49a7a0 | 265 | git_zstream stream; |
079afb18 SP |
266 | unsigned char fakebuf[4096], *in; |
267 | int st; | |
268 | ||
269 | memset(&stream, 0, sizeof(stream)); | |
39c68542 | 270 | git_inflate_init(&stream); |
079afb18 SP |
271 | do { |
272 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
273 | stream.next_in = in; | |
274 | stream.next_out = fakebuf; | |
275 | stream.avail_out = sizeof(fakebuf); | |
39c68542 | 276 | st = git_inflate(&stream, Z_FINISH); |
079afb18 SP |
277 | offset += stream.next_in - in; |
278 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
39c68542 | 279 | git_inflate_end(&stream); |
079afb18 SP |
280 | return (st == Z_STREAM_END && |
281 | stream.total_out == expect && | |
282 | stream.total_in == len) ? 0 : -1; | |
283 | } | |
284 | ||
98a3beab | 285 | static void copy_pack_data(struct hashfile *f, |
079afb18 SP |
286 | struct packed_git *p, |
287 | struct pack_window **w_curs, | |
6777a59f SP |
288 | off_t offset, |
289 | off_t len) | |
079afb18 SP |
290 | { |
291 | unsigned char *in; | |
ef49a7a0 | 292 | unsigned long avail; |
079afb18 SP |
293 | |
294 | while (len) { | |
295 | in = use_pack(p, w_curs, offset, &avail); | |
296 | if (avail > len) | |
ef49a7a0 | 297 | avail = (unsigned long)len; |
98a3beab | 298 | hashwrite(f, in, avail); |
079afb18 SP |
299 | offset += avail; |
300 | len -= avail; | |
301 | } | |
302 | } | |
303 | ||
1b4bb16b | 304 | /* Return 0 if we will bust the pack-size limit */ |
98a3beab | 305 | static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, |
c9018b03 | 306 | unsigned long limit, int usable_delta) |
c323ac7d | 307 | { |
c9018b03 | 308 | unsigned long size, datalen; |
2c5e2865 JK |
309 | unsigned char header[MAX_PACK_OBJECT_HEADER], |
310 | dheader[MAX_PACK_OBJECT_HEADER]; | |
6777a59f | 311 | unsigned hdrlen; |
2c5ef824 | 312 | enum object_type type; |
c9018b03 | 313 | void *buf; |
cf2ba13a | 314 | struct git_istream *st = NULL; |
41179100 | 315 | const unsigned hashsz = the_hash_algo->rawsz; |
c9018b03 NTND |
316 | |
317 | if (!usable_delta) { | |
fd9b1bae | 318 | if (oe_type(entry) == OBJ_BLOB && |
ac77d0c3 | 319 | oe_size_greater_than(&to_pack, entry, big_file_threshold) && |
c8123e72 MT |
320 | (st = open_istream(the_repository, &entry->idx.oid, &type, |
321 | &size, NULL)) != NULL) | |
cf2ba13a NTND |
322 | buf = NULL; |
323 | else { | |
b4f5aca4 | 324 | buf = read_object_file(&entry->idx.oid, &type, &size); |
cf2ba13a | 325 | if (!buf) |
e6a492b7 | 326 | die(_("unable to read %s"), |
327 | oid_to_hex(&entry->idx.oid)); | |
cf2ba13a | 328 | } |
c9018b03 NTND |
329 | /* |
330 | * make sure no cached delta data remains from a | |
331 | * previous attempt before a pack split occurred. | |
332 | */ | |
6a83d902 | 333 | FREE_AND_NULL(entry->delta_data); |
c9018b03 NTND |
334 | entry->z_delta_size = 0; |
335 | } else if (entry->delta_data) { | |
0aca34e8 | 336 | size = DELTA_SIZE(entry); |
c9018b03 NTND |
337 | buf = entry->delta_data; |
338 | entry->delta_data = NULL; | |
898eba5e | 339 | type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? |
c9018b03 NTND |
340 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
341 | } else { | |
342 | buf = get_delta(entry); | |
0aca34e8 | 343 | size = DELTA_SIZE(entry); |
898eba5e | 344 | type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? |
c9018b03 NTND |
345 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
346 | } | |
347 | ||
cf2ba13a NTND |
348 | if (st) /* large blob case, just assume we don't compress well */ |
349 | datalen = size; | |
350 | else if (entry->z_delta_size) | |
c9018b03 NTND |
351 | datalen = entry->z_delta_size; |
352 | else | |
353 | datalen = do_compress(&buf, size); | |
354 | ||
355 | /* | |
356 | * The object header is a byte of 'type' followed by zero or | |
357 | * more bytes of length. | |
358 | */ | |
7202a6fa JK |
359 | hdrlen = encode_in_pack_object_header(header, sizeof(header), |
360 | type, size); | |
c9018b03 NTND |
361 | |
362 | if (type == OBJ_OFS_DELTA) { | |
363 | /* | |
364 | * Deltas with relative base contain an additional | |
365 | * encoding of the relative offset for the delta | |
366 | * base from this object's position in the pack. | |
367 | */ | |
898eba5e | 368 | off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; |
c9018b03 NTND |
369 | unsigned pos = sizeof(dheader) - 1; |
370 | dheader[pos] = ofs & 127; | |
371 | while (ofs >>= 7) | |
372 | dheader[--pos] = 128 | (--ofs & 127); | |
41179100 | 373 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { |
cf2ba13a NTND |
374 | if (st) |
375 | close_istream(st); | |
c9018b03 NTND |
376 | free(buf); |
377 | return 0; | |
378 | } | |
98a3beab | 379 | hashwrite(f, header, hdrlen); |
380 | hashwrite(f, dheader + pos, sizeof(dheader) - pos); | |
c9018b03 NTND |
381 | hdrlen += sizeof(dheader) - pos; |
382 | } else if (type == OBJ_REF_DELTA) { | |
383 | /* | |
384 | * Deltas with a base reference contain | |
41179100 | 385 | * additional bytes for the base object ID. |
c9018b03 | 386 | */ |
41179100 | 387 | if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { |
cf2ba13a NTND |
388 | if (st) |
389 | close_istream(st); | |
c9018b03 NTND |
390 | free(buf); |
391 | return 0; | |
392 | } | |
98a3beab | 393 | hashwrite(f, header, hdrlen); |
42c8ce1c | 394 | hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); |
41179100 | 395 | hdrlen += hashsz; |
c9018b03 | 396 | } else { |
41179100 | 397 | if (limit && hdrlen + datalen + hashsz >= limit) { |
cf2ba13a NTND |
398 | if (st) |
399 | close_istream(st); | |
c9018b03 NTND |
400 | free(buf); |
401 | return 0; | |
402 | } | |
98a3beab | 403 | hashwrite(f, header, hdrlen); |
c9018b03 | 404 | } |
cf2ba13a | 405 | if (st) { |
188960b4 | 406 | datalen = write_large_blob_data(st, f, &entry->idx.oid); |
cf2ba13a NTND |
407 | close_istream(st); |
408 | } else { | |
98a3beab | 409 | hashwrite(f, buf, datalen); |
cf2ba13a NTND |
410 | free(buf); |
411 | } | |
c9018b03 NTND |
412 | |
413 | return hdrlen + datalen; | |
414 | } | |
415 | ||
416 | /* Return 0 if we will bust the pack-size limit */ | |
98a3beab | 417 | static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry, |
af92a645 | 418 | unsigned long limit, int usable_delta) |
c9018b03 | 419 | { |
43fa44fa | 420 | struct packed_git *p = IN_PACK(entry); |
c9018b03 NTND |
421 | struct pack_window *w_curs = NULL; |
422 | struct revindex_entry *revidx; | |
423 | off_t offset; | |
fd9b1bae | 424 | enum object_type type = oe_type(entry); |
211c61c6 | 425 | off_t datalen; |
2c5e2865 JK |
426 | unsigned char header[MAX_PACK_OBJECT_HEADER], |
427 | dheader[MAX_PACK_OBJECT_HEADER]; | |
c9018b03 | 428 | unsigned hdrlen; |
41179100 | 429 | const unsigned hashsz = the_hash_algo->rawsz; |
ac77d0c3 | 430 | unsigned long entry_size = SIZE(entry); |
c9018b03 | 431 | |
898eba5e NTND |
432 | if (DELTA(entry)) |
433 | type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? | |
c9018b03 | 434 | OBJ_OFS_DELTA : OBJ_REF_DELTA; |
7202a6fa | 435 | hdrlen = encode_in_pack_object_header(header, sizeof(header), |
ac77d0c3 | 436 | type, entry_size); |
c9018b03 NTND |
437 | |
438 | offset = entry->in_pack_offset; | |
439 | revidx = find_pack_revindex(p, offset); | |
440 | datalen = revidx[1].offset - offset; | |
441 | if (!pack_to_stdout && p->index_version > 1 && | |
442 | check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { | |
f616db6a | 443 | error(_("bad packed object CRC for %s"), |
e6a492b7 | 444 | oid_to_hex(&entry->idx.oid)); |
c9018b03 NTND |
445 | unuse_pack(&w_curs); |
446 | return write_no_reuse_object(f, entry, limit, usable_delta); | |
447 | } | |
448 | ||
449 | offset += entry->in_pack_header_size; | |
450 | datalen -= entry->in_pack_header_size; | |
451 | ||
452 | if (!pack_to_stdout && p->index_version == 1 && | |
ac77d0c3 | 453 | check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { |
f616db6a | 454 | error(_("corrupt packed object for %s"), |
e6a492b7 | 455 | oid_to_hex(&entry->idx.oid)); |
c9018b03 NTND |
456 | unuse_pack(&w_curs); |
457 | return write_no_reuse_object(f, entry, limit, usable_delta); | |
458 | } | |
459 | ||
460 | if (type == OBJ_OFS_DELTA) { | |
898eba5e | 461 | off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; |
c9018b03 NTND |
462 | unsigned pos = sizeof(dheader) - 1; |
463 | dheader[pos] = ofs & 127; | |
464 | while (ofs >>= 7) | |
465 | dheader[--pos] = 128 | (--ofs & 127); | |
41179100 | 466 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { |
c9018b03 NTND |
467 | unuse_pack(&w_curs); |
468 | return 0; | |
469 | } | |
98a3beab | 470 | hashwrite(f, header, hdrlen); |
471 | hashwrite(f, dheader + pos, sizeof(dheader) - pos); | |
c9018b03 NTND |
472 | hdrlen += sizeof(dheader) - pos; |
473 | reused_delta++; | |
474 | } else if (type == OBJ_REF_DELTA) { | |
41179100 | 475 | if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { |
c9018b03 NTND |
476 | unuse_pack(&w_curs); |
477 | return 0; | |
478 | } | |
98a3beab | 479 | hashwrite(f, header, hdrlen); |
42c8ce1c | 480 | hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); |
41179100 | 481 | hdrlen += hashsz; |
c9018b03 NTND |
482 | reused_delta++; |
483 | } else { | |
41179100 | 484 | if (limit && hdrlen + datalen + hashsz >= limit) { |
c9018b03 NTND |
485 | unuse_pack(&w_curs); |
486 | return 0; | |
487 | } | |
98a3beab | 488 | hashwrite(f, header, hdrlen); |
c9018b03 NTND |
489 | } |
490 | copy_pack_data(f, p, &w_curs, offset, datalen); | |
491 | unuse_pack(&w_curs); | |
492 | reused++; | |
493 | return hdrlen + datalen; | |
494 | } | |
495 | ||
496 | /* Return 0 if we will bust the pack-size limit */ | |
98a3beab | 497 | static off_t write_object(struct hashfile *f, |
af92a645 NTND |
498 | struct object_entry *entry, |
499 | off_t write_offset) | |
c9018b03 | 500 | { |
af92a645 NTND |
501 | unsigned long limit; |
502 | off_t len; | |
2c5ef824 | 503 | int usable_delta, to_reuse; |
c323ac7d | 504 | |
78d1e84f NP |
505 | if (!pack_to_stdout) |
506 | crc32_begin(f); | |
507 | ||
a2430dde | 508 | /* apply size limit if limited packsize and not first object */ |
a1e4760f NP |
509 | if (!pack_size_limit || !nr_written) |
510 | limit = 0; | |
511 | else if (pack_size_limit <= write_offset) | |
512 | /* | |
513 | * the earlier object did not fit the limit; avoid | |
514 | * mistaking this with unlimited (i.e. limit = 0). | |
515 | */ | |
516 | limit = 1; | |
517 | else | |
518 | limit = pack_size_limit - write_offset; | |
2c5ef824 | 519 | |
898eba5e | 520 | if (!DELTA(entry)) |
2c5ef824 NP |
521 | usable_delta = 0; /* no delta */ |
522 | else if (!pack_size_limit) | |
523 | usable_delta = 1; /* unlimited packfile */ | |
898eba5e | 524 | else if (DELTA(entry)->idx.offset == (off_t)-1) |
2c5ef824 | 525 | usable_delta = 0; /* base was written to another pack */ |
898eba5e | 526 | else if (DELTA(entry)->idx.offset) |
2c5ef824 NP |
527 | usable_delta = 1; /* base already exists in this pack */ |
528 | else | |
529 | usable_delta = 0; /* base could end up in another pack */ | |
530 | ||
a7de7130 | 531 | if (!reuse_object) |
fa736f72 | 532 | to_reuse = 0; /* explicit */ |
43fa44fa | 533 | else if (!IN_PACK(entry)) |
ab7cd7bb | 534 | to_reuse = 0; /* can't reuse what we don't have */ |
fd9b1bae NTND |
535 | else if (oe_type(entry) == OBJ_REF_DELTA || |
536 | oe_type(entry) == OBJ_OFS_DELTA) | |
17b08f2c DH |
537 | /* check_object() decided it for us ... */ |
538 | to_reuse = usable_delta; | |
539 | /* ... but pack split may override that */ | |
fd9b1bae | 540 | else if (oe_type(entry) != entry->in_pack_type) |
ab7cd7bb | 541 | to_reuse = 0; /* pack has delta which is unusable */ |
898eba5e | 542 | else if (DELTA(entry)) |
ab7cd7bb JH |
543 | to_reuse = 0; /* we want to pack afresh */ |
544 | else | |
545 | to_reuse = 1; /* we have it in-pack undeltified, | |
546 | * and we do not need to deltify it. | |
547 | */ | |
548 | ||
c9018b03 NTND |
549 | if (!to_reuse) |
550 | len = write_no_reuse_object(f, entry, limit, usable_delta); | |
551 | else | |
552 | len = write_reuse_object(f, entry, limit, usable_delta); | |
553 | if (!len) | |
554 | return 0; | |
64bd76b1 | 555 | |
17b08f2c | 556 | if (usable_delta) |
ab7cd7bb | 557 | written_delta++; |
3f9ac8d2 | 558 | written++; |
78d1e84f | 559 | if (!pack_to_stdout) |
aa7e44bf | 560 | entry->idx.crc32 = crc32_end(f); |
c9018b03 | 561 | return len; |
c323ac7d LT |
562 | } |
563 | ||
f63c79db JH |
564 | enum write_one_status { |
565 | WRITE_ONE_SKIP = -1, /* already written */ | |
566 | WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ | |
567 | WRITE_ONE_WRITTEN = 1, /* normal */ | |
568 | WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ | |
569 | }; | |
570 | ||
98a3beab | 571 | static enum write_one_status write_one(struct hashfile *f, |
f63c79db JH |
572 | struct object_entry *e, |
573 | off_t *offset) | |
9d5ab962 | 574 | { |
af92a645 | 575 | off_t size; |
f63c79db | 576 | int recursing; |
d7dd0223 | 577 | |
f63c79db JH |
578 | /* |
579 | * we set offset to 1 (which is an impossible value) to mark | |
580 | * the fact that this object is involved in "write its base | |
581 | * first before writing a deltified object" recursion. | |
582 | */ | |
583 | recursing = (e->idx.offset == 1); | |
584 | if (recursing) { | |
f616db6a | 585 | warning(_("recursive delta detected for object %s"), |
e6a492b7 | 586 | oid_to_hex(&e->idx.oid)); |
f63c79db JH |
587 | return WRITE_ONE_RECURSIVE; |
588 | } else if (e->idx.offset || e->preferred_base) { | |
589 | /* offset is non zero if object is written already. */ | |
590 | return WRITE_ONE_SKIP; | |
591 | } | |
d7dd0223 | 592 | |
720c9f7b | 593 | /* if we are deltified, write out base object first. */ |
898eba5e | 594 | if (DELTA(e)) { |
f63c79db | 595 | e->idx.offset = 1; /* now recurse */ |
898eba5e | 596 | switch (write_one(f, DELTA(e), offset)) { |
f63c79db JH |
597 | case WRITE_ONE_RECURSIVE: |
598 | /* we cannot depend on this one */ | |
898eba5e | 599 | SET_DELTA(e, NULL); |
f63c79db JH |
600 | break; |
601 | default: | |
602 | break; | |
603 | case WRITE_ONE_BREAK: | |
604 | e->idx.offset = recursing; | |
605 | return WRITE_ONE_BREAK; | |
606 | } | |
607 | } | |
d7dd0223 | 608 | |
6ed7f25e NP |
609 | e->idx.offset = *offset; |
610 | size = write_object(f, e, *offset); | |
17b08f2c | 611 | if (!size) { |
f63c79db JH |
612 | e->idx.offset = recursing; |
613 | return WRITE_ONE_BREAK; | |
17b08f2c | 614 | } |
79814f42 | 615 | written_list[nr_written++] = &e->idx; |
d7dd0223 NP |
616 | |
617 | /* make sure off_t is sufficiently large not to wrap */ | |
c03c8315 | 618 | if (signed_add_overflows(*offset, size)) |
f616db6a | 619 | die(_("pack too large for current definition of off_t")); |
6ed7f25e | 620 | *offset += size; |
f63c79db | 621 | return WRITE_ONE_WRITTEN; |
9d5ab962 JH |
622 | } |
623 | ||
d155254c | 624 | static int mark_tagged(const char *path, const struct object_id *oid, int flag, |
1b4bb16b JH |
625 | void *cb_data) |
626 | { | |
188960b4 | 627 | struct object_id peeled; |
3a37876b | 628 | struct object_entry *entry = packlist_find(&to_pack, oid); |
1b4bb16b JH |
629 | |
630 | if (entry) | |
631 | entry->tagged = 1; | |
b420d909 | 632 | if (!peel_ref(path, &peeled)) { |
3a37876b | 633 | entry = packlist_find(&to_pack, &peeled); |
1b4bb16b JH |
634 | if (entry) |
635 | entry->tagged = 1; | |
636 | } | |
637 | return 0; | |
638 | } | |
639 | ||
be126818 | 640 | static inline void add_to_write_order(struct object_entry **wo, |
92bef1a1 | 641 | unsigned int *endp, |
1b4bb16b JH |
642 | struct object_entry *e) |
643 | { | |
fe0ac2fb | 644 | if (e->filled || oe_layer(&to_pack, e) != write_layer) |
1b4bb16b JH |
645 | return; |
646 | wo[(*endp)++] = e; | |
647 | e->filled = 1; | |
648 | } | |
649 | ||
650 | static void add_descendants_to_write_order(struct object_entry **wo, | |
92bef1a1 | 651 | unsigned int *endp, |
1b4bb16b JH |
652 | struct object_entry *e) |
653 | { | |
f380872f DM |
654 | int add_to_order = 1; |
655 | while (e) { | |
656 | if (add_to_order) { | |
657 | struct object_entry *s; | |
658 | /* add this node... */ | |
659 | add_to_write_order(wo, endp, e); | |
660 | /* all its siblings... */ | |
898eba5e | 661 | for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { |
f380872f DM |
662 | add_to_write_order(wo, endp, s); |
663 | } | |
664 | } | |
665 | /* drop down a level to add left subtree nodes if possible */ | |
898eba5e | 666 | if (DELTA_CHILD(e)) { |
f380872f | 667 | add_to_order = 1; |
898eba5e | 668 | e = DELTA_CHILD(e); |
f380872f DM |
669 | } else { |
670 | add_to_order = 0; | |
671 | /* our sibling might have some children, it is next */ | |
898eba5e NTND |
672 | if (DELTA_SIBLING(e)) { |
673 | e = DELTA_SIBLING(e); | |
f380872f DM |
674 | continue; |
675 | } | |
676 | /* go back to our parent node */ | |
898eba5e NTND |
677 | e = DELTA(e); |
678 | while (e && !DELTA_SIBLING(e)) { | |
f380872f DM |
679 | /* we're on the right side of a subtree, keep |
680 | * going up until we can go right again */ | |
898eba5e | 681 | e = DELTA(e); |
f380872f DM |
682 | } |
683 | if (!e) { | |
684 | /* done- we hit our original root node */ | |
685 | return; | |
686 | } | |
687 | /* pass it off to sibling at this level */ | |
898eba5e | 688 | e = DELTA_SIBLING(e); |
f380872f DM |
689 | } |
690 | }; | |
1b4bb16b JH |
691 | } |
692 | ||
693 | static void add_family_to_write_order(struct object_entry **wo, | |
92bef1a1 | 694 | unsigned int *endp, |
1b4bb16b JH |
695 | struct object_entry *e) |
696 | { | |
697 | struct object_entry *root; | |
698 | ||
898eba5e | 699 | for (root = e; DELTA(root); root = DELTA(root)) |
1b4bb16b | 700 | ; /* nothing */ |
1b4bb16b JH |
701 | add_descendants_to_write_order(wo, endp, root); |
702 | } | |
703 | ||
f64ba53e | 704 | static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) |
1b4bb16b | 705 | { |
