]>
Commit | Line | Data |
---|---|---|
5d4a6003 | 1 | #include "builtin.h" |
c323ac7d | 2 | #include "cache.h" |
a74db82e | 3 | #include "attr.h" |
c323ac7d | 4 | #include "object.h" |
8e440259 PE |
5 | #include "blob.h" |
6 | #include "commit.h" | |
7 | #include "tag.h" | |
8 | #include "tree.h" | |
c323ac7d | 9 | #include "delta.h" |
a733cb60 | 10 | #include "pack.h" |
3449f8c4 | 11 | #include "pack-revindex.h" |
c38138cd | 12 | #include "csum-file.h" |
1b0c7174 | 13 | #include "tree-walk.h" |
b5d97e6b JH |
14 | #include "diff.h" |
15 | #include "revision.h" | |
16 | #include "list-objects.h" | |
2834bc27 | 17 | #include "pack-objects.h" |
96a02f8f | 18 | #include "progress.h" |
f0a24aa5 | 19 | #include "refs.h" |
cf2ba13a | 20 | #