f64ba53e | 706 | unsigned int i, last_untagged; |
2834bc27 | 707 | struct object_entry *objects = to_pack.objects; |
1b4bb16b | 708 | |
2834bc27 | 709 | for (i = 0; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
710 | if (objects[i].tagged) |
711 | break; | |
f64ba53e | 712 | add_to_write_order(wo, wo_end, &objects[i]); |
1b4bb16b | 713 | } |
38d4debb | 714 | last_untagged = i; |
1b4bb16b JH |
715 | |
716 | /* | |
717 | * Then fill all the tagged tips. | |
718 | */ | |
2834bc27 | 719 | for (; i < to_pack.nr_objects; i++) { |
1b4bb16b | 720 | if (objects[i].tagged) |
f64ba53e | 721 | add_to_write_order(wo, wo_end, &objects[i]); |
1b4bb16b JH |
722 | } |
723 | ||
724 | /* | |
725 | * And then all remaining commits and tags. | |
726 | */ | |
2834bc27 | 727 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
fd9b1bae NTND |
728 | if (oe_type(&objects[i]) != OBJ_COMMIT && |
729 | oe_type(&objects[i]) != OBJ_TAG) | |
1b4bb16b | 730 | continue; |
f64ba53e | 731 | add_to_write_order(wo, wo_end, &objects[i]); |
1b4bb16b JH |
732 | } |
733 | ||
734 | /* | |
735 | * And then all the trees. | |
736 | */ | |
2834bc27 | 737 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
fd9b1bae | 738 | if (oe_type(&objects[i]) != OBJ_TREE) |
1b4bb16b | 739 | continue; |
f64ba53e | 740 | add_to_write_order(wo, wo_end, &objects[i]); |
1b4bb16b JH |
741 | } |
742 | ||
743 | /* | |
744 | * Finally all the rest in really tight order | |
745 | */ | |
2834bc27 | 746 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
fe0ac2fb | 747 | if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) |
f64ba53e | 748 | add_family_to_write_order(wo, wo_end, &objects[i]); |
38d4debb | 749 | } |
f64ba53e CC |
750 | } |
751 | ||
752 | static struct object_entry **compute_write_order(void) | |
753 | { | |
28b8a730 | 754 | uint32_t max_layers = 1; |
f64ba53e CC |
755 | unsigned int i, wo_end; |
756 | ||
757 | struct object_entry **wo; | |
758 | struct object_entry *objects = to_pack.objects; | |
759 | ||
760 | for (i = 0; i < to_pack.nr_objects; i++) { | |
761 | objects[i].tagged = 0; | |
762 | objects[i].filled = 0; | |
763 | SET_DELTA_CHILD(&objects[i], NULL); | |
764 | SET_DELTA_SIBLING(&objects[i], NULL); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Fully connect delta_child/delta_sibling network. | |
769 | * Make sure delta_sibling is sorted in the original | |
770 | * recency order. | |
771 | */ | |
772 | for (i = to_pack.nr_objects; i > 0;) { | |
773 | struct object_entry *e = &objects[--i]; | |
774 | if (!DELTA(e)) | |
775 | continue; | |
776 | /* Mark me as the first child */ | |
777 | e->delta_sibling_idx = DELTA(e)->delta_child_idx; | |
778 | SET_DELTA_CHILD(DELTA(e), e); | |
38d4debb DM |
779 | } |
780 | ||
f64ba53e CC |
781 | /* |
782 | * Mark objects that are at the tip of tags. | |
783 | */ | |
784 | for_each_tag_ref(mark_tagged, NULL); | |
785 | ||
28b8a730 JK |
786 | if (use_delta_islands) |
787 | max_layers = compute_pack_layers(&to_pack); | |
788 | ||
f64ba53e CC |
789 | ALLOC_ARRAY(wo, to_pack.nr_objects); |
790 | wo_end = 0; | |
791 | ||
28b8a730 JK |
792 | for (; write_layer < max_layers; ++write_layer) |
793 | compute_layer_order(wo, &wo_end); | |
38d4debb | 794 | |
2834bc27 | 795 | if (wo_end != to_pack.nr_objects) |
f616db6a NTND |
796 | die(_("ordered %u objects, expected %"PRIu32), |
797 | wo_end, to_pack.nr_objects); | |
1b4bb16b JH |
798 | |
799 | return wo; | |
800 | } | |
801 | ||
bb514de3 JK |
802 | |
803 | /* | |
804 | * A reused set of objects. All objects in a chunk have the same | |
805 | * relative position in the original packfile and the generated | |
806 | * packfile. | |
807 | */ | |
808 | ||
809 | static struct reused_chunk { | |
810 | /* The offset of the first object of this chunk in the original | |
811 | * packfile. */ | |
812 | off_t original; | |
813 | /* The offset of the first object of this chunk in the generated | |
814 | * packfile minus "original". */ | |
815 | off_t difference; | |
816 | } *reused_chunks; | |
817 | static int reused_chunks_nr; | |
818 | static int reused_chunks_alloc; | |
819 | ||
820 | static void record_reused_object(off_t where, off_t offset) | |
821 | { | |
822 | if (reused_chunks_nr && reused_chunks[reused_chunks_nr-1].difference == offset) | |
823 | return; | |
824 | ||
825 | ALLOC_GROW(reused_chunks, reused_chunks_nr + 1, | |
826 | reused_chunks_alloc); | |
827 | reused_chunks[reused_chunks_nr].original = where; | |
828 | reused_chunks[reused_chunks_nr].difference = offset; | |
829 | reused_chunks_nr++; | |
830 | } | |
831 | ||
832 | /* | |
833 | * Binary search to find the chunk that "where" is in. Note | |
834 | * that we're not looking for an exact match, just the first | |
835 | * chunk that contains it (which implicitly ends at the start | |
836 | * of the next chunk. | |
837 | */ | |
838 | static off_t find_reused_offset(off_t where) | |
6b8fda2d | 839 | { |
bb514de3 JK |
840 | int lo = 0, hi = reused_chunks_nr; |
841 | while (lo < hi) { | |
842 | int mi = lo + ((hi - lo) / 2); | |
843 | if (where == reused_chunks[mi].original) | |
844 | return reused_chunks[mi].difference; | |
845 | if (where < reused_chunks[mi].original) | |
846 | hi = mi; | |
847 | else | |
848 | lo = mi + 1; | |
849 | } | |
6b8fda2d | 850 | |
bb514de3 JK |
851 | /* |
852 | * The first chunk starts at zero, so we can't have gone below | |
853 | * there. | |
854 | */ | |
855 | assert(lo); | |
856 | return reused_chunks[lo-1].difference; | |
857 | } | |
6b8fda2d | 858 | |
bb514de3 JK |
859 | static void write_reused_pack_one(size_t pos, struct hashfile *out, |
860 | struct pack_window **w_curs) | |
6b8fda2d | 861 | { |
bb514de3 JK |
862 | off_t offset, next, cur; |
863 | enum object_type type; | |
864 | unsigned long size; | |
6b8fda2d | 865 | |
bb514de3 JK |
866 | offset = reuse_packfile->revindex[pos].offset; |
867 | next = reuse_packfile->revindex[pos + 1].offset; | |
6b8fda2d | 868 | |
bb514de3 | 869 | record_reused_object(offset, offset - hashfile_total(out)); |
6b8fda2d | 870 | |
bb514de3 JK |
871 | cur = offset; |
872 | type = unpack_object_header(reuse_packfile, w_curs, &cur, &size); | |
873 | assert(type >= 0); | |
6b8fda2d | 874 | |
bb514de3 JK |
875 | if (type == OBJ_OFS_DELTA) { |
876 | off_t base_offset; | |
877 | off_t fixup; | |
878 | ||
879 | unsigned char header[MAX_PACK_OBJECT_HEADER]; | |
880 | unsigned len; | |
881 | ||
882 | base_offset = get_delta_base(reuse_packfile, w_curs, &cur, type, offset); | |
883 | assert(base_offset != 0); | |
884 | ||
885 | /* Convert to REF_DELTA if we must... */ | |
886 | if (!allow_ofs_delta) { | |
887 | int base_pos = find_revindex_position(reuse_packfile, base_offset); | |
f66d4e02 JK |
888 | struct object_id base_oid; |
889 | ||
890 | nth_packed_object_id(&base_oid, reuse_packfile, | |
891 | reuse_packfile->revindex[base_pos].nr); | |
bb514de3 JK |
892 | |
893 | len = encode_in_pack_object_header(header, sizeof(header), | |
894 | OBJ_REF_DELTA, size); | |
895 | hashwrite(out, header, len); | |
f8cb64e3 | 896 | hashwrite(out, base_oid.hash, the_hash_algo->rawsz); |
bb514de3 JK |
897 | copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur); |
898 | return; | |
899 | } | |
6b8fda2d | 900 | |
bb514de3 JK |
901 | /* Otherwise see if we need to rewrite the offset... */ |
902 | fixup = find_reused_offset(offset) - | |
903 | find_reused_offset(base_offset); | |
904 | if (fixup) { | |
905 | unsigned char ofs_header[10]; | |
906 | unsigned i, ofs_len; | |
907 | off_t ofs = offset - base_offset - fixup; | |
6b8fda2d | 908 | |
bb514de3 JK |
909 | len = encode_in_pack_object_header(header, sizeof(header), |
910 | OBJ_OFS_DELTA, size); | |
6b8fda2d | 911 | |
bb514de3 JK |
912 | i = sizeof(ofs_header) - 1; |
913 | ofs_header[i] = ofs & 127; | |
914 | while (ofs >>= 7) | |
915 | ofs_header[--i] = 128 | (--ofs & 127); | |
6b8fda2d | 916 | |
bb514de3 | 917 | ofs_len = sizeof(ofs_header) - i; |
6b8fda2d | 918 | |
bb514de3 JK |
919 | hashwrite(out, header, len); |
920 | hashwrite(out, ofs_header + sizeof(ofs_header) - ofs_len, ofs_len); | |
921 | copy_pack_data(out, reuse_packfile, w_curs, cur, next - cur); | |
922 | return; | |
923 | } | |
924 | ||
925 | /* ...otherwise we have no fixup, and can write it verbatim */ | |
926 | } | |
927 | ||
928 | copy_pack_data(out, reuse_packfile, w_curs, offset, next - offset); | |
929 | } | |
930 | ||
931 | static size_t write_reused_pack_verbatim(struct hashfile *out, | |
932 | struct pack_window **w_curs) | |
933 | { | |
934 | size_t pos = 0; | |
935 | ||
936 | while (pos < reuse_packfile_bitmap->word_alloc && | |
937 | reuse_packfile_bitmap->words[pos] == (eword_t)~0) | |
938 | pos++; | |
939 | ||
940 | if (pos) { | |
941 | off_t to_write; | |
942 | ||
943 | written = (pos * BITS_IN_EWORD); | |
944 | to_write = reuse_packfile->revindex[written].offset | |
945 | - sizeof(struct pack_header); | |
946 | ||
947 | /* We're recording one chunk, not one object. */ | |
948 | record_reused_object(sizeof(struct pack_header), 0); | |
949 | hashflush(out); | |
950 | copy_pack_data(out, reuse_packfile, w_curs, | |
951 | sizeof(struct pack_header), to_write); | |
657673f1 | 952 | |
657673f1 | 953 | display_progress(progress_state, written); |
6b8fda2d | 954 | } |
bb514de3 JK |
955 | return pos; |
956 | } | |
957 | ||
958 | static void write_reused_pack(struct hashfile *f) | |
959 | { | |
960 | size_t i = 0; | |
961 | uint32_t offset; | |
962 | struct pack_window *w_curs = NULL; | |
963 | ||
964 | if (allow_ofs_delta) | |
965 | i = write_reused_pack_verbatim(f, &w_curs); | |
966 | ||
967 | for (; i < reuse_packfile_bitmap->word_alloc; ++i) { | |
968 | eword_t word = reuse_packfile_bitmap->words[i]; | |
969 | size_t pos = (i * BITS_IN_EWORD); | |
6b8fda2d | 970 | |
bb514de3 JK |
971 | for (offset = 0; offset < BITS_IN_EWORD; ++offset) { |
972 | if ((word >> offset) == 0) | |
973 | break; | |
974 | ||
975 | offset += ewah_bit_ctz64(word >> offset); | |
976 | write_reused_pack_one(pos + offset, f, &w_curs); | |
977 | display_progress(progress_state, ++written); | |
978 | } | |
979 | } | |
6b8fda2d | 980 | |
bb514de3 | 981 | unuse_pack(&w_curs); |
6b8fda2d VM |
982 | } |
983 | ||
dd4b732d JT |
984 | static void write_excluded_by_configs(void) |
985 | { | |
986 | struct oidset_iter iter; | |
987 | const struct object_id *oid; | |
988 | ||
989 | oidset_iter_init(&excluded_by_config, &iter); | |
990 | while ((oid = oidset_iter_next(&iter))) { | |
991 | struct configured_exclusion *ex = | |
992 | oidmap_get(&configured_exclusions, oid); | |
993 | ||
994 | if (!ex) | |
995 | BUG("configured exclusion wasn't configured"); | |
996 | write_in_full(1, ex->pack_hash_hex, strlen(ex->pack_hash_hex)); | |
997 | write_in_full(1, " ", 1); | |
998 | write_in_full(1, ex->uri, strlen(ex->uri)); | |
999 | write_in_full(1, "\n", 1); | |
1000 | } | |
1001 | } | |
1002 | ||
9cea46cd EW |
1003 | static const char no_split_warning[] = N_( |
1004 | "disabling bitmap writing, packs are split due to pack.packSizeLimit" | |
1005 | ); | |
1006 | ||
d01fb92f | 1007 | static void write_pack_file(void) |
c323ac7d | 1008 | { |
ebe27b13 | 1009 | uint32_t i = 0, j; |
98a3beab | 1010 | struct hashfile *f; |
6ed7f25e | 1011 | off_t offset; |
ebe27b13 | 1012 | uint32_t nr_remaining = nr_result; |
f746bae8 | 1013 | time_t last_mtime = 0; |
1b4bb16b | 1014 | struct object_entry **write_order; |
81a216a5 | 1015 | |
bcd7954e | 1016 | if (progress > pack_to_stdout) |
754dbc43 | 1017 | progress_state = start_progress(_("Writing objects"), nr_result); |
b32fa95f | 1018 | ALLOC_ARRAY(written_list, to_pack.nr_objects); |
1b4bb16b | 1019 | write_order = compute_write_order(); |
183bdb2c | 1020 | |
ebe27b13 | 1021 | do { |
188960b4 | 1022 | struct object_id oid; |
7ba502c4 | 1023 | char *pack_tmp_name = NULL; |
aa7e44bf | 1024 | |
cdf9db3c | 1025 | if (pack_to_stdout) |
98a3beab | 1026 | f = hashfd_throughput(1, "<stdout>", progress_state); |
cdf9db3c JH |
1027 | else |
1028 | f = create_tmp_packfile(&pack_tmp_name); | |
ebe27b13 | 1029 | |
c0ad4657 | 1030 | offset = write_pack_header(f, nr_remaining); |
6b8fda2d VM |
1031 | |
1032 | if (reuse_packfile) { | |
6b8fda2d | 1033 | assert(pack_to_stdout); |
bb514de3 JK |
1034 | write_reused_pack(f); |
1035 | offset = hashfile_total(f); | |
6b8fda2d VM |
1036 | } |
1037 | ||
ebe27b13 | 1038 | nr_written = 0; |
2834bc27 | 1039 | for (; i < to_pack.nr_objects; i++) { |
1b4bb16b | 1040 | struct object_entry *e = write_order[i]; |
cddec4f8 | 1041 | if (write_one(f, e, &offset) == WRITE_ONE_BREAK) |
720c9f7b NP |
1042 | break; |
1043 | display_progress(progress_state, written); | |
1044 | } | |
c553ca25 | 1045 | |
ebe27b13 DH |
1046 | /* |
1047 | * Did we write the wrong # entries in the header? | |
1048 | * If so, rewrite it like in fast-import | |
1049 | */ | |
54352bb2 | 1050 | if (pack_to_stdout) { |
cfe83216 | 1051 | finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); |
54352bb2 | 1052 | } else if (nr_written == nr_remaining) { |
cfe83216 | 1053 | finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); |
ebe27b13 | 1054 | } else { |
f2af9f5e | 1055 | int fd = finalize_hashfile(f, oid.hash, 0); |
188960b4 | 1056 | fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, |
1057 | nr_written, oid.hash, offset); | |
7ba502c4 | 1058 | close(fd); |
9cea46cd | 1059 | if (write_bitmap_index) { |
25575015 JK |
1060 | if (write_bitmap_index != WRITE_BITMAP_QUIET) |
1061 | warning(_(no_split_warning)); | |
9cea46cd EW |
1062 | write_bitmap_index = 0; |
1063 | } | |
ebe27b13 DH |
1064 | } |
1065 | ||
1066 | if (!pack_to_stdout) { | |
f746bae8 | 1067 | struct stat st; |
58892711 | 1068 | struct strbuf tmpname = STRBUF_INIT; |
d01fb92f | 1069 | |
f746bae8 NP |
1070 | /* |
1071 | * Packs are runtime accessed in their mtime | |
1072 | * order since newer packs are more likely to contain | |
1073 | * younger objects. So if we are creating multiple | |
1074 | * packs then we should modify the mtime of later ones | |
1075 | * to preserve this property. | |
1076 | */ | |
0e990530 | 1077 | if (stat(pack_tmp_name, &st) < 0) { |
f616db6a | 1078 | warning_errno(_("failed to stat %s"), pack_tmp_name); |
f746bae8 NP |
1079 | } else if (!last_mtime) { |
1080 | last_mtime = st.st_mtime; | |
1081 | } else { | |
1082 | struct utimbuf utb; | |
1083 | utb.actime = st.st_atime; | |
1084 | utb.modtime = --last_mtime; | |
0e990530 | 1085 | if (utime(pack_tmp_name, &utb) < 0) |
f616db6a | 1086 | warning_errno(_("failed utime() on %s"), pack_tmp_name); |
f746bae8 NP |
1087 | } |
1088 | ||
58892711 | 1089 | strbuf_addf(&tmpname, "%s-", base_name); |
7cc8f971 VM |
1090 | |
1091 | if (write_bitmap_index) { | |
188960b4 | 1092 | bitmap_writer_set_checksum(oid.hash); |
06af3bba NTND |
1093 | bitmap_writer_build_type_index( |
1094 | &to_pack, written_list, nr_written); | |
7cc8f971 VM |
1095 | } |
1096 | ||
58892711 | 1097 | finish_tmp_packfile(&tmpname, pack_tmp_name, |
0e990530 | 1098 | written_list, nr_written, |
188960b4 | 1099 | &pack_idx_opts, oid.hash); |
7cc8f971 VM |
1100 | |
1101 | if (write_bitmap_index) { | |
188960b4 | 1102 | strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); |
7cc8f971 VM |
1103 | |
1104 | stop_progress(&progress_state); | |
1105 | ||
1106 | bitmap_writer_show_progress(progress); | |
1107 | bitmap_writer_reuse_bitmaps(&to_pack); | |
1108 | bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); | |
1109 | bitmap_writer_build(&to_pack); | |
ae4f07fb | 1110 | bitmap_writer_finish(written_list, nr_written, |
58892711 | 1111 | tmpname.buf, write_bitmap_options); |
7cc8f971 VM |
1112 | write_bitmap_index = 0; |
1113 | } | |
1114 | ||
58892711 | 1115 | strbuf_release(&tmpname); |
7ba502c4 | 1116 | free(pack_tmp_name); |
188960b4 | 1117 | puts(oid_to_hex(&oid)); |
ebe27b13 DH |
1118 | } |
1119 | ||
1120 | /* mark written objects as written to previous pack */ | |
1121 | for (j = 0; j < nr_written; j++) { | |
79814f42 | 1122 | written_list[j]->offset = (off_t)-1; |
ebe27b13 DH |
1123 | } |
1124 | nr_remaining -= nr_written; | |
2834bc27 | 1125 | } while (nr_remaining && i < to_pack.nr_objects); |
ebe27b13 DH |
1126 | |
1127 | free(written_list); | |
1b4bb16b | 1128 | free(write_order); |
4d4fcc54 | 1129 | stop_progress(&progress_state); |
67c08ce1 | 1130 | if (written != nr_result) |
f616db6a NTND |
1131 | die(_("wrote %"PRIu32" objects while expecting %"PRIu32), |
1132 | written, nr_result); | |
9ed87902 JT |
1133 | trace2_data_intmax("pack-objects", the_repository, |
1134 | "write_pack_file/wrote", nr_result); | |
c323ac7d LT |
1135 | } |
1136 | ||
a74db82e JH |
1137 | static int no_try_delta(const char *path) |
1138 | { | |
2aef63d3 | 1139 | static struct attr_check *check; |
a74db82e | 1140 | |
2aef63d3 JH |
1141 | if (!check) |
1142 | check = attr_check_initl("delta", NULL); | |
f8adbec9 | 1143 | git_check_attr(the_repository->index, path, check); |
2aef63d3 | 1144 | if (ATTR_FALSE(check->items[0].value)) |
a74db82e JH |
1145 | return 1; |
1146 | return 0; | |
1147 | } | |
1148 | ||
ce2bc424 JK |
1149 | /* |
1150 | * When adding an object, check whether we have already added it | |
1151 | * to our packing list. If so, we can skip. However, if we are | |
1152 | * being asked to excludei t, but the previous mention was to include | |
1153 | * it, make sure to adjust its flags and tweak our numbers accordingly. | |
1154 | * | |
1155 | * As an optimization, we pass out the index position where we would have | |
1156 | * found the item, since that saves us from having to look it up again a | |
1157 | * few lines later when we want to add the new entry. | |
1158 | */ | |
188960b4 | 1159 | static int have_duplicate_entry(const struct object_id *oid, |
3a37876b | 1160 | int exclude) |
c323ac7d | 1161 | { |
c323ac7d | 1162 | struct object_entry *entry; |
29b734e4 | 1163 | |
92fb0db9 JK |
1164 | if (reuse_packfile_bitmap && |
1165 | bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid)) | |
1166 | return 1; | |
1167 | ||
3a37876b | 1168 | entry = packlist_find(&to_pack, oid); |
ce2bc424 | 1169 | if (!entry) |
29b734e4 | 1170 | return 0; |
ce2bc424 JK |
1171 | |
1172 | if (exclude) { | |
1173 | if (!entry->preferred_base) | |
1174 | nr_result--; | |
1175 | entry->preferred_base = 1; | |
29b734e4 | 1176 | } |
c323ac7d | 1177 | |
ce2bc424 JK |
1178 | return 1; |
1179 | } | |
1180 | ||
702d1b95 KS |
1181 | static int want_found_object(int exclude, struct packed_git *p) |
1182 | { | |
1183 | if (exclude) | |
1184 | return 1; | |
1185 | if (incremental) | |
1186 | return 0; | |
1187 | ||
1188 | /* | |
1189 | * When asked to do --local (do not include an object that appears in a | |
1190 | * pack we borrow from elsewhere) or --honor-pack-keep (do not include | |
1191 | * an object that appears in a pack marked with .keep), finding a pack | |
1192 | * that matches the criteria is sufficient for us to decide to omit it. | |
1193 | * However, even if this pack does not satisfy the criteria, we need to | |
1194 | * make sure no copy of this object appears in _any_ pack that makes us | |
1195 | * to omit the object, so we need to check all the packs. | |
1196 | * | |
1197 | * We can however first check whether these options can possible matter; | |
1198 | * if they do not matter we know we want the object in generated pack. | |
1199 | * Otherwise, we signal "-1" at the end to tell the caller that we do | |
1200 | * not know either way, and it needs to check more packs. | |
1201 | */ | |
ed7e5fc3 NTND |
1202 | if (!ignore_packed_keep_on_disk && |
1203 | !ignore_packed_keep_in_core && | |
702d1b95 KS |
1204 | (!local || !have_non_local_packs)) |
1205 | return 1; | |
1206 | ||
1207 | if (local && !p->pack_local) | |
1208 | return 0; | |
ed7e5fc3 NTND |
1209 | if (p->pack_local && |
1210 | ((ignore_packed_keep_on_disk && p->pack_keep) || | |
1211 | (ignore_packed_keep_in_core && p->pack_keep_in_core))) | |
702d1b95 KS |
1212 | return 0; |
1213 | ||
1214 | /* we don't know yet; keep looking for more packs */ | |
1215 | return -1; | |
1216 | } | |
1217 | ||
ce2bc424 JK |
1218 | /* |
1219 | * Check whether we want the object in the pack (e.g., we do not want | |
1220 | * objects found in non-local stores if the "--local" option was used). | |
1221 | * | |
702d1b95 KS |
1222 | * If the caller already knows an existing pack it wants to take the object |
1223 | * from, that is passed in *found_pack and *found_offset; otherwise this | |
1224 | * function finds if there is any pack that has the object and returns the pack | |
1225 | * and its offset in these variables. | |
ce2bc424 | 1226 | */ |
188960b4 | 1227 | static int want_object_in_pack(const struct object_id *oid, |
ce2bc424 JK |
1228 | int exclude, |
1229 | struct packed_git **found_pack, | |
1230 | off_t *found_offset) | |
1231 | { | |
702d1b95 | 1232 | int want; |
8865859d | 1233 | struct list_head *pos; |
6a22d521 | 1234 | struct multi_pack_index *m; |
ce2bc424 | 1235 | |
6862ebbf | 1236 | if (!exclude && local && has_loose_object_nonlocal(oid)) |
daae0625 BC |
1237 | return 0; |
1238 | ||
702d1b95 KS |
1239 | /* |
1240 | * If we already know the pack object lives in, start checks from that | |
1241 | * pack - in the usual case when neither --local was given nor .keep files | |
1242 | * are present we will determine the answer right now. | |
1243 | */ | |
1244 | if (*found_pack) { | |
1245 | want = want_found_object(exclude, *found_pack); | |
1246 | if (want != -1) | |
1247 | return want; | |
1248 | } | |
6a22d521 DS |
1249 | |
1250 | for (m = get_multi_pack_index(the_repository); m; m = m->next) { | |
1251 | struct pack_entry e; | |
64404a24 | 1252 | if (fill_midx_entry(the_repository, oid, &e, m)) { |
6a22d521 DS |
1253 | struct packed_git *p = e.p; |
1254 | off_t offset; | |
1255 | ||
1256 | if (p == *found_pack) | |
1257 | offset = *found_offset; | |
1258 | else | |
1259 | offset = find_pack_entry_one(oid->hash, p); | |
1260 | ||
1261 | if (offset) { | |
1262 | if (!*found_pack) { | |
1263 | if (!is_pack_valid(p)) | |
1264 | continue; | |
1265 | *found_offset = offset; | |
1266 | *found_pack = p; | |
1267 | } | |
1268 | want = want_found_object(exclude, p); | |
1269 | if (want != -1) | |
1270 | return want; | |
1271 | } | |
1272 | } | |
1273 | } | |
1274 | ||
a80d72db | 1275 | list_for_each(pos, get_packed_git_mru(the_repository)) { |
ec2dd32c | 1276 | struct packed_git *p = list_entry(pos, struct packed_git, mru); |
702d1b95 KS |
1277 | off_t offset; |
1278 | ||
1279 | if (p == *found_pack) | |
1280 | offset = *found_offset; | |
1281 | else | |
188960b4 | 1282 | offset = find_pack_entry_one(oid->hash, p); |
702d1b95 | 1283 | |
5c49c116 | 1284 | if (offset) { |
ce2bc424 | 1285 | if (!*found_pack) { |
319b678a | 1286 | if (!is_pack_valid(p)) |
4c080182 | 1287 | continue; |
ce2bc424 JK |
1288 | *found_offset = offset; |
1289 | *found_pack = p; | |
64560374 | 1290 | } |
702d1b95 | 1291 | want = want_found_object(exclude, p); |
e6e24c94 | 1292 | if (!exclude && want > 0) |
a80d72db SB |
1293 | list_move(&p->mru, |
1294 | get_packed_git_mru(the_repository)); | |
702d1b95 KS |
1295 | if (want != -1) |
1296 | return want; | |
64560374 LT |
1297 | } |
1298 | } | |
eb019375 | 1299 | |
dd4b732d JT |
1300 | if (uri_protocols.nr) { |
1301 | struct configured_exclusion *ex = | |
1302 | oidmap_get(&configured_exclusions, oid); | |
1303 | int i; | |
1304 | const char *p; | |
1305 | ||
1306 | if (ex) { | |
1307 | for (i = 0; i < uri_protocols.nr; i++) { | |
1308 | if (skip_prefix(ex->uri, | |
1309 | uri_protocols.items[i].string, | |
1310 | &p) && | |
1311 | *p == ':') { | |
1312 | oidset_insert(&excluded_by_config, oid); | |
1313 | return 0; | |
1314 | } | |
1315 | } | |
1316 | } | |
1317 | } | |
1318 | ||
ce2bc424 JK |
1319 | return 1; |
1320 | } | |
1321 | ||
188960b4 | 1322 | static void create_object_entry(const struct object_id *oid, |
ce2bc424 JK |
1323 | enum object_type type, |
1324 | uint32_t hash, | |
1325 | int exclude, | |
1326 | int no_try_delta, | |
ce2bc424 JK |
1327 | struct packed_git *found_pack, |
1328 | off_t found_offset) | |
1329 | { | |
1330 | struct object_entry *entry; | |
29b734e4 | 1331 | |
3a37876b | 1332 | entry = packlist_alloc(&to_pack, oid); |
27225f2e | 1333 | entry->hash = hash; |
fd9b1bae | 1334 | oe_set_type(entry, type); |
29b734e4 NP |
1335 | if (exclude) |
1336 | entry->preferred_base = 1; | |
81a216a5 NP |
1337 | else |
1338 | nr_result++; | |
29b734e4 | 1339 | if (found_pack) { |
43fa44fa | 1340 | oe_set_in_pack(&to_pack, entry, found_pack); |
29b734e4 NP |
1341 | entry->in_pack_offset = found_offset; |
1342 | } | |
7a979d99 | 1343 | |
ce2bc424 JK |
1344 | entry->no_try_delta = no_try_delta; |
1345 | } | |
29b734e4 | 1346 | |
373c67da JK |
1347 | static const char no_closure_warning[] = N_( |
1348 | "disabling bitmap writing, as some objects are not being packed" | |
1349 | ); | |
1350 | ||
188960b4 | 1351 | static int add_object_entry(const struct object_id *oid, enum object_type type, |
ce2bc424 JK |
1352 | const char *name, int exclude) |
1353 | { | |
702d1b95 KS |
1354 | struct packed_git *found_pack = NULL; |
1355 | off_t found_offset = 0; | |
7a979d99 | 1356 | |
5af05043 NTND |
1357 | display_progress(progress_state, ++nr_seen); |
1358 | ||
3a37876b | 1359 | if (have_duplicate_entry(oid, exclude)) |
ce2bc424 | 1360 | return 0; |
a74db82e | 1361 | |
188960b4 | 1362 | if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) { |
373c67da JK |
1363 | /* The pack is missing an object, so it will not have closure */ |
1364 | if (write_bitmap_index) { | |
25575015 JK |
1365 | if (write_bitmap_index != WRITE_BITMAP_QUIET) |
1366 | warning(_(no_closure_warning)); | |
373c67da JK |
1367 | write_bitmap_index = 0; |
1368 | } | |
ce2bc424 | 1369 | return 0; |
373c67da | 1370 | } |
29b734e4 | 1371 | |
188960b4 | 1372 | create_object_entry(oid, type, pack_name_hash(name), |
ce2bc424 | 1373 | exclude, name && no_try_delta(name), |
3a37876b | 1374 | found_pack, found_offset); |
29b734e4 | 1375 | return 1; |
c323ac7d LT |
1376 | } |
1377 | ||
20664967 | 1378 | static int add_object_entry_from_bitmap(const struct object_id *oid, |
6b8fda2d VM |
1379 | enum object_type type, |
1380 | int flags, uint32_t name_hash, | |
1381 | struct packed_git *pack, off_t offset) | |
1382 | { | |
5af05043 NTND |
1383 | display_progress(progress_state, ++nr_seen); |
1384 | ||
3a37876b | 1385 | if (have_duplicate_entry(oid, 0)) |
6b8fda2d VM |
1386 | return 0; |
1387 | ||
188960b4 | 1388 | if (!want_object_in_pack(oid, 0, &pack, &offset)) |
702d1b95 KS |
1389 | return 0; |
1390 | ||
3a37876b | 1391 | create_object_entry(oid, type, name_hash, 0, 0, pack, offset); |
29b734e4 | 1392 | return 1; |
c323ac7d LT |
1393 | } |
1394 | ||
5379a5c5 | 1395 | struct pbase_tree_cache { |
188960b4 | 1396 | struct object_id oid; |
5379a5c5 JH |
1397 | int ref; |
1398 | int temporary; | |
1399 | void *tree_data; | |
1400 | unsigned long tree_size; | |
1401 | }; | |
1402 | ||
1403 | static struct pbase_tree_cache *(pbase_tree_cache[256]); | |
188960b4 | 1404 | static int pbase_tree_cache_ix(const struct object_id *oid) |
5379a5c5 | 1405 | { |
188960b4 | 1406 | return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache); |
5379a5c5 JH |
1407 | } |
1408 | static int pbase_tree_cache_ix_incr(int ix) | |
1409 | { | |
1410 | return (ix+1) % ARRAY_SIZE(pbase_tree_cache); | |
1411 | } | |
1412 | ||
1413 | static struct pbase_tree { | |
1414 | struct pbase_tree *next; | |
1415 | /* This is a phony "cache" entry; we are not | |
01689909 | 1416 | * going to evict it or find it through _get() |
5379a5c5 JH |
1417 | * mechanism -- this is for the toplevel node that |
1418 | * would almost always change with any commit. | |
1419 | */ | |
1420 | struct pbase_tree_cache pcache; | |
1421 | } *pbase_tree; | |
1422 | ||
188960b4 | 1423 | static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid) |
5379a5c5 JH |
1424 | { |
1425 | struct pbase_tree_cache *ent, *nent; | |
1426 | void *data; | |
1427 | unsigned long size; | |
21666f1a | 1428 | enum object_type type; |
5379a5c5 | 1429 | int neigh; |
188960b4 | 1430 | int my_ix = pbase_tree_cache_ix(oid); |
5379a5c5 JH |
1431 | int available_ix = -1; |
1432 | ||
1433 | /* pbase-tree-cache acts as a limited hashtable. | |
1434 | * your object will be found at your index or within a few | |
1435 | * slots after that slot if it is cached. | |
1436 | */ | |
1437 | for (neigh = 0; neigh < 8; neigh++) { | |
1438 | ent = pbase_tree_cache[my_ix]; | |
4a7e27e9 | 1439 | if (ent && oideq(&ent->oid, oid)) { |
5379a5c5 JH |
1440 | ent->ref++; |
1441 | return ent; | |
1442 | } | |
1443 | else if (((available_ix < 0) && (!ent || !ent->ref)) || | |
1444 | ((0 <= available_ix) && | |
1445 | (!ent && pbase_tree_cache[available_ix]))) | |
1446 | available_ix = my_ix; | |
1447 | if (!ent) | |
1448 | break; | |
1449 | my_ix = pbase_tree_cache_ix_incr(my_ix); | |
1450 | } | |
1451 | ||
1452 | /* Did not find one. Either we got a bogus request or | |
1453 | * we need to read and perhaps cache. | |
1454 | */ | |
b4f5aca4 | 1455 | data = read_object_file(oid, &type, &size); |
5379a5c5 JH |
1456 | if (!data) |
1457 | return NULL; | |
21666f1a | 1458 | if (type != OBJ_TREE) { |
5379a5c5 JH |
1459 | free(data); |
1460 | return NULL; | |
1461 | } | |
1462 | ||
1463 | /* We need to either cache or return a throwaway copy */ | |
1464 | ||
1465 | if (available_ix < 0) | |
1466 | ent = NULL; | |
1467 | else { | |
1468 | ent = pbase_tree_cache[available_ix]; | |
1469 | my_ix = available_ix; | |
1470 | } | |
1471 | ||
1472 | if (!ent) { | |
1473 | nent = xmalloc(sizeof(*nent)); | |
1474 | nent->temporary = (available_ix < 0); | |
1475 | } | |
1476 | else { | |
1477 | /* evict and reuse */ | |
1478 | free(ent->tree_data); | |
1479 | nent = ent; | |
1480 | } | |
188960b4 | 1481 | oidcpy(&nent->oid, oid); |
5379a5c5 JH |
1482 | nent->tree_data = data; |
1483 | nent->tree_size = size; | |
1484 | nent->ref = 1; | |
1485 | if (!nent->temporary) | |
1486 | pbase_tree_cache[my_ix] = nent; | |
1487 | return nent; | |
1488 | } | |
1489 | ||
1490 | static void pbase_tree_put(struct pbase_tree_cache *cache) | |
1491 | { | |
1492 | if (!cache->temporary) { | |
1493 | cache->ref--; | |
1494 | return; | |
1495 | } | |
1496 | free(cache->tree_data); | |
1497 | free(cache); | |
1498 | } | |
1499 | ||
1500 | static int name_cmp_len(const char *name) | |
1501 | { | |
1502 | int i; | |
1503 | for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++) | |
1504 | ; | |
1505 | return i; | |
1506 | } | |
1507 | ||
1508 | static void add_pbase_object(struct tree_desc *tree, | |
5379a5c5 | 1509 | const char *name, |
ce0bd642 LT |
1510 | int cmplen, |
1511 | const char *fullname) | |
3f9ac8d2 | 1512 | { |
4c068a98 | 1513 | struct name_entry entry; |
8a5a8d6c | 1514 | int cmp; |
4c068a98 LT |
1515 | |
1516 | while (tree_entry(tree,&entry)) { | |
1211be6b LT |
1517 | if (S_ISGITLINK(entry.mode)) |
1518 | continue; | |
0de16337 | 1519 | cmp = tree_entry_len(&entry) != cmplen ? 1 : |
8a5a8d6c NP |
1520 | memcmp(name, entry.path, cmplen); |
1521 | if (cmp > 0) | |
7a979d99 | 1522 | continue; |
8a5a8d6c NP |
1523 | if (cmp < 0) |
1524 | return; | |
5379a5c5 | 1525 | if (name[cmplen] != '/') { |
ea82b2a0 | 1526 | add_object_entry(&entry.oid, |
4d1012c3 | 1527 | object_type(entry.mode), |
bc32fed5 | 1528 | fullname, 1); |
5379a5c5 JH |
1529 | return; |
1530 | } | |
8a5a8d6c | 1531 | if (S_ISDIR(entry.mode)) { |
7a979d99 | 1532 | struct tree_desc sub; |
5379a5c5 JH |
1533 | struct pbase_tree_cache *tree; |
1534 | const char *down = name+cmplen+1; | |
1535 | int downlen = name_cmp_len(down); | |
1536 | ||
ea82b2a0 | 1537 | tree = pbase_tree_get(&entry.oid); |
5379a5c5 JH |
1538 | if (!tree) |
1539 | return; | |
6fda5e51 | 1540 | init_tree_desc(&sub, tree->tree_data, tree->tree_size); |
5379a5c5 | 1541 | |
ce0bd642 | 1542 | add_pbase_object(&sub, down, downlen, fullname); |
5379a5c5 JH |
1543 | pbase_tree_put(tree); |
1544 | } | |
1545 | } | |
1546 | } | |
1d6b38cc | 1547 | |
5379a5c5 JH |
1548 | static unsigned *done_pbase_paths; |
1549 | static int done_pbase_paths_num; | |
1550 | static int done_pbase_paths_alloc; | |
1551 | static int done_pbase_path_pos(unsigned hash) | |
1552 | { | |
1553 | int lo = 0; | |
1554 | int hi = done_pbase_paths_num; | |
1555 | while (lo < hi) { | |
19716b21 | 1556 | int mi = lo + (hi - lo) / 2; |
5379a5c5 JH |
1557 | if (done_pbase_paths[mi] == hash) |
1558 | return mi; | |
1559 | if (done_pbase_paths[mi] < hash) | |
1560 | hi = mi; | |
1561 | else | |
1562 | lo = mi + 1; | |
1563 | } | |
1564 | return -lo-1; | |
1565 | } | |
1566 | ||
1567 | static int check_pbase_path(unsigned hash) | |
1568 | { | |
c7b07805 | 1569 | int pos = done_pbase_path_pos(hash); |
5379a5c5 JH |
1570 | if (0 <= pos) |
1571 | return 1; | |
1572 | pos = -pos - 1; | |
25e19407 DD |
1573 | ALLOC_GROW(done_pbase_paths, |
1574 | done_pbase_paths_num + 1, | |
1575 | done_pbase_paths_alloc); | |
5379a5c5 JH |
1576 | done_pbase_paths_num++; |
1577 | if (pos < done_pbase_paths_num) | |
f331ab9d RS |
1578 | MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos, |
1579 | done_pbase_paths_num - pos - 1); | |
5379a5c5 JH |
1580 | done_pbase_paths[pos] = hash; |
1581 | return 0; | |
1582 | } | |
1583 | ||
bc32fed5 | 1584 | static void add_preferred_base_object(const char *name) |
5379a5c5 JH |
1585 | { |
1586 | struct pbase_tree *it; | |
8a5a8d6c | 1587 | int cmplen; |
68fb36eb | 1588 | unsigned hash = pack_name_hash(name); |
5379a5c5 | 1589 | |
8a5a8d6c | 1590 | if (!num_preferred_base || check_pbase_path(hash)) |
5379a5c5 JH |
1591 | return; |
1592 | ||
8a5a8d6c | 1593 | cmplen = name_cmp_len(name); |
5379a5c5 JH |
1594 | for (it = pbase_tree; it; it = it->next) { |
1595 | if (cmplen == 0) { | |
188960b4 | 1596 | add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1); |
5379a5c5 JH |
1597 | } |
1598 | else { | |
1599 | struct tree_desc tree; | |
6fda5e51 | 1600 | init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size); |
ce0bd642 | 1601 | add_pbase_object(&tree, name, cmplen, name); |
7a979d99 | 1602 | } |
3f9ac8d2 | 1603 | } |
3f9ac8d2 JH |
1604 | } |
1605 | ||
188960b4 | 1606 | static void add_preferred_base(struct object_id *oid) |
3f9ac8d2 | 1607 | { |
5379a5c5 JH |
1608 | struct pbase_tree *it; |
1609 | void *data; | |
1610 | unsigned long size; | |
188960b4 | 1611 | struct object_id tree_oid; |
1d6b38cc | 1612 | |
8d1d8f83 JH |
1613 | if (window <= num_preferred_base++) |
1614 | return; | |
1615 | ||
d3b4705a NTND |
1616 | data = read_object_with_reference(the_repository, oid, |
1617 | tree_type, &size, &tree_oid); | |
5379a5c5 | 1618 | if (!data) |
7a979d99 | 1619 | return; |
5379a5c5 JH |
1620 | |
1621 | for (it = pbase_tree; it; it = it->next) { | |
4a7e27e9 | 1622 | if (oideq(&it->pcache.oid, &tree_oid)) { |
5379a5c5 JH |
1623 | free(data); |
1624 | return; | |
1625 | } | |
1626 | } | |
1627 | ||
1628 | it = xcalloc(1, sizeof(*it)); | |
1629 | it->next = pbase_tree; | |
1630 | pbase_tree = it; | |
1631 | ||
188960b4 | 1632 | oidcpy(&it->pcache.oid, &tree_oid); |
5379a5c5 JH |
1633 | it->pcache.tree_data = data; |
1634 | it->pcache.tree_size = size; | |
3f9ac8d2 JH |
1635 | } |
1636 | ||
0ef95f72 NP |
1637 | static void cleanup_preferred_base(void) |
1638 | { | |
1639 | struct pbase_tree *it; | |
1640 | unsigned i; | |
1641 | ||
1642 | it = pbase_tree; | |
1643 | pbase_tree = NULL; | |
1644 | while (it) { | |
095b3b2c BW |
1645 | struct pbase_tree *tmp = it; |
1646 | it = tmp->next; | |
1647 | free(tmp->pcache.tree_data); | |
1648 | free(tmp); | |
0ef95f72 NP |
1649 | } |
1650 | ||
1651 | for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) { | |
1652 | if (!pbase_tree_cache[i]) | |
1653 | continue; | |
1654 | free(pbase_tree_cache[i]->tree_data); | |
6a83d902 | 1655 | FREE_AND_NULL(pbase_tree_cache[i]); |
0ef95f72 NP |
1656 | } |
1657 | ||
6a83d902 | 1658 | FREE_AND_NULL(done_pbase_paths); |
0ef95f72 NP |
1659 | done_pbase_paths_num = done_pbase_paths_alloc = 0; |
1660 | } | |
1661 | ||
2fa233a5 JK |
1662 | /* |
1663 | * Return 1 iff the object specified by "delta" can be sent | |
1664 | * literally as a delta against the base in "base_sha1". If | |
1665 | * so, then *base_out will point to the entry in our packing | |
1666 | * list, or NULL if we must use the external-base list. | |
1667 | * | |
1668 | * Depth value does not matter - find_deltas() will | |
1669 | * never consider reused delta as the base object to | |
1670 | * deltify other objects against, in order to avoid | |
1671 | * circular deltas. | |
1672 | */ | |
3f83fd5e | 1673 | static int can_reuse_delta(const struct object_id *base_oid, |
2fa233a5 JK |
1674 | struct object_entry *delta, |
1675 | struct object_entry **base_out) | |
1676 | { | |
1677 | struct object_entry *base; | |
3df28cae | 1678 | |
2fa233a5 JK |
1679 | /* |
1680 | * First see if we're already sending the base (or it's explicitly in | |
1681 | * our "excluded" list). | |
1682 | */ | |
3f83fd5e | 1683 | base = packlist_find(&to_pack, base_oid); |
2fa233a5 JK |
1684 | if (base) { |
1685 | if (!in_same_island(&delta->idx.oid, &base->idx.oid)) | |
1686 | return 0; | |
1687 | *base_out = base; | |
1688 | return 1; | |
1689 | } | |
1690 | ||
1691 | /* | |
1692 | * Otherwise, reachability bitmaps may tell us if the receiver has it, | |
1693 | * even if it was buried too deep in history to make it into the | |
1694 | * packing list. | |
1695 | */ | |
3f83fd5e | 1696 | if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, base_oid)) { |
2fa233a5 | 1697 | if (use_delta_islands) { |
3f83fd5e | 1698 | if (!in_same_island(&delta->idx.oid, base_oid)) |
2fa233a5 JK |
1699 | return 0; |
1700 | } | |
1701 | *base_out = NULL; | |
1702 | return 1; | |
1703 | } | |
1704 | ||
1705 | return 0; | |
1706 | } | |
1707 | ||
e00549aa JT |
1708 | static void prefetch_to_pack(uint32_t object_index_start) { |
1709 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
1710 | uint32_t i; | |
1711 | ||
1712 | for (i = object_index_start; i < to_pack.nr_objects; i++) { | |
1713 | struct object_entry *entry = to_pack.objects + i; | |
1714 | ||
1715 | if (!oid_object_info_extended(the_repository, | |
1716 | &entry->idx.oid, | |
1717 | NULL, | |
1718 | OBJECT_INFO_FOR_PREFETCH)) | |
1719 | continue; | |
1720 | oid_array_append(&to_fetch, &entry->idx.oid); | |
1721 | } | |
1722 | promisor_remote_get_direct(the_repository, | |
1723 | to_fetch.oid, to_fetch.nr); | |
1724 | oid_array_clear(&to_fetch); | |
1725 | } | |
1726 | ||
1727 | static void check_object(struct object_entry *entry, uint32_t object_index) | |
c323ac7d | 1728 | { |
ac77d0c3 | 1729 | unsigned long canonical_size; |
8d5cf957 JT |
1730 | enum object_type type; |
1731 | struct object_info oi = {.typep = &type, .sizep = &canonical_size}; | |
ac77d0c3 | 1732 | |
43fa44fa NTND |
1733 | if (IN_PACK(entry)) { |
1734 | struct packed_git *p = IN_PACK(entry); | |
03e79c88 | 1735 | struct pack_window *w_curs = NULL; |
3f83fd5e JK |
1736 | int have_base = 0; |
1737 | struct object_id base_ref; | |
5c49c116 NP |
1738 | struct object_entry *base_entry; |
1739 | unsigned long used, used_0; | |
ef49a7a0 | 1740 | unsigned long avail; |
5c49c116 NP |
1741 | off_t ofs; |
1742 | unsigned char *buf, c; | |
fd9b1bae | 1743 | enum object_type type; |
27a7d067 | 1744 | unsigned long in_pack_size; |
780e6e73 | 1745 | |
6777a59f | 1746 | buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail); |
ab7cd7bb | 1747 | |
5c49c116 | 1748 | /* |
fa736f72 NP |
1749 | * We want in_pack_type even if we do not reuse delta |
1750 | * since non-delta representations could still be reused. | |
ab7cd7bb | 1751 | */ |
09ded04b | 1752 | used = unpack_object_header_buffer(buf, avail, |
fd9b1bae | 1753 | &type, |
27a7d067 | 1754 | &in_pack_size); |
03d66015 NP |
1755 | if (used == 0) |
1756 | goto give_up; | |
ab7cd7bb | 1757 | |
fd9b1bae NTND |
1758 | if (type < 0) |
1759 | BUG("invalid type %d", type); | |
1760 | entry->in_pack_type = type; | |
1761 | ||
5c49c116 NP |
1762 | /* |
1763 | * Determine if this is a delta and if so whether we can | |
1764 | * reuse it or not. Otherwise let's find out as cheaply as | |
1765 | * possible what the actual type and size for this object is. | |
3f9ac8d2 | 1766 | */ |
5c49c116 NP |
1767 | switch (entry->in_pack_type) { |
1768 | default: | |
1769 | /* Not a delta hence we've already got all we need. */ | |
fd9b1bae | 1770 | oe_set_type(entry, entry->in_pack_type); |
ac77d0c3 | 1771 | SET_SIZE(entry, in_pack_size); |
5c49c116 | 1772 | entry->in_pack_header_size = used; |
fd9b1bae | 1773 | if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB) |
03d66015 | 1774 | goto give_up; |
5c49c116 NP |
1775 | unuse_pack(&w_curs); |
1776 | return; | |
1777 | case OBJ_REF_DELTA: | |
3f83fd5e JK |
1778 | if (reuse_delta && !entry->preferred_base) { |
1779 | oidread(&base_ref, | |
1780 | use_pack(p, &w_curs, | |
1781 | entry->in_pack_offset + used, | |
1782 | NULL)); | |
1783 | have_base = 1; | |
1784 | } | |
41179100 | 1785 | entry->in_pack_header_size = used + the_hash_algo->rawsz; |
5c49c116 NP |
1786 | break; |
1787 | case OBJ_OFS_DELTA: | |
1788 | buf = use_pack(p, &w_curs, | |
1789 | entry->in_pack_offset + used, NULL); | |
1790 | used_0 = 0; | |
1791 | c = buf[used_0++]; | |
1792 | ofs = c & 127; | |
1793 | while (c & 128) { | |
1794 | ofs += 1; | |
03d66015 | 1795 | if (!ofs || MSB(ofs, 7)) { |
f616db6a | 1796 | error(_("delta base offset overflow in pack for %s"), |
e6a492b7 | 1797 | oid_to_hex(&entry->idx.oid)); |
03d66015 NP |
1798 | goto give_up; |
1799 | } | |
5c49c116 NP |
1800 | c = buf[used_0++]; |
1801 | ofs = (ofs << 7) + (c & 127); | |
780e6e73 | 1802 | } |
5c49c116 | 1803 | ofs = entry->in_pack_offset - ofs; |
03d66015 | 1804 | if (ofs <= 0 || ofs >= entry->in_pack_offset) { |
f616db6a | 1805 | error(_("delta base offset out of bound for %s"), |
e6a492b7 | 1806 | oid_to_hex(&entry->idx.oid)); |
03d66015 NP |
1807 | goto give_up; |
1808 | } | |
a7de7130 | 1809 | if (reuse_delta && !entry->preferred_base) { |
3449f8c4 NP |
1810 | struct revindex_entry *revidx; |
1811 | revidx = find_pack_revindex(p, ofs); | |
08698b1e NP |
1812 | if (!revidx) |
1813 | goto give_up; | |
3f83fd5e JK |
1814 | if (!nth_packed_object_id(&base_ref, p, revidx->nr)) |
1815 | have_base = 1; | |
3449f8c4 | 1816 | } |
5c49c116 NP |
1817 | entry->in_pack_header_size = used + used_0; |
1818 | break; | |
780e6e73 | 1819 | } |
780e6e73 | 1820 | |
3f83fd5e JK |
1821 | if (have_base && |
1822 | can_reuse_delta(&base_ref, entry, &base_entry)) { | |
fd9b1bae | 1823 | oe_set_type(entry, entry->in_pack_type); |
ac77d0c3 | 1824 | SET_SIZE(entry, in_pack_size); /* delta size */ |
0aca34e8 | 1825 | SET_DELTA_SIZE(entry, in_pack_size); |
6a1e32d5 JK |
1826 | |
1827 | if (base_entry) { | |
1828 | SET_DELTA(entry, base_entry); | |
1829 | entry->delta_sibling_idx = base_entry->delta_child_idx; | |
1830 | SET_DELTA_CHILD(base_entry, entry); | |
1831 | } else { | |
a93c141d | 1832 | SET_DELTA_EXT(entry, &base_ref); |
6a1e32d5 JK |
1833 | } |
1834 | ||
5c49c116 NP |
1835 | unuse_pack(&w_curs); |
1836 | return; | |
1837 | } | |
ab7cd7bb | 1838 | |
fd9b1bae | 1839 | if (oe_type(entry)) { |
27a7d067 NTND |
1840 | off_t delta_pos; |
1841 | ||
5c49c116 NP |
1842 | /* |
1843 | * This must be a delta and we already know what the | |
1844 | * final object type is. Let's extract the actual | |
1845 | * object size from the delta header. | |
1846 | */ | |
27a7d067 | 1847 | delta_pos = entry->in_pack_offset + entry->in_pack_header_size; |
ac77d0c3 NTND |
1848 | canonical_size = get_size_from_delta(p, &w_curs, delta_pos); |
1849 | if (canonical_size == 0) | |
03d66015 | 1850 | goto give_up; |
ac77d0c3 | 1851 | SET_SIZE(entry, canonical_size); |
5c49c116 | 1852 | unuse_pack(&w_curs); |
3f9ac8d2 JH |
1853 | return; |
1854 | } | |
5c49c116 NP |
1855 | |
1856 | /* | |
1857 | * No choice but to fall back to the recursive delta walk | |
c93206b4 | 1858 | * with oid_object_info() to find about the object type |
5c49c116 NP |
1859 | * at this point... |
1860 | */ | |
03d66015 | 1861 | give_up: |
5c49c116 | 1862 | unuse_pack(&w_curs); |
36e4d74a | 1863 | } |
3f9ac8d2 | 1864 | |
8d5cf957 | 1865 | if (oid_object_info_extended(the_repository, &entry->idx.oid, &oi, |
e00549aa JT |
1866 | OBJECT_INFO_SKIP_FETCH_OBJECT | OBJECT_INFO_LOOKUP_REPLACE) < 0) { |
1867 | if (has_promisor_remote()) { | |
1868 | prefetch_to_pack(object_index); | |
1869 | if (oid_object_info_extended(the_repository, &entry->idx.oid, &oi, | |
1870 | OBJECT_INFO_SKIP_FETCH_OBJECT | OBJECT_INFO_LOOKUP_REPLACE) < 0) | |
1871 | type = -1; | |
1872 | } else { | |
1873 | type = -1; | |
1874 | } | |
1875 | } | |
8d5cf957 | 1876 | oe_set_type(entry, type); |
ac77d0c3 NTND |
1877 | if (entry->type_valid) { |
1878 | SET_SIZE(entry, canonical_size); | |
1879 | } else { | |
1880 | /* | |
1881 | * Bad object type is checked in prepare_pack(). This is | |
1882 | * to permit a missing preferred base object to be ignored | |
1883 | * as a preferred base. Doing so can result in a larger | |
1884 | * pack file, but the transfer will still take place. | |
1885 | */ | |
1886 | } | |
3f9ac8d2 JH |
1887 | } |
1888 | ||
5c49c116 NP |
1889 | static int pack_offset_sort(const void *_a, const void *_b) |
1890 | { | |
1891 | const struct object_entry *a = *(struct object_entry **)_a; | |
1892 | const struct object_entry *b = *(struct object_entry **)_b; | |
43fa44fa NTND |
1893 | const struct packed_git *a_in_pack = IN_PACK(a); |
1894 | const struct packed_git *b_in_pack = IN_PACK(b); | |
5c49c116 NP |
1895 | |
1896 | /* avoid filesystem trashing with loose objects */ | |
43fa44fa | 1897 | if (!a_in_pack && !b_in_pack) |
e6a492b7 | 1898 | return oidcmp(&a->idx.oid, &b->idx.oid); |
5c49c116 | 1899 | |
43fa44fa | 1900 | if (a_in_pack < b_in_pack) |
5c49c116 | 1901 | return -1; |
43fa44fa | 1902 | if (a_in_pack > b_in_pack) |
5c49c116 NP |
1903 | return 1; |
1904 | return a->in_pack_offset < b->in_pack_offset ? -1 : | |
1905 | (a->in_pack_offset > b->in_pack_offset); | |
1906 | } | |
1907 | ||
4cf2143e JK |
1908 | /* |
1909 | * Drop an on-disk delta we were planning to reuse. Naively, this would | |
1910 | * just involve blanking out the "delta" field, but we have to deal | |
1911 | * with some extra book-keeping: | |
1912 | * | |
1913 | * 1. Removing ourselves from the delta_sibling linked list. | |
1914 | * | |
1915 | * 2. Updating our size/type to the non-delta representation. These were | |
1916 | * either not recorded initially (size) or overwritten with the delta type | |
1917 | * (type) when check_object() decided to reuse the delta. | |
7dbabbbe JK |
1918 | * |
1919 | * 3. Resetting our delta depth, as we are now a base object. | |
4cf2143e JK |
1920 | */ |
1921 | static void drop_reused_delta(struct object_entry *entry) | |
1922 | { | |
898eba5e | 1923 | unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx; |
4cf2143e | 1924 | struct object_info oi = OBJECT_INFO_INIT; |
fd9b1bae | 1925 | enum object_type type; |
ac77d0c3 | 1926 | unsigned long size; |
4cf2143e | 1927 | |
898eba5e NTND |
1928 | while (*idx) { |
1929 | struct object_entry *oe = &to_pack.objects[*idx - 1]; | |
4cf2143e | 1930 | |
898eba5e NTND |
1931 | if (oe == entry) |
1932 | *idx = oe->delta_sibling_idx; | |
4cf2143e | 1933 | else |
898eba5e | 1934 | idx = &oe->delta_sibling_idx; |
4cf2143e | 1935 | } |
898eba5e | 1936 | SET_DELTA(entry, NULL); |
7dbabbbe | 1937 | entry->depth = 0; |
4cf2143e | 1938 | |
ac77d0c3 | 1939 | oi.sizep = &size; |
fd9b1bae | 1940 | oi.typep = &type; |
ad635e82 | 1941 | if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) { |
4cf2143e JK |
1942 | /* |
1943 | * We failed to get the info from this pack for some reason; | |
c93206b4 | 1944 | * fall back to oid_object_info, which may find another copy. |
fd9b1bae | 1945 | * And if that fails, the error will be recorded in oe_type(entry) |
4cf2143e JK |
1946 | * and dealt with in prepare_pack(). |
1947 | */ | |
ad635e82 JH |
1948 | oe_set_type(entry, |
1949 | oid_object_info(the_repository, &entry->idx.oid, &size)); | |
fd9b1bae NTND |
1950 | } else { |
1951 | oe_set_type(entry, type); | |
4cf2143e | 1952 | } |
ac77d0c3 | 1953 | SET_SIZE(entry, size); |
4cf2143e JK |
1954 | } |
1955 | ||
1956 | /* | |
1957 | * Follow the chain of deltas from this entry onward, throwing away any links | |
1958 | * that cause us to hit a cycle (as determined by the DFS state flags in | |
1959 | * the entries). | |
7dbabbbe JK |
1960 | * |
1961 | * We also detect too-long reused chains that would violate our --depth | |
1962 | * limit. | |
4cf2143e JK |
1963 | */ |
1964 | static void break_delta_chains(struct object_entry *entry) | |
1965 | { | |
42b766d7 JK |
1966 | /* |
1967 | * The actual depth of each object we will write is stored as an int, | |
1968 | * as it cannot exceed our int "depth" limit. But before we break | |
1969 | * changes based no that limit, we may potentially go as deep as the | |
1970 | * number of objects, which is elsewhere bounded to a uint32_t. | |
1971 | */ | |
1972 | uint32_t total_depth; | |
1973 | struct object_entry *cur, *next; | |
1974 | ||
1975 | for (cur = entry, total_depth = 0; | |
1976 | cur; | |
898eba5e | 1977 | cur = DELTA(cur), total_depth++) { |
42b766d7 JK |
1978 | if (cur->dfs_state == DFS_DONE) { |
1979 | /* | |
1980 | * We've already seen this object and know it isn't | |
1981 | * part of a cycle. We do need to append its depth | |
1982 | * to our count. | |
1983 | */ | |
1984 | total_depth += cur->depth; | |
1985 | break; | |
1986 | } | |
4cf2143e | 1987 | |
4cf2143e | 1988 | /* |
42b766d7 JK |
1989 | * We break cycles before looping, so an ACTIVE state (or any |
1990 | * other cruft which made its way into the state variable) | |
1991 | * is a bug. | |
4cf2143e | 1992 | */ |
42b766d7 | 1993 | if (cur->dfs_state != DFS_NONE) |
033abf97 | 1994 | BUG("confusing delta dfs state in first pass: %d", |
42b766d7 | 1995 | cur->dfs_state); |
4cf2143e | 1996 | |
4cf2143e | 1997 | /* |
42b766d7 JK |
1998 | * Now we know this is the first time we've seen the object. If |
1999 | * it's not a delta, we're done traversing, but we'll mark it | |
2000 | * done to save time on future traversals. | |
4cf2143e | 2001 | */ |
898eba5e | 2002 | if (!DELTA(cur)) { |
42b766d7 JK |
2003 | cur->dfs_state = DFS_DONE; |
2004 | break; | |
2005 | } | |
4cf2143e | 2006 | |
4cf2143e | 2007 | /* |
42b766d7 JK |
2008 | * Mark ourselves as active and see if the next step causes |
2009 | * us to cycle to another active object. It's important to do | |
2010 | * this _before_ we loop, because it impacts where we make the | |
2011 | * cut, and thus how our total_depth counter works. | |
2012 | * E.g., We may see a partial loop like: | |
2013 | * | |
2014 | * A -> B -> C -> D -> B | |
2015 | * | |
2016 | * Cutting B->C breaks the cycle. But now the depth of A is | |
2017 | * only 1, and our total_depth counter is at 3. The size of the | |
2018 | * error is always one less than the size of the cycle we | |
2019 | * broke. Commits C and D were "lost" from A's chain. | |
2020 | * | |
2021 | * If we instead cut D->B, then the depth of A is correct at 3. | |
2022 | * We keep all commits in the chain that we examined. | |
4cf2143e | 2023 | */ |
42b766d7 | 2024 | cur->dfs_state = DFS_ACTIVE; |
898eba5e | 2025 | if (DELTA(cur)->dfs_state == DFS_ACTIVE) { |
42b766d7 JK |
2026 | drop_reused_delta(cur); |
2027 | cur->dfs_state = DFS_DONE; | |
2028 | break; | |
7dbabbbe | 2029 | } |
42b766d7 | 2030 | } |
7dbabbbe | 2031 | |
42b766d7 JK |
2032 | /* |
2033 | * And now that we've gone all the way to the bottom of the chain, we | |
2034 | * need to clear the active flags and set the depth fields as | |
2035 | * appropriate. Unlike the loop above, which can quit when it drops a | |
2036 | * delta, we need to keep going to look for more depth cuts. So we need | |
2037 | * an extra "next" pointer to keep going after we reset cur->delta. | |
2038 | */ | |
2039 | for (cur = entry; cur; cur = next) { | |
898eba5e | 2040 | next = DELTA(cur); |
4cf2143e | 2041 | |
42b766d7 JK |
2042 | /* |
2043 | * We should have a chain of zero or more ACTIVE states down to | |
2044 | * a final DONE. We can quit after the DONE, because either it | |
2045 | * has no bases, or we've already handled them in a previous | |
2046 | * call. | |
2047 | */ | |
2048 | if (cur->dfs_state == DFS_DONE) | |
2049 | break; | |
2050 | else if (cur->dfs_state != DFS_ACTIVE) | |
033abf97 | 2051 | BUG("confusing delta dfs state in second pass: %d", |
42b766d7 | 2052 | cur->dfs_state); |
4cf2143e | 2053 | |
4cf2143e | 2054 | /* |
42b766d7 JK |
2055 | * If the total_depth is more than depth, then we need to snip |
2056 | * the chain into two or more smaller chains that don't exceed | |
2057 | * the maximum depth. Most of the resulting chains will contain | |
2058 | * (depth + 1) entries (i.e., depth deltas plus one base), and | |
2059 | * the last chain (i.e., the one containing entry) will contain | |
2060 | * whatever entries are left over, namely | |
2061 | * (total_depth % (depth + 1)) of them. | |
2062 | * | |
2063 | * Since we are iterating towards decreasing depth, we need to | |
2064 | * decrement total_depth as we go, and we need to write to the | |
2065 | * entry what its final depth will be after all of the | |
2066 | * snipping. Since we're snipping into chains of length (depth | |
2067 | * + 1) entries, the final depth of an entry will be its | |
2068 | * original depth modulo (depth + 1). Any time we encounter an | |
2069 | * entry whose final depth is supposed to be zero, we snip it | |
2070 | * from its delta base, thereby making it so. | |
4cf2143e | 2071 | */ |
42b766d7 JK |
2072 | cur->depth = (total_depth--) % (depth + 1); |
2073 | if (!cur->depth) | |
2074 | drop_reused_delta(cur); | |
2075 | ||
2076 | cur->dfs_state = DFS_DONE; | |
4cf2143e JK |
2077 | } |
2078 | } | |
2079 | ||
c323ac7d LT |
2080 | static void get_object_details(void) |
2081 | { | |
7cadf491 | 2082 | uint32_t i; |
5c49c116 NP |
2083 | struct object_entry **sorted_by_offset; |
2084 | ||
5af05043 NTND |
2085 | if (progress) |
2086 | progress_state = start_progress(_("Counting objects"), | |
2087 | to_pack.nr_objects); | |
2088 | ||
2834bc27 VM |
2089 | sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *)); |
2090 | for (i = 0; i < to_pack.nr_objects; i++) | |
2091 | sorted_by_offset[i] = to_pack.objects + i; | |
9ed0d8d6 | 2092 | QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort); |
c323ac7d | 2093 | |
2834bc27 | 2094 | for (i = 0; i < to_pack.nr_objects; i++) { |
15366280 | 2095 | struct object_entry *entry = sorted_by_offset[i]; |
e00549aa | 2096 | check_object(entry, i); |
ac77d0c3 NTND |
2097 | if (entry->type_valid && |
2098 | oe_size_greater_than(&to_pack, entry, big_file_threshold)) | |
15366280 | 2099 | entry->no_try_delta = 1; |
5af05043 | 2100 | display_progress(progress_state, i + 1); |
15366280 | 2101 | } |
5af05043 | 2102 | stop_progress(&progress_state); |
3449f8c4 | 2103 | |
4cf2143e JK |
2104 | /* |
2105 | * This must happen in a second pass, since we rely on the delta | |
2106 | * information for the whole list being completed. | |
2107 | */ | |
2108 | for (i = 0; i < to_pack.nr_objects; i++) | |
2109 | break_delta_chains(&to_pack.objects[i]); | |
2110 | ||
5c49c116 | 2111 | free(sorted_by_offset); |
c323ac7d LT |
2112 | } |
2113 | ||
b904166c NP |
2114 | /* |
2115 | * We search for deltas in a list sorted by type, by filename hash, and then | |
2116 | * by size, so that we see progressively smaller and smaller files. | |
2117 | * That's because we prefer deltas to be from the bigger file | |
2118 | * to the smaller -- deletes are potentially cheaper, but perhaps | |
2119 | * more importantly, the bigger file is likely the more recent | |
2120 | * one. The deepest deltas are therefore the oldest objects which are | |
2121 | * less susceptible to be accessed often. | |
2122 | */ | |
9668cf59 | 2123 | static int type_size_sort(const void *_a, const void *_b) |
c323ac7d | 2124 | { |
9668cf59 NP |
2125 | const struct object_entry *a = *(struct object_entry **)_a; |
2126 | const struct object_entry *b = *(struct object_entry **)_b; | |
33de80b1 SL |
2127 | const enum object_type a_type = oe_type(a); |
2128 | const enum object_type b_type = oe_type(b); | |
2129 | const unsigned long a_size = SIZE(a); | |
2130 | const unsigned long b_size = SIZE(b); | |
9668cf59 | 2131 | |
fd9b1bae | 2132 | if (a_type > b_type) |
27225f2e | 2133 | return -1; |
fd9b1bae | 2134 | if (a_type < b_type) |
27225f2e | 2135 | return 1; |
b904166c | 2136 | if (a->hash > b->hash) |
7a979d99 | 2137 | return -1; |
b904166c | 2138 | if (a->hash < b->hash) |
7a979d99 | 2139 | return 1; |
b904166c | 2140 | if (a->preferred_base > b->preferred_base) |
c323ac7d | 2141 | return -1; |
b904166c NP |
2142 | if (a->preferred_base < b->preferred_base) |
2143 | return 1; | |
28b8a730 | 2144 | if (use_delta_islands) { |
33de80b1 | 2145 | const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid); |
28b8a730 JK |
2146 | if (island_cmp) |
2147 | return island_cmp; | |
2148 | } | |
ac77d0c3 | 2149 | if (a_size > b_size) |
b904166c | 2150 | return -1; |
ac77d0c3 | 2151 | if (a_size < b_size) |
c323ac7d | 2152 | return 1; |
b904166c | 2153 | return a < b ? -1 : (a > b); /* newest first */ |
c323ac7d LT |
2154 | } |
2155 | ||
2156 | struct unpacked { | |
2157 | struct object_entry *entry; | |
2158 | void *data; | |
f6c7081a | 2159 | struct delta_index *index; |
5a235b5e | 2160 | unsigned depth; |
c323ac7d LT |
2161 | }; |
2162 | ||
d250626c NP |
2163 | static int delta_cacheable(unsigned long src_size, unsigned long trg_size, |
2164 | unsigned long delta_size) | |
074b2eea MK |
2165 | { |
2166 | if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size) | |
2167 | return 0; | |
2168 | ||
e3dfddb3 MK |
2169 | if (delta_size < cache_max_small_delta_size) |
2170 | return 1; | |
2171 | ||
074b2eea MK |
2172 | /* cache delta, if objects are large enough compared to delta size */ |
2173 | if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10)) | |
2174 | return 1; | |
2175 | ||
2176 | return 0; | |
2177 | } | |
2178 | ||
ffbd51cc | 2179 | /* Protect delta_cache_size */ |
44626dc7 | 2180 | static pthread_mutex_t cache_mutex; |
3c701839 NP |
2181 | #define cache_lock() pthread_mutex_lock(&cache_mutex) |
2182 | #define cache_unlock() pthread_mutex_unlock(&cache_mutex) | |
2183 | ||
ffbd51cc NTND |
2184 | /* |
2185 | * Protect object list partitioning (e.g. struct thread_param) and | |
2186 | * progress_state | |
2187 | */ | |
44626dc7 | 2188 | static pthread_mutex_t progress_mutex; |
8ecce684 NP |
2189 | #define progress_lock() pthread_mutex_lock(&progress_mutex) |
2190 | #define progress_unlock() pthread_mutex_unlock(&progress_mutex) | |
2191 | ||
ffbd51cc NTND |
2192 | /* |
2193 | * Access to struct object_entry is unprotected since each thread owns | |
2194 | * a portion of the main object list. Just don't access object entries | |
2195 | * ahead in the list because they can be stolen and would need | |
2196 | * progress_mutex for protection. | |
2197 | */ | |
8ecce684 | 2198 | |
ac77d0c3 NTND |
2199 | /* |
2200 | * Return the size of the object without doing any delta | |
2201 | * reconstruction (so non-deltas are true object sizes, but deltas | |
2202 | * return the size of the delta data). | |
2203 | */ | |
2204 | unsigned long oe_get_size_slow(struct packing_data *pack, | |
2205 | const struct object_entry *e) | |
2206 | { | |
2207 | struct packed_git *p; | |
2208 | struct pack_window *w_curs; | |
2209 | unsigned char *buf; | |
2210 | enum object_type type; | |
2211 | unsigned long used, avail, size; | |
2212 | ||
2213 | if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) { | |
edb673cf | 2214 | packing_data_lock(&to_pack); |
ad635e82 | 2215 | if (oid_object_info(the_repository, &e->idx.oid, &size) < 0) |
ac77d0c3 NTND |
2216 | die(_("unable to get size of %s"), |
2217 | oid_to_hex(&e->idx.oid)); | |
edb673cf | 2218 | packing_data_unlock(&to_pack); |
ac77d0c3 NTND |
2219 | return size; |
2220 | } | |
2221 | ||
2222 | p = oe_in_pack(pack, e); | |
2223 | if (!p) | |
2224 | BUG("when e->type is a delta, it must belong to a pack"); | |
2225 | ||
edb673cf | 2226 | packing_data_lock(&to_pack); |
ac77d0c3 NTND |
2227 | w_curs = NULL; |
2228 | buf = use_pack(p, &w_curs, e->in_pack_offset, &avail); | |
2229 | used = unpack_object_header_buffer(buf, avail, &type, &size); | |
2230 | if (used == 0) | |
2231 | die(_("unable to parse object header of %s"), | |
2232 | oid_to_hex(&e->idx.oid)); | |
2233 | ||
2234 | unuse_pack(&w_curs); | |
edb673cf | 2235 | packing_data_unlock(&to_pack); |
ac77d0c3 NTND |
2236 | return size; |
2237 | } | |
2238 | ||
f6c7081a | 2239 | static int try_delta(struct unpacked *trg, struct unpacked *src, |
ef0316fc | 2240 | unsigned max_depth, unsigned long *mem_usage) |
c323ac7d | 2241 | { |
f6c7081a NP |
2242 | struct object_entry *trg_entry = trg->entry; |
2243 | struct object_entry *src_entry = src->entry; | |
560b25a8 | 2244 | unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz; |
c83f032e | 2245 | unsigned ref_depth; |
21666f1a | 2246 | enum object_type type; |
c323ac7d LT |
2247 | void *delta_buf; |
2248 | ||
2249 | /* Don't bother doing diffs between different types */ | |
fd9b1bae | 2250 | if (oe_type(trg_entry) != oe_type(src_entry)) |
c323ac7d LT |
2251 | return -1; |
2252 | ||
51d1e83f | 2253 | /* |
15f07e06 JK |
2254 | * We do not bother to try a delta that we discarded on an |
2255 | * earlier try, but only when reusing delta data. Note that | |
2256 | * src_entry that is marked as the preferred_base should always | |
2257 | * be considered, as even if we produce a suboptimal delta against | |
2258 | * it, we will still save the transfer cost, as we already know | |
2259 | * the other side has it and we won't send src_entry at all. | |
51d1e83f | 2260 | */ |
43fa44fa NTND |
2261 | if (reuse_delta && IN_PACK(trg_entry) && |
2262 | IN_PACK(trg_entry) == IN_PACK(src_entry) && | |
15f07e06 | 2263 | !src_entry->preferred_base && |
e9195b58 JH |
2264 | trg_entry->in_pack_type != OBJ_REF_DELTA && |
2265 | trg_entry->in_pack_type != OBJ_OFS_DELTA) | |
51d1e83f LT |
2266 | return 0; |
2267 | ||
898b14ce | 2268 | /* Let's not bust the allowed depth. */ |
5a235b5e | 2269 | if (src->depth >= max_depth) |
d116a45a | 2270 | return 0; |
c323ac7d | 2271 | |
c3b06a69 | 2272 | /* Now some size filtering heuristics. */ |
ac77d0c3 | 2273 | trg_size = SIZE(trg_entry); |
898eba5e | 2274 | if (!DELTA(trg_entry)) { |
41179100 | 2275 | max_size = trg_size/2 - the_hash_algo->rawsz; |
c83f032e NP |
2276 | ref_depth = 1; |
2277 | } else { | |
0aca34e8 | 2278 | max_size = DELTA_SIZE(trg_entry); |
5a235b5e | 2279 | ref_depth = trg->depth; |
c83f032e | 2280 | } |
720fe22d | 2281 | max_size = (uint64_t)max_size * (max_depth - src->depth) / |
c83f032e | 2282 | (max_depth - ref_depth + 1); |
c3b06a69 NP |
2283 | if (max_size == 0) |
2284 | return 0; | |
ac77d0c3 | 2285 | src_size = SIZE(src_entry); |
560b25a8 | 2286 | sizediff = src_size < trg_size ? trg_size - src_size : 0; |
27225f2e | 2287 | if (sizediff >= max_size) |
f527cb8c | 2288 | return 0; |
a1dab41a BD |
2289 | if (trg_size < src_size / 32) |
2290 | return 0; | |
f6c7081a | 2291 | |
28b8a730 JK |
2292 | if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid)) |
2293 | return 0; | |
2294 | ||
560b25a8 NP |
2295 | /* Load data if not already done */ |
2296 | if (!trg->data) { | |
edb673cf | 2297 | packing_data_lock(&to_pack); |
b4f5aca4 | 2298 | trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz); |
edb673cf | 2299 | packing_data_unlock(&to_pack); |
2e3404c3 | 2300 | if (!trg->data) |
f616db6a | 2301 | die(_("object %s cannot be read"), |
e6a492b7 | 2302 | oid_to_hex(&trg_entry->idx.oid)); |
560b25a8 | 2303 | if (sz != trg_size) |
ca473cef TB |
2304 | die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"), |
2305 | oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz, | |
2306 | (uintmax_t)trg_size); | |
ef0316fc | 2307 | *mem_usage += sz; |
560b25a8 NP |
2308 | } |
2309 | if (!src->data) { | |
edb673cf | 2310 | packing_data_lock(&to_pack); |
b4f5aca4 | 2311 | src->data = read_object_file(&src_entry->idx.oid, &type, &sz); |
edb673cf | 2312 | packing_data_unlock(&to_pack); |
71064a95 NP |
2313 | if (!src->data) { |
2314 | if (src_entry->preferred_base) { | |
2315 | static int warned = 0; | |
2316 | if (!warned++) | |
f616db6a | 2317 | warning(_("object %s cannot be read"), |
e6a492b7 | 2318 | oid_to_hex(&src_entry->idx.oid)); |
71064a95 NP |
2319 | /* |
2320 | * Those objects are not included in the | |
2321 | * resulting pack. Be resilient and ignore | |
2322 | * them if they can't be read, in case the | |
2323 | * pack could be created nevertheless. | |
2324 | */ | |
2325 | return 0; | |
2326 | } | |
f616db6a | 2327 | die(_("object %s cannot be read"), |
e6a492b7 | 2328 | oid_to_hex(&src_entry->idx.oid)); |
71064a95 | 2329 | } |
560b25a8 | 2330 | if (sz != src_size) |
ca473cef TB |
2331 | die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"), |
2332 | oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz, | |
2333 | (uintmax_t)src_size); | |
ef0316fc | 2334 | *mem_usage += sz; |
560b25a8 NP |
2335 | } |
2336 | if (!src->index) { | |
2337 | src->index = create_delta_index(src->data, src_size); | |
a588d88a MK |
2338 | if (!src->index) { |
2339 | static int warned = 0; | |
2340 | if (!warned++) | |
f616db6a | 2341 | warning(_("suboptimal pack - out of memory")); |
a588d88a MK |
2342 | return 0; |
2343 | } | |
ef0316fc | 2344 | *mem_usage += sizeof_delta_index(src->index); |
560b25a8 NP |
2345 | } |
2346 | ||
2347 | delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size); | |
c323ac7d | 2348 | if (!delta_buf) |
75c42d8c | 2349 | return 0; |
f6c7081a | 2350 | |
898eba5e | 2351 | if (DELTA(trg_entry)) { |
848d732c | 2352 | /* Prefer only shallower same-sized deltas. */ |
0aca34e8 | 2353 | if (delta_size == DELTA_SIZE(trg_entry) && |
5a235b5e | 2354 | src->depth + 1 >= trg->depth) { |
848d732c BD |
2355 | free(delta_buf); |
2356 | return 0; | |
2357 | } | |
074b2eea | 2358 | } |
9e2d57a0 | 2359 | |
3c701839 NP |
2360 | /* |
2361 | * Handle memory allocation outside of the cache | |
2362 | * accounting lock. Compiler will optimize the strangeness | |
7eb151d6 | 2363 | * away when NO_PTHREADS is defined. |
3c701839 | 2364 | */ |
8e0f7003 | 2365 | free(trg_entry->delta_data); |
3c701839 | 2366 | cache_lock(); |
9e2d57a0 | 2367 | if (trg_entry->delta_data) { |
0aca34e8 | 2368 | delta_cache_size -= DELTA_SIZE(trg_entry); |
9e2d57a0 NP |
2369 | trg_entry->delta_data = NULL; |
2370 | } | |
d250626c | 2371 | if (delta_cacheable(src_size, trg_size, delta_size)) { |
b7a28f78 | 2372 | delta_cache_size += delta_size; |
3c701839 NP |
2373 | cache_unlock(); |
2374 | trg_entry->delta_data = xrealloc(delta_buf, delta_size); | |
2375 | } else { | |
2376 | cache_unlock(); | |
074b2eea | 2377 | free(delta_buf); |
3c701839 NP |
2378 | } |
2379 | ||
898eba5e | 2380 | SET_DELTA(trg_entry, src_entry); |
0aca34e8 | 2381 | SET_DELTA_SIZE(trg_entry, delta_size); |
b7a28f78 NP |
2382 | trg->depth = src->depth + 1; |
2383 | ||
f6c7081a | 2384 | return 1; |
c323ac7d LT |
2385 | } |
2386 | ||
898b14ce | 2387 | static unsigned int check_delta_limit(struct object_entry *me, unsigned int n) |
b2504a0d | 2388 | { |
898eba5e | 2389 | struct object_entry *child = DELTA_CHILD(me); |
898b14ce NP |
2390 | unsigned int m = n; |
2391 | while (child) { | |
33de80b1 | 2392 | const unsigned int c = check_delta_limit(child, n + 1); |
898b14ce NP |
2393 | if (m < c) |
2394 | m = c; | |
898eba5e | 2395 | child = DELTA_SIBLING(child); |
898b14ce NP |
2396 | } |
2397 | return m; | |
b2504a0d NP |
2398 | } |
2399 | ||
75ad235c | 2400 | static unsigned long free_unpacked(struct unpacked *n) |
a97773ce | 2401 | { |
ef0316fc | 2402 | unsigned long freed_mem = sizeof_delta_index(n->index); |
a97773ce BD |
2403 | free_delta_index(n->index); |
2404 | n->index = NULL; | |
2405 | if (n->data) { | |
ac77d0c3 | 2406 | freed_mem += SIZE(n->entry); |
6a83d902 | 2407 | FREE_AND_NULL(n->data); |
a97773ce BD |
2408 | } |
2409 | n->entry = NULL; | |
7d7baa5e | 2410 | n->depth = 0; |
ef0316fc | 2411 | return freed_mem; |
a97773ce BD |
2412 | } |
2413 | ||
384b32c0 | 2414 | static void find_deltas(struct object_entry **list, unsigned *list_size, |
e334977d | 2415 | int window, int depth, unsigned *processed) |
c323ac7d | 2416 | { |
384b32c0 | 2417 | uint32_t i, idx = 0, count = 0; |
7cadf491 | 2418 | struct unpacked *array; |
ef0316fc | 2419 | unsigned long mem_usage = 0; |
c323ac7d | 2420 | |
19d4b416 | 2421 | array = xcalloc(window, sizeof(struct unpacked)); |
21fcd1bd | 2422 | |
384b32c0 | 2423 | for (;;) { |
421b488a | 2424 | struct object_entry *entry; |
c323ac7d | 2425 | struct unpacked *n = array + idx; |
ef0316fc | 2426 | int j, max_depth, best_base = -1; |
c323ac7d | 2427 | |
384b32c0 NP |
2428 | progress_lock(); |
2429 | if (!*list_size) { | |
2430 | progress_unlock(); | |
2431 | break; | |
2432 | } | |
421b488a | 2433 | entry = *list++; |
384b32c0 NP |
2434 | (*list_size)--; |
2435 | if (!entry->preferred_base) { | |
2436 | (*processed)++; | |
2437 | display_progress(progress_state, *processed); | |
2438 | } | |
2439 | progress_unlock(); | |
2440 | ||
ef0316fc | 2441 | mem_usage -= free_unpacked(n); |
c323ac7d | 2442 | n->entry = entry; |
ab7cd7bb | 2443 | |
a97773ce | 2444 | while (window_memory_limit && |
ef0316fc | 2445 | mem_usage > window_memory_limit && |
a97773ce | 2446 | count > 1) { |
33de80b1 | 2447 | const uint32_t tail = (idx + window - count) % window; |
75ad235c | 2448 | mem_usage -= free_unpacked(array + tail); |
a97773ce BD |
2449 | count--; |
2450 | } | |
2451 | ||
75d39853 NP |
2452 | /* We do not compute delta to *create* objects we are not |
2453 | * going to pack. | |
2454 | */ | |
2455 | if (entry->preferred_base) | |
2456 | goto next; | |
2457 | ||
898b14ce NP |
2458 | /* |
2459 | * If the current object is at pack edge, take the depth the | |
2460 | * objects that depend on the current object into account | |
2461 | * otherwise they would become too deep. | |
2462 | */ | |
2463 | max_depth = depth; | |
898eba5e | 2464 | if (DELTA_CHILD(entry)) { |
898b14ce NP |
2465 | max_depth -= check_delta_limit(entry, 0); |
2466 | if (max_depth <= 0) | |
2467 | goto next; | |
2468 | } | |
2469 | ||
78817c15 LT |
2470 | j = window; |
2471 | while (--j > 0) { | |
77639870 | 2472 | int ret; |
7cadf491 | 2473 | uint32_t other_idx = idx + j; |
c323ac7d | 2474 | struct unpacked *m; |
78817c15 LT |
2475 | if (other_idx >= window) |
2476 | other_idx -= window; | |
c323ac7d LT |
2477 | m = array + other_idx; |
2478 | if (!m->entry) | |
2479 | break; | |
ef0316fc | 2480 | ret = try_delta(n, m, max_depth, &mem_usage); |
77639870 | 2481 | if (ret < 0) |
c323ac7d | 2482 | break; |
77639870 JH |
2483 | else if (ret > 0) |
2484 | best_base = other_idx; | |
c323ac7d | 2485 | } |
898b14ce | 2486 | |
ed4a9031 NP |
2487 | /* |
2488 | * If we decided to cache the delta data, then it is best | |
2489 | * to compress it right away. First because we have to do | |
2490 | * it anyway, and doing it here while we're threaded will | |
2491 | * save a lot of time in the non threaded write phase, | |
2492 | * as well as allow for caching more deltas within | |
2493 | * the same cache size limit. | |
2494 | * ... | |
2495 | * But only if not writing to stdout, since in that case | |
2496 | * the network is most likely throttling writes anyway, | |
2497 | * and therefore it is best to go to the write phase ASAP | |
2498 | * instead, as we can afford spending more time compressing | |
2499 | * between writes at that moment. | |
2500 | */ | |
2501 | if (entry->delta_data && !pack_to_stdout) { | |
0cb3c142 NTND |
2502 | unsigned long size; |
2503 | ||
0aca34e8 | 2504 | size = do_compress(&entry->delta_data, DELTA_SIZE(entry)); |
0cb3c142 NTND |
2505 | if (size < (1U << OE_Z_DELTA_BITS)) { |
2506 | entry->z_delta_size = size; | |
2507 | cache_lock(); | |
0aca34e8 | 2508 | delta_cache_size -= DELTA_SIZE(entry); |
0cb3c142 NTND |
2509 | delta_cache_size += entry->z_delta_size; |
2510 | cache_unlock(); | |
2511 | } else { | |
2512 | FREE_AND_NULL(entry->delta_data); | |
2513 | entry->z_delta_size = 0; | |
2514 | } | |
ed4a9031 NP |
2515 | } |
2516 | ||
70ca1a3f JH |
2517 | /* if we made n a delta, and if n is already at max |
2518 | * depth, leaving it in the window is pointless. we | |
2519 | * should evict it first. | |
70ca1a3f | 2520 | */ |
898eba5e | 2521 | if (DELTA(entry) && max_depth <= n->depth) |
70ca1a3f | 2522 | continue; |
ff45715c | 2523 | |
77639870 JH |
2524 | /* |
2525 | * Move the best delta base up in the window, after the | |
2526 | * currently deltified object, to keep it longer. It will | |
2527 | * be the first base object to be attempted next. | |
2528 | */ | |
898eba5e | 2529 | if (DELTA(entry)) { |
77639870 JH |
2530 | struct unpacked swap = array[best_base]; |
2531 | int dist = (window + idx - best_base) % window; | |
2532 | int dst = best_base; | |
2533 | while (dist--) { | |
2534 | int src = (dst + 1) % window; | |
2535 | array[dst] = array[src]; | |
2536 | dst = src; | |
2537 | } | |
2538 | array[dst] = swap; | |
2539 | } | |
2540 | ||
898b14ce | 2541 | next: |
521a4f4c | 2542 | idx++; |
a97773ce BD |
2543 | if (count + 1 < window) |
2544 | count++; | |
521a4f4c LT |
2545 | if (idx >= window) |
2546 | idx = 0; | |
384b32c0 | 2547 | } |
adee7bdf | 2548 | |
f6c7081a | 2549 | for (i = 0; i < window; ++i) { |
ff45715c | 2550 | free_delta_index(array[i].index); |
adee7bdf | 2551 | free(array[i].data); |
f6c7081a | 2552 | } |
adee7bdf | 2553 | free(array); |
c323ac7d LT |
2554 | } |
2555 | ||
50f22ada | 2556 | /* |
ffbd51cc NTND |
2557 | * The main object list is split into smaller lists, each is handed to |
2558 | * one worker. | |
2559 | * | |
50f22ada JS |
2560 | * The main thread waits on the condition that (at least) one of the workers |
2561 | * has stopped working (which is indicated in the .working member of | |
2562 | * struct thread_params). | |
ffbd51cc | 2563 | * |
50f22ada JS |
2564 | * When a work thread has completed its work, it sets .working to 0 and |
2565 | * signals the main thread and waits on the condition that .data_ready | |
2566 | * becomes 1. | |
ffbd51cc NTND |
2567 | * |
2568 | * The main thread steals half of the work from the worker that has | |
2569 | * most work left to hand it to the idle worker. | |
50f22ada JS |
2570 | */ |
2571 | ||
8ecce684 NP |
2572 | struct thread_params { |
2573 | pthread_t thread; | |
2574 | struct object_entry **list; | |
2575 | unsigned list_size; | |
384b32c0 | 2576 | unsigned remaining; |
8ecce684 NP |
2577 | int window; |
2578 | int depth; | |
50f22ada JS |
2579 | int working; |
2580 | int data_ready; | |
2581 | pthread_mutex_t mutex; | |
2582 | pthread_cond_t cond; | |
8ecce684 NP |
2583 | unsigned *processed; |
2584 | }; | |
2585 | ||
44626dc7 AH |
2586 | static pthread_cond_t progress_cond; |
2587 | ||
2588 | /* | |
2589 | * Mutex and conditional variable can't be statically-initialized on Windows. | |
2590 | */ | |
2591 | static void init_threaded_search(void) | |
2592 | { | |
44626dc7 AH |
2593 | pthread_mutex_init(&cache_mutex, NULL); |
2594 | pthread_mutex_init(&progress_mutex, NULL); | |
2595 | pthread_cond_init(&progress_cond, NULL); | |
2596 | } | |
2597 | ||
2598 | static void cleanup_threaded_search(void) | |
2599 | { | |
2600 | pthread_cond_destroy(&progress_cond); | |
44626dc7 AH |
2601 | pthread_mutex_destroy(&cache_mutex); |
2602 | pthread_mutex_destroy(&progress_mutex); | |
2603 | } | |
c2a33679 | 2604 | |
8ecce684 NP |
2605 | static void *threaded_find_deltas(void *arg) |
2606 | { | |
c2a33679 NP |
2607 | struct thread_params *me = arg; |
2608 | ||
0c2ad00b | 2609 | progress_lock(); |
50f22ada | 2610 | while (me->remaining) { |
0c2ad00b MÅ |
2611 | progress_unlock(); |
2612 | ||
384b32c0 | 2613 | find_deltas(me->list, &me->remaining, |
c2a33679 | 2614 | me->window, me->depth, me->processed); |
50f22ada JS |
2615 | |
2616 | progress_lock(); | |
2617 | me->working = 0; | |
2618 | pthread_cond_signal(&progress_cond); | |
2619 | progress_unlock(); | |
2620 | ||
2621 | /* | |
2622 | * We must not set ->data_ready before we wait on the | |
2623 | * condition because the main thread may have set it to 1 | |
2624 | * before we get here. In order to be sure that new | |
2625 | * work is available if we see 1 in ->data_ready, it | |
2626 | * was initialized to 0 before this thread was spawned | |
2627 | * and we reset it to 0 right away. | |
2628 | */ | |
2629 | pthread_mutex_lock(&me->mutex); | |
2630 | while (!me->data_ready) | |
2631 | pthread_cond_wait(&me->cond, &me->mutex); | |
2632 | me->data_ready = 0; | |
2633 | pthread_mutex_unlock(&me->mutex); | |
0c2ad00b MÅ |
2634 | |
2635 | progress_lock(); | |
c2a33679 | 2636 | } |
0c2ad00b | 2637 | progress_unlock(); |
50f22ada JS |
2638 | /* leave ->working 1 so that this doesn't get more work assigned */ |
2639 | return NULL; | |
8ecce684 NP |
2640 | } |
2641 | ||
8ecce684 NP |
2642 | static void ll_find_deltas(struct object_entry **list, unsigned list_size, |
2643 | int window, int depth, unsigned *processed) | |
2644 | { | |
dcda3614 | 2645 | struct thread_params *p; |
384b32c0 | 2646 | int i, ret, active_threads = 0; |
c2a33679 | 2647 | |
44626dc7 AH |
2648 | init_threaded_search(); |
2649 | ||
367f4a43 | 2650 | if (delta_search_threads <= 1) { |
384b32c0 | 2651 | find_deltas(list, &list_size, window, depth, processed); |
44626dc7 | 2652 | cleanup_threaded_search(); |
367f4a43 NP |
2653 | return; |
2654 | } | |
43cc2b42 | 2655 | if (progress > pack_to_stdout) |
f616db6a | 2656 | fprintf_ln(stderr, _("Delta compression using up to %d threads"), |
1a07e59c | 2657 | delta_search_threads); |
dcda3614 | 2658 | p = xcalloc(delta_search_threads, sizeof(*p)); |
367f4a43 | 2659 | |
50f22ada | 2660 | /* Partition the work amongst work threads. */ |
367f4a43 | 2661 | for (i = 0; i < delta_search_threads; i++) { |
50f22ada JS |
2662 | unsigned sub_size = list_size / (delta_search_threads - i); |
2663 | ||
bf874896 NP |
2664 | /* don't use too small segments or no deltas will be found */ |
2665 | if (sub_size < 2*window && i+1 < delta_search_threads) | |
2666 | sub_size = 0; | |
2667 | ||
8ecce684 NP |
2668 | p[i].window = window; |
2669 | p[i].depth = depth; | |
2670 | p[i].processed = processed; | |
50f22ada JS |
2671 | p[i].working = 1; |
2672 | p[i].data_ready = 0; | |
c2a33679 | 2673 | |
59921b4b | 2674 | /* try to split chunks on "path" boundaries */ |
6fc74703 NP |
2675 | while (sub_size && sub_size < list_size && |
2676 | list[sub_size]->hash && | |
384b32c0 NP |
2677 | list[sub_size]->hash == list[sub_size-1]->hash) |
2678 | sub_size++; | |
2679 | ||
50f22ada JS |
2680 | p[i].list = list; |
2681 | p[i].list_size = sub_size; | |
2682 | p[i].remaining = sub_size; | |
59921b4b | 2683 | |
384b32c0 NP |
2684 | list += sub_size; |
2685 | list_size -= sub_size; | |
2686 | } | |
2687 | ||
50f22ada JS |
2688 | /* Start work threads. */ |
2689 | for (i = 0; i < delta_search_threads; i++) { | |
2690 | if (!p[i].list_size) | |
2691 | continue; | |
68e6a4f8 JS |
2692 | pthread_mutex_init(&p[i].mutex, NULL); |
2693 | pthread_cond_init(&p[i].cond, NULL); | |
50f22ada JS |
2694 | ret = pthread_create(&p[i].thread, NULL, |
2695 | threaded_find_deltas, &p[i]); | |
2696 | if (ret) | |
f616db6a | 2697 | die(_("unable to create thread: %s"), strerror(ret)); |
50f22ada JS |
2698 | active_threads++; |
2699 | } | |
2700 | ||
384b32c0 NP |
2701 | /* |
2702 | * Now let's wait for work completion. Each time a thread is done | |
2703 | * with its work, we steal half of the remaining work from the | |
2704 | * thread with the largest number of unprocessed objects and give | |
2705 | * it to that newly idle thread. This ensure good load balancing | |
2706 | * until the remaining object list segments are simply too short | |
2707 | * to be worth splitting anymore. | |
2708 | */ | |
50f22ada JS |
2709 | while (active_threads) { |
2710 | struct thread_params *target = NULL; | |
384b32c0 NP |
2711 | struct thread_params *victim = NULL; |
2712 | unsigned sub_size = 0; | |
384b32c0 NP |
2713 | |
2714 | progress_lock(); | |
50f22ada JS |
2715 | for (;;) { |
2716 | for (i = 0; !target && i < delta_search_threads; i++) | |
2717 | if (!p[i].working) | |
2718 | target = &p[i]; | |
2719 | if (target) | |
2720 | break; | |
2721 | pthread_cond_wait(&progress_cond, &progress_mutex); | |
2722 | } | |
2723 | ||
384b32c0 NP |
2724 | for (i = 0; i < delta_search_threads; i++) |
2725 | if (p[i].remaining > 2*window && | |
2726 | (!victim || victim->remaining < p[i].remaining)) | |
2727 | victim = &p[i]; | |
2728 | if (victim) { | |
2729 | sub_size = victim->remaining / 2; | |
2730 | list = victim->list + victim->list_size - sub_size; | |
2731 | while (sub_size && list[0]->hash && | |
2732 | list[0]->hash == list[-1]->hash) { | |
2733 | list++; | |
2734 | sub_size--; | |
2735 | } | |
eb9688ff NP |
2736 | if (!sub_size) { |
2737 | /* | |
2738 | * It is possible for some "paths" to have | |
2739 | * so many objects that no hash boundary | |
2740 | * might be found. Let's just steal the | |
2741 | * exact half in that case. | |
2742 | */ | |
2743 | sub_size = victim->remaining / 2; | |
2744 | list -= sub_size; | |
2745 | } | |
384b32c0 NP |
2746 | target->list = list; |
2747 | victim->list_size -= sub_size; | |
2748 | victim->remaining -= sub_size; | |
2749 | } | |
384b32c0 NP |
2750 | target->list_size = sub_size; |
2751 | target->remaining = sub_size; | |
50f22ada JS |
2752 | target->working = 1; |
2753 | progress_unlock(); | |
2754 | ||
2755 | pthread_mutex_lock(&target->mutex); | |
2756 | target->data_ready = 1; | |
2757 | pthread_cond_signal(&target->cond); | |
2758 | pthread_mutex_unlock(&target->mutex); | |
c2a33679 | 2759 | |
384b32c0 | 2760 | if (!sub_size) { |
b81d9af7 | 2761 | pthread_join(target->thread, NULL); |
50f22ada JS |
2762 | pthread_cond_destroy(&target->cond); |
2763 | pthread_mutex_destroy(&target->mutex); | |
384b32c0 | 2764 | active_threads--; |
c2a33679 | 2765 | } |
50f22ada | 2766 | } |
44626dc7 | 2767 | cleanup_threaded_search(); |
dcda3614 | 2768 | free(p); |
8ecce684 NP |
2769 | } |
2770 | ||
ff483026 JK |
2771 | static int obj_is_packed(const struct object_id *oid) |
2772 | { | |
a14aebea | 2773 | return packlist_find(&to_pack, oid) || |
92fb0db9 JK |
2774 | (reuse_packfile_bitmap && |
2775 | bitmap_walk_contains(bitmap_git, reuse_packfile_bitmap, oid)); | |
ff483026 JK |
2776 | } |
2777 | ||
b773ddea JK |
2778 | static void add_tag_chain(const struct object_id *oid) |
2779 | { | |
2780 | struct tag *tag; | |
2781 | ||
2782 | /* | |
2783 | * We catch duplicates already in add_object_entry(), but we'd | |
2784 | * prefer to do this extra check to avoid having to parse the | |
2785 | * tag at all if we already know that it's being packed (e.g., if | |
2786 | * it was included via bitmaps, we would not have parsed it | |
2787 | * previously). | |
2788 | */ | |
ff483026 | 2789 | if (obj_is_packed(oid)) |
b773ddea JK |
2790 | return; |
2791 | ||
ce71efb7 | 2792 | tag = lookup_tag(the_repository, oid); |
b773ddea JK |
2793 | while (1) { |
2794 | if (!tag || parse_tag(tag) || !tag->tagged) | |
f616db6a | 2795 | die(_("unable to pack objects reachable from tag %s"), |
b773ddea JK |
2796 | oid_to_hex(oid)); |
2797 | ||
188960b4 | 2798 | add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0); |
b773ddea JK |
2799 | |
2800 | if (tag->tagged->type != OBJ_TAG) | |
2801 | return; | |
2802 | ||
2803 | tag = (struct tag *)tag->tagged; | |
2804 | } | |
2805 | } | |
2806 | ||
d155254c | 2807 | static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data) |
f0a24aa5 | 2808 | { |
d155254c | 2809 | struct object_id peeled; |
f0a24aa5 | 2810 | |
59556548 | 2811 | if (starts_with(path, "refs/tags/") && /* is a tag? */ |
b420d909 | 2812 | !peel_ref(path, &peeled) && /* peelable? */ |
ff483026 | 2813 | obj_is_packed(&peeled)) /* object packed? */ |
b773ddea | 2814 | add_tag_chain(oid); |
f0a24aa5 SP |
2815 | return 0; |
2816 | } | |
2817 | ||
f3123c4a JH |
2818 | static void prepare_pack(int window, int depth) |
2819 | { | |
9668cf59 | 2820 | struct object_entry **delta_list; |
6e1c2344 RJ |
2821 | uint32_t i, nr_deltas; |
2822 | unsigned n; | |
9668cf59 | 2823 | |
28b8a730 | 2824 | if (use_delta_islands) |
385cb64f | 2825 | resolve_tree_islands(the_repository, progress, &to_pack); |
28b8a730 | 2826 | |
3f9ac8d2 | 2827 | get_object_details(); |
9668cf59 | 2828 | |
0e8189e2 NP |
2829 | /* |
2830 | * If we're locally repacking then we need to be doubly careful | |
2831 | * from now on in order to make sure no stealth corruption gets | |
2832 | * propagated to the new pack. Clients receiving streamed packs | |
2833 | * should validate everything they get anyway so no need to incur | |
2834 | * the additional cost here in that case. | |
2835 | */ | |
2836 | if (!pack_to_stdout) | |
2837 | do_check_packed_object_crc = 1; | |
2838 | ||
2834bc27 | 2839 | if (!to_pack.nr_objects || !window || !depth) |
9668cf59 NP |
2840 | return; |
2841 | ||
b32fa95f | 2842 | ALLOC_ARRAY(delta_list, to_pack.nr_objects); |
75d39853 NP |
2843 | nr_deltas = n = 0; |
2844 | ||
2834bc27 VM |
2845 | for (i = 0; i < to_pack.nr_objects; i++) { |
2846 | struct object_entry *entry = to_pack.objects + i; | |
75d39853 | 2847 | |
898eba5e | 2848 | if (DELTA(entry)) |
75d39853 | 2849 | /* This happens if we decided to reuse existing |
a7de7130 | 2850 | * delta from a pack. "reuse_delta &&" is implied. |
75d39853 NP |
2851 | */ |
2852 | continue; | |
2853 | ||
ac77d0c3 NTND |
2854 | if (!entry->type_valid || |
2855 | oe_size_less_than(&to_pack, entry, 50)) | |
75d39853 NP |
2856 | continue; |
2857 | ||
2858 | if (entry->no_try_delta) | |
2859 | continue; | |
2860 | ||
6d6f9cdd | 2861 | if (!entry->preferred_base) { |
75d39853 | 2862 | nr_deltas++; |
fd9b1bae | 2863 | if (oe_type(entry) < 0) |
f616db6a | 2864 | die(_("unable to get type of object %s"), |
e6a492b7 | 2865 | oid_to_hex(&entry->idx.oid)); |
eede9f42 | 2866 | } else { |
fd9b1bae | 2867 | if (oe_type(entry) < 0) { |
eede9f42 NP |
2868 | /* |
2869 | * This object is not found, but we | |
2870 | * don't have to include it anyway. | |
2871 | */ | |
2872 | continue; | |
2873 | } | |
6d6f9cdd | 2874 | } |
75d39853 NP |
2875 | |
2876 | delta_list[n++] = entry; | |
2877 | } | |
2878 | ||
2f8b8947 | 2879 | if (nr_deltas && n > 1) { |
e334977d | 2880 | unsigned nr_done = 0; |
bb514de3 | 2881 | |
e334977d | 2882 | if (progress) |
754dbc43 | 2883 | progress_state = start_progress(_("Compressing objects"), |
dc6a0757 | 2884 | nr_deltas); |
9ed0d8d6 | 2885 | QSORT(delta_list, n, type_size_sort); |
8ecce684 | 2886 | ll_find_deltas(delta_list, n, window+1, depth, &nr_done); |
4d4fcc54 | 2887 | stop_progress(&progress_state); |
e334977d | 2888 | if (nr_done != nr_deltas) |
f616db6a | 2889 | die(_("inconsistency with delta count")); |
75d39853 | 2890 | } |
9668cf59 | 2891 | free(delta_list); |
f3123c4a JH |
2892 | } |
2893 | ||
ef90d6d4 | 2894 | static int git_pack_config(const char *k, const char *v, void *cb) |
4812a93a | 2895 | { |
eeefa7c9 | 2896 | if (!strcmp(k, "pack.window")) { |
4812a93a JK |
2897 | window = git_config_int(k, v); |
2898 | return 0; | |
2899 | } | |
a97773ce BD |
2900 | if (!strcmp(k, "pack.windowmemory")) { |
2901 | window_memory_limit = git_config_ulong(k, v); | |
2902 | return 0; | |
2903 | } | |
2904 | if (!strcmp(k, "pack.depth")) { | |
842aaf93 TT |
2905 | depth = git_config_int(k, v); |
2906 | return 0; | |
2907 | } | |
074b2eea MK |
2908 | if (!strcmp(k, "pack.deltacachesize")) { |
2909 | max_delta_cache_size = git_config_int(k, v); | |
2910 | return 0; | |
2911 | } | |
e3dfddb3 MK |
2912 | if (!strcmp(k, "pack.deltacachelimit")) { |
2913 | cache_max_small_delta_size = git_config_int(k, v); | |
2914 | return 0; | |
2915 | } | |
ae4f07fb VM |
2916 | if (!strcmp(k, "pack.writebitmaphashcache")) { |
2917 | if (git_config_bool(k, v)) | |
2918 | write_bitmap_options |= BITMAP_OPT_HASH_CACHE; | |
2919 | else | |
2920 | write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE; | |
2921 | } | |
6b8fda2d | 2922 | if (!strcmp(k, "pack.usebitmaps")) { |
645c432d | 2923 | use_bitmap_index_default = git_config_bool(k, v); |
6b8fda2d VM |
2924 | return 0; |
2925 | } | |
e704fc79 JK |
2926 | if (!strcmp(k, "pack.allowpackreuse")) { |
2927 | allow_pack_reuse = git_config_bool(k, v); | |
2928 | return 0; | |
2929 | } | |
693b86ff NP |
2930 | if (!strcmp(k, "pack.threads")) { |
2931 | delta_search_threads = git_config_int(k, v); | |
833e3df1 | 2932 | if (delta_search_threads < 0) |
f616db6a | 2933 | die(_("invalid number of threads specified (%d)"), |
693b86ff | 2934 | delta_search_threads); |
9c897c5c | 2935 | if (!HAVE_THREADS && delta_search_threads != 1) { |
f616db6a | 2936 | warning(_("no threads support, ignoring %s"), k); |
2e96d815 ÆAB |
2937 | delta_search_threads = 0; |
2938 | } | |
693b86ff NP |
2939 | return 0; |
2940 | } | |
4d00bda2 | 2941 | if (!strcmp(k, "pack.indexversion")) { |
ebcfb379 JH |
2942 | pack_idx_opts.version = git_config_int(k, v); |
2943 | if (pack_idx_opts.version > 2) | |
f616db6a | 2944 | die(_("bad pack.indexversion=%"PRIu32), |
ebcfb379 | 2945 | pack_idx_opts.version); |
4d00bda2 NP |
2946 | return 0; |
2947 | } | |
dd4b732d JT |
2948 | if (!strcmp(k, "uploadpack.blobpackfileuri")) { |
2949 | struct configured_exclusion *ex = xmalloc(sizeof(*ex)); | |
2950 | const char *oid_end, *pack_end; | |
2951 | /* | |
2952 | * Stores the pack hash. This is not a true object ID, but is | |
2953 | * of the same form. | |
2954 | */ | |
2955 | struct object_id pack_hash; | |
2956 | ||
2957 | if (parse_oid_hex(v, &ex->e.oid, &oid_end) || | |
2958 | *oid_end != ' ' || | |
2959 | parse_oid_hex(oid_end + 1, &pack_hash, &pack_end) || | |
2960 | *pack_end != ' ') | |
2961 | die(_("value of uploadpack.blobpackfileuri must be " | |
2962 | "of the form '<object-hash> <pack-hash> <uri>' (got '%s')"), v); | |
2963 | if (oidmap_get(&configured_exclusions, &ex->e.oid)) | |
2964 | die(_("object already configured in another " | |
2965 | "uploadpack.blobpackfileuri (got '%s')"), v); | |
2966 | ex->pack_hash_hex = xcalloc(1, pack_end - oid_end); | |
2967 | memcpy(ex->pack_hash_hex, oid_end + 1, pack_end - oid_end - 1); | |
2968 | ex->uri = xstrdup(pack_end + 1); | |
2969 | oidmap_put(&configured_exclusions, ex); | |
2970 | } | |
ef90d6d4 | 2971 | return git_default_config(k, v, cb); |
4812a93a JK |
2972 | } |
2973 | ||
b5d97e6b | 2974 | static void read_object_list_from_stdin(void) |
c323ac7d | 2975 | { |
188960b4 | 2976 | char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2]; |
2977 | struct object_id oid; | |
2978 | const char *p; | |
b2504a0d | 2979 | |
da93d12b | 2980 | for (;;) { |
da93d12b LT |
2981 | if (!fgets(line, sizeof(line), stdin)) { |
2982 | if (feof(stdin)) | |
2983 | break; | |
2984 | if (!ferror(stdin)) | |
1a07e59c | 2985 | die("BUG: fgets returned NULL, not EOF, not error!"); |
687dd75c | 2986 | if (errno != EINTR) |
d824cbba | 2987 | die_errno("fgets"); |
687dd75c JH |
2988 | clearerr(stdin); |
2989 | continue; | |
da93d12b | 2990 | } |
7a979d99 | 2991 | if (line[0] == '-') { |
188960b4 | 2992 | if (get_oid_hex(line+1, &oid)) |
f616db6a | 2993 | die(_("expected edge object ID, got garbage:\n %s"), |
b5d97e6b | 2994 | line); |
188960b4 | 2995 | add_preferred_base(&oid); |
7a979d99 | 2996 | continue; |
21fcd1bd | 2997 | } |
188960b4 | 2998 | if (parse_oid_hex(line, &oid, &p)) |
f616db6a | 2999 | die(_("expected object ID, got garbage:\n %s"), line); |
b5d97e6b | 3000 | |
188960b4 | 3001 | add_preferred_base_object(p + 1); |
fd9b1bae | 3002 | add_object_entry(&oid, OBJ_NONE, p + 1, 0); |
c323ac7d | 3003 | } |
b5d97e6b JH |
3004 | } |
3005 | ||
95308d64 | 3006 | /* Remember to update object flag allocation in object.h */ |
08cdfb13 JH |
3007 | #define OBJECT_ADDED (1u<<20) |
3008 | ||
11c211fa | 3009 | static void show_commit(struct commit *commit, void *data) |
b5d97e6b | 3010 | { |
188960b4 | 3011 | add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0); |
08cdfb13 | 3012 | commit->object.flags |= OBJECT_ADDED; |
7cc8f971 VM |
3013 | |
3014 | if (write_bitmap_index) | |
3015 | index_commit_for_bitmap(commit); | |
28b8a730 JK |
3016 | |
3017 | if (use_delta_islands) | |
3018 | propagate_island_marks(commit); | |
b5d97e6b JH |
3019 | } |
3020 | ||
de1e67d0 | 3021 | static void show_object(struct object *obj, const char *name, void *data) |
b5d97e6b | 3022 | { |
8d2dfc49 | 3023 | add_preferred_base_object(name); |
188960b4 | 3024 | add_object_entry(&obj->oid, obj->type, name, 0); |
8d2dfc49 | 3025 | obj->flags |= OBJECT_ADDED; |
28b8a730 JK |
3026 | |
3027 | if (use_delta_islands) { | |
3028 | const char *p; | |
39490536 | 3029 | unsigned depth; |
28b8a730 JK |
3030 | struct object_entry *ent; |
3031 | ||
39490536 JK |
3032 | /* the empty string is a root tree, which is depth 0 */ |
3033 | depth = *name ? 1 : 0; | |
28b8a730 JK |
3034 | for (p = strchr(name, '/'); p; p = strchr(p + 1, '/')) |
3035 | depth++; | |
3036 | ||
3a37876b | 3037 | ent = packlist_find(&to_pack, &obj->oid); |
108f5303 CC |
3038 | if (ent && depth > oe_tree_depth(&to_pack, ent)) |
3039 | oe_set_tree_depth(&to_pack, ent, depth); | |
28b8a730 | 3040 | } |
b5d97e6b JH |
3041 | } |
3042 | ||
9535ce73 JH |
3043 | static void show_object__ma_allow_any(struct object *obj, const char *name, void *data) |
3044 | { | |
3045 | assert(arg_missing_action == MA_ALLOW_ANY); | |
3046 | ||
3047 | /* | |
3048 | * Quietly ignore ALL missing objects. This avoids problems with | |
3049 | * staging them now and getting an odd error later. | |
3050 | */ | |
ee47243d | 3051 | if (!has_object(the_repository, &obj->oid, 0)) |
9535ce73 JH |
3052 | return; |
3053 | ||
3054 | show_object(obj, name, data); | |
3055 | } | |
3056 | ||
0c16cd49 JT |
3057 | static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data) |
3058 | { | |
3059 | assert(arg_missing_action == MA_ALLOW_PROMISOR); | |
3060 | ||
3061 | /* | |
3062 | * Quietly ignore EXPECTED missing objects. This avoids problems with | |
3063 | * staging them now and getting an odd error later. | |
3064 | */ | |
ee47243d | 3065 | if (!has_object(the_repository, &obj->oid, 0) && is_promisor_object(&obj->oid)) |
0c16cd49 JT |
3066 | return; |
3067 | ||
3068 | show_object(obj, name, data); | |
3069 | } | |
3070 | ||
9535ce73 JH |
3071 | static int option_parse_missing_action(const struct option *opt, |
3072 | const char *arg, int unset) | |
3073 | { | |
3074 | assert(arg); | |
3075 | assert(!unset); | |
3076 | ||
3077 | if (!strcmp(arg, "error")) { | |
3078 | arg_missing_action = MA_ERROR; | |
3079 | fn_show_object = show_object; | |
3080 | return 0; | |
3081 | } | |
3082 | ||
3083 | if (!strcmp(arg, "allow-any")) { | |
3084 | arg_missing_action = MA_ALLOW_ANY; | |
0c16cd49 | 3085 | fetch_if_missing = 0; |
9535ce73 JH |
3086 | fn_show_object = show_object__ma_allow_any; |
3087 | return 0; | |
3088 | } | |
3089 | ||
0c16cd49 JT |
3090 | if (!strcmp(arg, "allow-promisor")) { |
3091 | arg_missing_action = MA_ALLOW_PROMISOR; | |
3092 | fetch_if_missing = 0; | |
3093 | fn_show_object = show_object__ma_allow_promisor; | |
3094 | return 0; | |
3095 | } | |
3096 | ||
9535ce73 JH |
3097 | die(_("invalid value for --missing")); |
3098 | return 0; | |
3099 | } | |
3100 | ||
8d1d8f83 JH |
3101 | static void show_edge(struct commit *commit) |
3102 | { | |
188960b4 | 3103 | add_preferred_base(&commit->object.oid); |
8d1d8f83 JH |
3104 | } |
3105 | ||
08cdfb13 JH |
3106 | struct in_pack_object { |
3107 | off_t offset; | |
3108 | struct object *object; | |
3109 | }; | |
3110 | ||
3111 | struct in_pack { | |
071bcaab RJ |
3112 | unsigned int alloc; |
3113 | unsigned int nr; | |
08cdfb13 JH |
3114 | struct in_pack_object *array; |
3115 | }; | |
3116 | ||
3117 | static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack) | |
3118 | { | |
ed1c9977 | 3119 | in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p); |
08cdfb13 JH |
3120 | in_pack->array[in_pack->nr].object = object; |
3121 | in_pack->nr++; | |
3122 | } | |
3123 | ||
3124 | /* | |
3125 | * Compare the objects in the offset order, in order to emulate the | |
f18d244a | 3126 | * "git rev-list --objects" output that produced the pack originally. |
08cdfb13 JH |
3127 | */ |
3128 | static int ofscmp(const void *a_, const void *b_) | |
3129 | { | |
3130 | struct in_pack_object *a = (struct in_pack_object *)a_; | |
3131 | struct in_pack_object *b = (struct in_pack_object *)b_; | |
3132 | ||
3133 | if (a->offset < b->offset) | |
3134 | return -1; | |
3135 | else if (a->offset > b->offset) | |
3136 | return 1; | |
3137 | else | |
f2fd0760 | 3138 | return oidcmp(&a->object->oid, &b->object->oid); |
08cdfb13 JH |
3139 | } |
3140 | ||
7a2a7216 | 3141 | static void add_objects_in_unpacked_packs(void) |
08cdfb13 JH |
3142 | { |
3143 | struct packed_git *p; | |
3144 | struct in_pack in_pack; | |
3145 | uint32_t i; | |
3146 | ||
3147 | memset(&in_pack, 0, sizeof(in_pack)); | |
3148 | ||
454ea2e4 | 3149 | for (p = get_all_packs(the_repository); p; p = p->next) { |
188960b4 | 3150 | struct object_id oid; |
08cdfb13 JH |
3151 | struct object *o; |
3152 | ||
ed7e5fc3 | 3153 | if (!p->pack_local || p->pack_keep || p->pack_keep_in_core) |
08cdfb13 JH |
3154 | continue; |
3155 | if (open_pack_index(p)) | |
f616db6a | 3156 | die(_("cannot open pack index")); |
08cdfb13 JH |
3157 | |
3158 | ALLOC_GROW(in_pack.array, | |
3159 | in_pack.nr + p->num_objects, | |
3160 | in_pack.alloc); | |
3161 | ||
3162 | for (i = 0; i < p->num_objects; i++) { | |
0763671b | 3163 | nth_packed_object_id(&oid, p, i); |
0ebbcf70 | 3164 | o = lookup_unknown_object(&oid); |
08cdfb13 JH |
3165 | if (!(o->flags & OBJECT_ADDED)) |
3166 | mark_in_pack_object(o, p, &in_pack); | |
3167 | o->flags |= OBJECT_ADDED; | |
3168 | } | |
3169 | } | |
3170 | ||
3171 | if (in_pack.nr) { | |
9ed0d8d6 | 3172 | QSORT(in_pack.array, in_pack.nr, ofscmp); |
08cdfb13 JH |
3173 | for (i = 0; i < in_pack.nr; i++) { |
3174 | struct object *o = in_pack.array[i].object; | |
188960b4 | 3175 | add_object_entry(&o->oid, o->type, "", 0); |
08cdfb13 JH |
3176 | } |
3177 | } | |
3178 | free(in_pack.array); | |
3179 | } | |
3180 | ||
76c1d9a0 | 3181 | static int add_loose_object(const struct object_id *oid, const char *path, |
e26a8c47 JK |
3182 | void *data) |
3183 | { | |
0df8e965 | 3184 | enum object_type type = oid_object_info(the_repository, oid, NULL); |
e26a8c47 JK |
3185 | |
3186 | if (type < 0) { | |
f616db6a | 3187 | warning(_("loose object at %s could not be examined"), path); |
e26a8c47 JK |
3188 | return 0; |
3189 | } | |
3190 | ||
188960b4 | 3191 | add_object_entry(oid, type, "", 0); |
e26a8c47 JK |
3192 | return 0; |
3193 | } | |
3194 | ||
3195 | /* | |
3196 | * We actually don't even have to worry about reachability here. | |
3197 | * add_object_entry will weed out duplicates, so we just add every | |
3198 | * loose object we find. | |
3199 | */ | |
3200 | static void add_unreachable_loose_objects(void) | |
3201 | { | |
3202 | for_each_loose_file_in_objdir(get_object_directory(), | |
3203 | add_loose_object, | |
3204 | NULL, NULL, NULL); | |
3205 | } | |
3206 | ||
188960b4 | 3207 | static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid) |
094085e3 BC |
3208 | { |
3209 | static struct packed_git *last_found = (void *)1; | |
3210 | struct packed_git *p; | |
3211 | ||
a80d72db | 3212 | p = (last_found != (void *)1) ? last_found : |
454ea2e4 | 3213 | get_all_packs(the_repository); |
094085e3 BC |
3214 | |
3215 | while (p) { | |
ed7e5fc3 NTND |
3216 | if ((!p->pack_local || p->pack_keep || |
3217 | p->pack_keep_in_core) && | |
188960b4 | 3218 | find_pack_entry_one(oid->hash, p)) { |
094085e3 BC |
3219 | last_found = p; |
3220 | return 1; | |
3221 | } | |
3222 | if (p == last_found) | |
454ea2e4 | 3223 | p = get_all_packs(the_repository); |
094085e3 BC |
3224 | else |
3225 | p = p->next; | |
3226 | if (p == last_found) | |
3227 | p = p->next; | |
3228 | } | |
3229 | return 0; | |
3230 | } | |
3231 | ||
abcb8655 JK |
3232 | /* |
3233 | * Store a list of sha1s that are should not be discarded | |
3234 | * because they are either written too recently, or are | |
3235 | * reachable from another object that was. | |
3236 | * | |
3237 | * This is filled by get_object_list. | |
3238 | */ | |
910650d2 | 3239 | static struct oid_array recent_objects; |
abcb8655 | 3240 | |
4ce3621a | 3241 | static int loosened_object_can_be_discarded(const struct object_id *oid, |
dddbad72 | 3242 | timestamp_t mtime) |
d0d46abc JK |
3243 | { |
3244 | if (!unpack_unreachable_expiration) | |
3245 | return 0; | |
3246 | if (mtime > unpack_unreachable_expiration) | |
3247 | return 0; | |
910650d2 | 3248 | if (oid_array_lookup(&recent_objects, oid) >= 0) |
abcb8655 | 3249 | return 0; |
d0d46abc JK |
3250 | return 1; |
3251 | } | |
3252 | ||
7a2a7216 | 3253 | static void loosen_unused_packed_objects(void) |
ca11b212 NP |
3254 | { |
3255 | struct packed_git *p; | |
3256 | uint32_t i; | |
4ce3621a | 3257 | struct object_id oid; |
ca11b212 | 3258 | |
454ea2e4 | 3259 | for (p = get_all_packs(the_repository); p; p = p->next) { |
ed7e5fc3 | 3260 | if (!p->pack_local || p->pack_keep || p->pack_keep_in_core) |
ca11b212 NP |
3261 | continue; |
3262 | ||
3263 | if (open_pack_index(p)) | |
f616db6a | 3264 | die(_("cannot open pack index")); |
ca11b212 NP |
3265 | |
3266 | for (i = 0; i < p->num_objects; i++) { | |
0763671b | 3267 | nth_packed_object_id(&oid, p, i); |
3a37876b | 3268 | if (!packlist_find(&to_pack, &oid) && |
188960b4 | 3269 | !has_sha1_pack_kept_or_nonlocal(&oid) && |
4ce3621a | 3270 | !loosened_object_can_be_discarded(&oid, p->mtime)) |
4bdb70a4 | 3271 | if (force_object_loose(&oid, p->mtime)) |
f616db6a | 3272 | die(_("unable to force loose object")); |
ca11b212 NP |
3273 | } |
3274 | } | |
3275 | } | |
3276 | ||
69e4b342 | 3277 | /* |
645c432d KS |
3278 | * This tracks any options which pack-reuse code expects to be on, or which a |
3279 | * reader of the pack might not understand, and which would therefore prevent | |
3280 | * blind reuse of what we have on disk. | |
69e4b342 JK |
3281 | */ |
3282 | static int pack_options_allow_reuse(void) | |
3283 | { | |
e704fc79 JK |
3284 | return allow_pack_reuse && |
3285 | pack_to_stdout && | |
ed7e5fc3 NTND |
3286 | !ignore_packed_keep_on_disk && |
3287 | !ignore_packed_keep_in_core && | |
9df4a607 JK |
3288 | (!local || !have_non_local_packs) && |
3289 | !incremental; | |
69e4b342 JK |
3290 | } |
3291 | ||
6b8fda2d VM |
3292 | static int get_object_list_from_bitmap(struct rev_info *revs) |
3293 | { | |
3ab3185f | 3294 | if (!(bitmap_git = prepare_bitmap_walk(revs, &filter_options))) |
6b8fda2d VM |
3295 | return -1; |
3296 | ||
69e4b342 JK |
3297 | if (pack_options_allow_reuse() && |
3298 | !reuse_partial_packfile_from_bitmap( | |
3ae5fa07 | 3299 | bitmap_git, |
6b8fda2d VM |
3300 | &reuse_packfile, |
3301 | &reuse_packfile_objects, | |
bb514de3 | 3302 | &reuse_packfile_bitmap)) { |
6b8fda2d VM |
3303 | assert(reuse_packfile_objects); |
3304 | nr_result += reuse_packfile_objects; | |
78d2214e | 3305 | display_progress(progress_state, nr_result); |
6b8fda2d VM |
3306 | } |
3307 | ||
4eb707eb JK |
3308 | traverse_bitmap_commit_list(bitmap_git, revs, |
3309 | &add_object_entry_from_bitmap); | |
6b8fda2d VM |
3310 | return 0; |
3311 | } | |
3312 | ||
abcb8655 | 3313 | static void record_recent_object(struct object *obj, |
de1e67d0 | 3314 | const char *name, |
abcb8655 JK |
3315 | void *data) |
3316 | { | |
910650d2 | 3317 | oid_array_append(&recent_objects, &obj->oid); |
abcb8655 JK |
3318 | } |
3319 | ||
3320 | static void record_recent_commit(struct commit *commit, void *data) | |
3321 | { | |
910650d2 | 3322 | oid_array_append(&recent_objects, &commit->object.oid); |
abcb8655 JK |
3323 | } |
3324 | ||
8d1d8f83 | 3325 | static void get_object_list(int ac, const char **av) |
b5d97e6b JH |
3326 | { |
3327 | struct rev_info revs; | |
bbcde41a MD |
3328 | struct setup_revision_opt s_r_opt = { |
3329 | .allow_exclude_promisor_objects = 1, | |
3330 | }; | |
b5d97e6b | 3331 | char line[1000]; |
b5d97e6b | 3332 | int flags = 0; |
a4544b31 | 3333 | int save_warning; |
b5d97e6b | 3334 | |
2abf3503 | 3335 | repo_init_revisions(the_repository, &revs, NULL); |
b5d97e6b | 3336 | save_commit_buffer = 0; |
bbcde41a | 3337 | setup_revisions(ac, av, &revs, &s_r_opt); |
b5d97e6b | 3338 | |
b790e0f6 | 3339 | /* make sure shallows are read */ |
c8813487 | 3340 | is_repository_shallow(the_repository); |
b790e0f6 | 3341 | |
a4544b31 DS |
3342 | save_warning = warn_on_object_refname_ambiguity; |
3343 | warn_on_object_refname_ambiguity = 0; | |
3344 | ||
b5d97e6b JH |
3345 | while (fgets(line, sizeof(line), stdin) != NULL) { |
3346 | int len = strlen(line); | |
872c930d | 3347 | if (len && line[len - 1] == '\n') |
b5d97e6b JH |
3348 | line[--len] = 0; |
3349 | if (!len) | |
3350 | break; | |
3351 | if (*line == '-') { | |
3352 | if (!strcmp(line, "--not")) { | |
3353 | flags ^= UNINTERESTING; | |
7cc8f971 | 3354 | write_bitmap_index = 0; |
b5d97e6b JH |
3355 | continue; |
3356 | } | |
b790e0f6 | 3357 | if (starts_with(line, "--shallow ")) { |
e92b848c | 3358 | struct object_id oid; |
3359 | if (get_oid_hex(line + 10, &oid)) | |
4279000d | 3360 | die("not an object name '%s'", line + 10); |
19143f13 | 3361 | register_shallow(the_repository, &oid); |
f7f91086 | 3362 | use_bitmap_index = 0; |
b790e0f6 NTND |
3363 | continue; |
3364 | } | |
f616db6a | 3365 | die(_("not a rev '%s'"), line); |
b5d97e6b | 3366 | } |
8e676e8b | 3367 | if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME)) |
f616db6a | 3368 | die(_("bad revision '%s'"), line); |
b5d97e6b JH |
3369 | } |
3370 | ||
a4544b31 DS |
3371 | warn_on_object_refname_ambiguity = save_warning; |
3372 | ||
6b8fda2d VM |
3373 | if (use_bitmap_index && !get_object_list_from_bitmap(&revs)) |
3374 | return; | |
3375 | ||
28b8a730 | 3376 | if (use_delta_islands) |
bdbdf42f | 3377 | load_delta_islands(the_repository, progress); |
28b8a730 | 3378 | |
3d51e1b5 | 3379 | if (prepare_revision_walk(&revs)) |
f616db6a | 3380 | die(_("revision walk setup failed")); |
4f6d26b1 | 3381 | mark_edges_uninteresting(&revs, show_edge, sparse); |
9535ce73 JH |
3382 | |
3383 | if (!fn_show_object) | |
3384 | fn_show_object = show_object; | |
3385 | traverse_commit_list_filtered(&filter_options, &revs, | |
3386 | show_commit, fn_show_object, NULL, | |
3387 | NULL); | |
08cdfb13 | 3388 | |
abcb8655 JK |
3389 | if (unpack_unreachable_expiration) { |
3390 | revs.ignore_missing_links = 1; | |
3391 | if (add_unseen_recent_objects_to_traversal(&revs, | |
3392 | unpack_unreachable_expiration)) | |
f616db6a | 3393 | die(_("unable to add recent objects")); |
abcb8655 | 3394 | if (prepare_revision_walk(&revs)) |
f616db6a | 3395 | die(_("revision walk setup failed")); |
abcb8655 JK |
3396 | traverse_commit_list(&revs, record_recent_commit, |
3397 | record_recent_object, NULL); | |
3398 | } | |
3399 | ||
08cdfb13 | 3400 | if (keep_unreachable) |
7a2a7216 | 3401 | add_objects_in_unpacked_packs(); |
e26a8c47 JK |
3402 | if (pack_loose_unreachable) |
3403 | add_unreachable_loose_objects(); | |
ca11b212 | 3404 | if (unpack_unreachable) |
7a2a7216 | 3405 | loosen_unused_packed_objects(); |
abcb8655 | 3406 | |
910650d2 | 3407 | oid_array_clear(&recent_objects); |
b5d97e6b JH |
3408 | } |
3409 | ||
ed7e5fc3 NTND |
3410 | static void add_extra_kept_packs(const struct string_list *names) |
3411 | { | |
3412 | struct packed_git *p; | |
3413 | ||
3414 | if (!names->nr) | |
3415 | return; | |
3416 | ||
454ea2e4 | 3417 | for (p = get_all_packs(the_repository); p; p = p->next) { |
ed7e5fc3 NTND |
3418 | const char *name = basename(p->pack_name); |
3419 | int i; | |
3420 | ||
3421 | if (!p->pack_local) | |
3422 | continue; | |
3423 | ||
3424 | for (i = 0; i < names->nr; i++) | |
3425 | if (!fspathcmp(name, names->items[i].string)) | |
3426 | break; | |
3427 | ||
3428 | if (i < names->nr) { | |
3429 | p->pack_keep_in_core = 1; | |
3430 | ignore_packed_keep_in_core = 1; | |
3431 | continue; | |
3432 | } | |
3433 | } | |
3434 | } | |
3435 | ||
99fb6e04 NTND |
3436 | static int option_parse_index_version(const struct option *opt, |
3437 | const char *arg, int unset) | |
3438 | { | |
3439 | char *c; | |
3440 | const char *val = arg; | |
517fe807 JK |
3441 | |
3442 | BUG_ON_OPT_NEG(unset); | |
3443 | ||
99fb6e04 NTND |
3444 | pack_idx_opts.version = strtoul(val, &c, 10); |
3445 | if (pack_idx_opts.version > 2) | |
3446 | die(_("unsupported index version %s"), val); | |
3447 | if (*c == ',' && c[1]) | |
3448 | pack_idx_opts.off32_limit = strtoul(c+1, &c, 0); | |
3449 | if (*c || pack_idx_opts.off32_limit & 0x80000000) | |
3450 | die(_("bad index version '%s'"), val); | |
3451 | return 0; | |
3452 | } | |
3453 | ||
7e52f566 JK |
3454 | static int option_parse_unpack_unreachable(const struct option *opt, |
3455 | const char *arg, int unset) | |
3456 | { | |
3457 | if (unset) { | |
3458 | unpack_unreachable = 0; | |
3459 | unpack_unreachable_expiration = 0; | |
3460 | } | |
3461 | else { | |
3462 | unpack_unreachable = 1; | |
3463 | if (arg) | |
3464 | unpack_unreachable_expiration = approxidate(arg); | |
3465 | } | |
3466 | return 0; | |
3467 | } | |
3468 | ||
b5d97e6b JH |
3469 | int cmd_pack_objects(int argc, const char **argv, const char *prefix) |
3470 | { | |
b5d97e6b | 3471 | int use_internal_rev_list = 0; |
2dacf26d | 3472 | int shallow = 0; |
4f366275 | 3473 | int all_progress_implied = 0; |
22f9b7f3 | 3474 | struct strvec rp = STRVEC_INIT; |
99fb6e04 | 3475 | int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0; |
c90f9e13 | 3476 | int rev_list_index = 0; |
ed7e5fc3 | 3477 | struct string_list keep_pack_list = STRING_LIST_INIT_NODUP; |
99fb6e04 NTND |
3478 | struct option pack_objects_options[] = { |
3479 | OPT_SET_INT('q', "quiet", &progress, | |
4c688120 | 3480 | N_("do not show progress meter"), 0), |
99fb6e04 | 3481 | OPT_SET_INT(0, "progress", &progress, |
4c688120 | 3482 | N_("show progress meter"), 1), |
99fb6e04 | 3483 | OPT_SET_INT(0, "all-progress", &progress, |
4c688120 | 3484 | N_("show progress meter during object writing phase"), 2), |
99fb6e04 NTND |
3485 | OPT_BOOL(0, "all-progress-implied", |
3486 | &all_progress_implied, | |
4c688120 | 3487 | N_("similar to --all-progress when progress meter is shown")), |
203c8533 | 3488 | OPT_CALLBACK_F(0, "index-version", NULL, N_("<version>[,<offset>]"), |
4c688120 | 3489 | N_("write the pack index file in the specified idx format version"), |
203c8533 | 3490 | PARSE_OPT_NONEG, option_parse_index_version), |
2a514ed8 CB |
3491 | OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit, |
3492 | N_("maximum size of each output pack file")), | |
99fb6e04 | 3493 | OPT_BOOL(0, "local", &local, |
4c688120 | 3494 | N_("ignore borrowed objects from alternate object store")), |
99fb6e04 | 3495 | OPT_BOOL(0, "incremental", &incremental, |
4c688120 | 3496 | N_("ignore packed objects")), |
99fb6e04 | 3497 | OPT_INTEGER(0, "window", &window, |
4c688120 | 3498 | N_("limit pack window by objects")), |
2a514ed8 CB |
3499 | OPT_MAGNITUDE(0, "window-memory", &window_memory_limit, |
3500 | N_("limit pack window by memory in addition to object limit")), | |
99fb6e04 | 3501 | OPT_INTEGER(0, "depth", &depth, |
4c688120 | 3502 | N_("maximum length of delta chain allowed in the resulting pack")), |
99fb6e04 | 3503 | OPT_BOOL(0, "reuse-delta", &reuse_delta, |
4c688120 | 3504 | N_("reuse existing deltas")), |
99fb6e04 | 3505 | OPT_BOOL(0, "reuse-object", &reuse_object, |
4c688120 | 3506 | N_("reuse existing objects")), |
99fb6e04 | 3507 | OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta, |
4c688120 | 3508 | N_("use OFS_DELTA objects")), |
99fb6e04 | 3509 | OPT_INTEGER(0, "threads", &delta_search_threads, |
4c688120 | 3510 | N_("use threads when searching for best delta matches")), |
99fb6e04 | 3511 | OPT_BOOL(0, "non-empty", &non_empty, |
4c688120 | 3512 | N_("do not create an empty pack output")), |
99fb6e04 | 3513 | OPT_BOOL(0, "revs", &use_internal_rev_list, |
4c688120 | 3514 | N_("read revision arguments from standard input")), |
3e4a67b4 NTND |
3515 | OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked, |
3516 | N_("limit the objects to those that are not yet packed"), | |
3517 | 1, PARSE_OPT_NONEG), | |
3518 | OPT_SET_INT_F(0, "all", &rev_list_all, | |
3519 | N_("include objects reachable from any reference"), | |
3520 | 1, PARSE_OPT_NONEG), | |
3521 | OPT_SET_INT_F(0, "reflog", &rev_list_reflog, | |
3522 | N_("include objects referred by reflog entries"), | |
3523 | 1, PARSE_OPT_NONEG), | |
3524 | OPT_SET_INT_F(0, "indexed-objects", &rev_list_index, | |
3525 | N_("include objects referred to by the index"), | |
3526 | 1, PARSE_OPT_NONEG), | |
99fb6e04 | 3527 | OPT_BOOL(0, "stdout", &pack_to_stdout, |
4c688120 | 3528 | N_("output pack to stdout")), |
99fb6e04 | 3529 | OPT_BOOL(0, "include-tag", &include_tag, |
4c688120 | 3530 | N_("include tag objects that refer to objects to be packed")), |
99fb6e04 | 3531 | OPT_BOOL(0, "keep-unreachable", &keep_unreachable, |
4c688120 | 3532 | N_("keep unreachable objects")), |
e26a8c47 JK |
3533 | OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable, |
3534 | N_("pack loose unreachable objects")), | |
203c8533 | 3535 | OPT_CALLBACK_F(0, "unpack-unreachable", NULL, N_("time"), |
4c688120 | 3536 | N_("unpack unreachable objects newer than <time>"), |
203c8533 | 3537 | PARSE_OPT_OPTARG, option_parse_unpack_unreachable), |
4f6d26b1 DS |
3538 | OPT_BOOL(0, "sparse", &sparse, |
3539 | N_("use the sparse reachability algorithm")), | |
99fb6e04 | 3540 | OPT_BOOL(0, "thin", &thin, |
4c688120 | 3541 | N_("create thin packs")), |
2dacf26d | 3542 | OPT_BOOL(0, "shallow", &shallow, |
3543 | N_("create packs suitable for shallow fetches")), | |
ed7e5fc3 | 3544 | OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk, |
4c688120 | 3545 | N_("ignore packs that have companion .keep file")), |
ed7e5fc3 NTND |
3546 | OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"), |
3547 | N_("ignore this pack")), | |
99fb6e04 | 3548 | OPT_INTEGER(0, "compression", &pack_compression_level, |
4c688120 | 3549 | N_("pack compression level")), |
99fb6e04 | 3550 | OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents, |
4c688120 | 3551 | N_("do not hide commits by grafts"), 0), |
6b8fda2d VM |
3552 | OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index, |
3553 | N_("use a bitmap index if available to speed up counting objects")), | |
25575015 JK |
3554 | OPT_SET_INT(0, "write-bitmap-index", &write_bitmap_index, |
3555 | N_("write a bitmap index together with the pack index"), | |
3556 | WRITE_BITMAP_TRUE), | |
3557 | OPT_SET_INT_F(0, "write-bitmap-index-quiet", | |
3558 | &write_bitmap_index, | |
3559 | N_("write a bitmap index if possible"), | |
3560 | WRITE_BITMAP_QUIET, PARSE_OPT_HIDDEN), | |
9535ce73 | 3561 | OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options), |
203c8533 | 3562 | OPT_CALLBACK_F(0, "missing", NULL, N_("action"), |
9535ce73 | 3563 | N_("handling for missing objects"), PARSE_OPT_NONEG, |
203c8533 | 3564 | option_parse_missing_action), |
0c16cd49 JT |
3565 | OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects, |
3566 | N_("do not pack objects in promisor packfiles")), | |
28b8a730 JK |
3567 | OPT_BOOL(0, "delta-islands", &use_delta_islands, |
3568 | N_("respect islands during delta compression")), | |
dd4b732d JT |
3569 | OPT_STRING_LIST(0, "uri-protocol", &uri_protocols, |
3570 | N_("protocol"), | |
3571 | N_("exclude any configured uploadpack.blobpackfileuri with this protocol")), | |
99fb6e04 NTND |
3572 | OPT_END(), |
3573 | }; | |
8d1d8f83 | 3574 | |
0c6804ab NTND |
3575 | if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS)) |
3576 | BUG("too many dfs states, increase OE_DFS_STATE_BITS"); | |
3577 | ||
6ebd1caf | 3578 | read_replace_refs = 0; |
dae556bd | 3579 | |
2d657ab9 | 3580 | sparse = git_env_bool("GIT_TEST_PACK_SPARSE", -1); |
7211b9e7 | 3581 | prepare_repo_settings(the_repository); |
2d657ab9 | 3582 | if (sparse < 0) |
7211b9e7 DS |
3583 | sparse = the_repository->settings.pack_use_sparse; |
3584 | ||
ebcfb379 | 3585 | reset_pack_idx_option(&pack_idx_opts); |
ef90d6d4 | 3586 | git_config(git_pack_config, NULL); |
b5d97e6b JH |
3587 | |
3588 | progress = isatty(2); | |
99fb6e04 NTND |
3589 | argc = parse_options(argc, argv, prefix, pack_objects_options, |
3590 | pack_usage, 0); | |
b5d97e6b | 3591 | |
99fb6e04 NTND |
3592 | if (argc) { |
3593 | base_name = argv[0]; | |
3594 | argc--; | |
b5d97e6b | 3595 | } |
99fb6e04 NTND |
3596 | if (pack_to_stdout != !base_name || argc) |
3597 | usage_with_options(pack_usage, pack_objects_options); | |
b5d97e6b | 3598 | |
b5c0cbd8 NTND |
3599 | if (depth >= (1 << OE_DEPTH_BITS)) { |
3600 | warning(_("delta chain depth %d is too deep, forcing %d"), | |
3601 | depth, (1 << OE_DEPTH_BITS) - 1); | |
3602 | depth = (1 << OE_DEPTH_BITS) - 1; | |
3603 | } | |
0cb3c142 NTND |
3604 | if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) { |
3605 | warning(_("pack.deltaCacheLimit is too high, forcing %d"), | |
3606 | (1U << OE_Z_DELTA_BITS) - 1); | |
3607 | cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1; | |
3608 | } | |
b5c0cbd8 | 3609 | |
22f9b7f3 | 3610 | strvec_push(&rp, "pack-objects"); |
99fb6e04 NTND |
3611 | if (thin) { |
3612 | use_internal_rev_list = 1; | |
22f9b7f3 | 3613 | strvec_push(&rp, shallow |
2dacf26d | 3614 | ? "--objects-edge-aggressive" |
3615 | : "--objects-edge"); | |
99fb6e04 | 3616 | } else |
22f9b7f3 | 3617 | strvec_push(&rp, "--objects"); |
b5d97e6b | 3618 | |
99fb6e04 NTND |
3619 | if (rev_list_all) { |
3620 | use_internal_rev_list = 1; | |
22f9b7f3 | 3621 | strvec_push(&rp, "--all"); |
99fb6e04 NTND |
3622 | } |
3623 | if (rev_list_reflog) { | |
3624 | use_internal_rev_list = 1; | |
22f9b7f3 | 3625 | strvec_push(&rp, "--reflog"); |
99fb6e04 | 3626 | } |
c90f9e13 JK |
3627 | if (rev_list_index) { |
3628 | use_internal_rev_list = 1; | |
22f9b7f3 | 3629 | strvec_push(&rp, "--indexed-objects"); |
99fb6e04 NTND |
3630 | } |
3631 | if (rev_list_unpacked) { | |
3632 | use_internal_rev_list = 1; | |
22f9b7f3 | 3633 | strvec_push(&rp, "--unpacked"); |
99fb6e04 | 3634 | } |
b5d97e6b | 3635 | |
0c16cd49 JT |
3636 | if (exclude_promisor_objects) { |
3637 | use_internal_rev_list = 1; | |
3638 | fetch_if_missing = 0; | |
22f9b7f3 | 3639 | strvec_push(&rp, "--exclude-promisor-objects"); |
0c16cd49 | 3640 | } |
58bd77b6 NTND |
3641 | if (unpack_unreachable || keep_unreachable || pack_loose_unreachable) |
3642 | use_internal_rev_list = 1; | |
0c16cd49 | 3643 | |
99fb6e04 NTND |
3644 | if (!reuse_object) |
3645 | reuse_delta = 0; | |
3646 | if (pack_compression_level == -1) | |
3647 | pack_compression_level = Z_DEFAULT_COMPRESSION; | |
3648 | else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION) | |
f616db6a | 3649 | die(_("bad pack compression level %d"), pack_compression_level); |
0c45d258 JH |
3650 | |
3651 | if (!delta_search_threads) /* --threads=0 means autodetect */ | |
3652 | delta_search_threads = online_cpus(); | |
3653 | ||
9c897c5c | 3654 | if (!HAVE_THREADS && delta_search_threads != 1) |
f616db6a | 3655 | warning(_("no threads support, ignoring --threads")); |
2b84b5a8 JS |
3656 | if (!pack_to_stdout && !pack_size_limit) |
3657 | pack_size_limit = pack_size_limit_cfg; | |
01c12a23 | 3658 | if (pack_to_stdout && pack_size_limit) |
f616db6a | 3659 | die(_("--max-pack-size cannot be used to build a pack for transfer")); |
07cf0f24 | 3660 | if (pack_size_limit && pack_size_limit < 1024*1024) { |
f616db6a | 3661 | warning(_("minimum pack size limit is 1 MiB")); |
07cf0f24 NP |
3662 | pack_size_limit = 1024*1024; |
3663 | } | |
01c12a23 | 3664 | |
8d1d8f83 | 3665 | if (!pack_to_stdout && thin) |
f616db6a | 3666 | die(_("--thin cannot be used to build an indexable pack")); |
b5d97e6b | 3667 | |
ca11b212 | 3668 | if (keep_unreachable && unpack_unreachable) |
f616db6a | 3669 | die(_("--keep-unreachable and --unpack-unreachable are incompatible")); |
b1e757f3 JK |
3670 | if (!rev_list_all || !rev_list_reflog || !rev_list_index) |
3671 | unpack_unreachable_expiration = 0; | |
ca11b212 | 3672 | |
9535ce73 JH |
3673 | if (filter_options.choice) { |
3674 | if (!pack_to_stdout) | |
f616db6a | 3675 | die(_("cannot use --filter without --stdout")); |
9535ce73 JH |
3676 | } |
3677 | ||
645c432d KS |
3678 | /* |
3679 | * "soft" reasons not to use bitmaps - for on-disk repack by default we want | |
3680 | * | |
3681 | * - to produce good pack (with bitmap index not-yet-packed objects are | |
3682 | * packed in suboptimal order). | |
3683 | * | |
3684 | * - to use more robust pack-generation codepath (avoiding possible | |
3685 | * bugs in bitmap code and possible bitmap index corruption). | |
3686 | */ | |
3687 | if (!pack_to_stdout) | |
3688 | use_bitmap_index_default = 0; | |
3689 | ||
3690 | if (use_bitmap_index < 0) | |
3691 | use_bitmap_index = use_bitmap_index_default; | |
3692 | ||
3693 | /* "hard" reasons not to use bitmaps; these just won't work at all */ | |
c8813487 | 3694 | if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository)) |
6b8fda2d VM |
3695 | use_bitmap_index = 0; |
3696 | ||
7cc8f971 VM |
3697 | if (pack_to_stdout || !rev_list_all) |
3698 | write_bitmap_index = 0; | |
3699 | ||
28b8a730 | 3700 | if (use_delta_islands) |
22f9b7f3 | 3701 | strvec_push(&rp, "--topo-order"); |
28b8a730 | 3702 | |
4f366275 NP |
3703 | if (progress && all_progress_implied) |
3704 | progress = 2; | |
3705 | ||
ed7e5fc3 NTND |
3706 | add_extra_kept_packs(&keep_pack_list); |
3707 | if (ignore_packed_keep_on_disk) { | |
56dfeb62 | 3708 | struct packed_git *p; |
454ea2e4 | 3709 | for (p = get_all_packs(the_repository); p; p = p->next) |
56dfeb62 JK |
3710 | if (p->pack_local && p->pack_keep) |
3711 | break; | |
3712 | if (!p) /* no keep-able packs found */ | |
ed7e5fc3 | 3713 | ignore_packed_keep_on_disk = 0; |
56dfeb62 JK |
3714 | } |
3715 | if (local) { | |
3716 | /* | |
ed7e5fc3 NTND |
3717 | * unlike ignore_packed_keep_on_disk above, we do not |
3718 | * want to unset "local" based on looking at packs, as | |
3719 | * it also covers non-local objects | |
56dfeb62 JK |
3720 | */ |
3721 | struct packed_git *p; | |
454ea2e4 | 3722 | for (p = get_all_packs(the_repository); p; p = p->next) { |
56dfeb62 JK |
3723 | if (!p->pack_local) { |
3724 | have_non_local_packs = 1; | |
3725 | break; | |
3726 | } | |
3727 | } | |
3728 | } | |
b5d97e6b | 3729 | |
ae417807 DS |
3730 | trace2_region_enter("pack-objects", "enumerate-objects", |
3731 | the_repository); | |
7c141127 | 3732 | prepare_packing_data(the_repository, &to_pack); |
43fa44fa | 3733 | |
13aaf148 | 3734 | if (progress) |
5af05043 | 3735 | progress_state = start_progress(_("Enumerating objects"), 0); |
b5d97e6b JH |
3736 | if (!use_internal_rev_list) |
3737 | read_object_list_from_stdin(); | |
8d1d8f83 | 3738 | else { |
d70a9eb6 | 3739 | get_object_list(rp.nr, rp.v); |
22f9b7f3 | 3740 | strvec_clear(&rp); |
8d1d8f83 | 3741 | } |
0ef95f72 | 3742 | cleanup_preferred_base(); |
d155254c MH |
3743 | if (include_tag && nr_result) |
3744 | for_each_ref(add_ref_tag, NULL); | |
4d4fcc54 | 3745 | stop_progress(&progress_state); |
ae417807 DS |
3746 | trace2_region_leave("pack-objects", "enumerate-objects", |
3747 | the_repository); | |
96a02f8f | 3748 | |
f0b0af1b | 3749 | if (non_empty && !nr_result) |
1c4a2912 | 3750 | return 0; |
ae417807 DS |
3751 | if (nr_result) { |
3752 | trace2_region_enter("pack-objects", "prepare-pack", | |
3753 | the_repository); | |
f7ae6a93 | 3754 | prepare_pack(window, depth); |
ae417807 DS |
3755 | trace2_region_leave("pack-objects", "prepare-pack", |
3756 | the_repository); | |
3757 | } | |
3758 | ||
3759 | trace2_region_enter("pack-objects", "write-pack-file", the_repository); | |
dd4b732d | 3760 | write_excluded_by_configs(); |
d01fb92f | 3761 | write_pack_file(); |
ae417807 DS |
3762 | trace2_region_leave("pack-objects", "write-pack-file", the_repository); |
3763 | ||
ab7cd7bb | 3764 | if (progress) |
f616db6a NTND |
3765 | fprintf_ln(stderr, |
3766 | _("Total %"PRIu32" (delta %"PRIu32")," | |
bab28d9f JK |
3767 | " reused %"PRIu32" (delta %"PRIu32")," |
3768 | " pack-reused %"PRIu32), | |
3769 | written, written_delta, reused, reused_delta, | |
3770 | reuse_packfile_objects); | |
c323ac7d LT |
3771 | return 0; |
3772 | } |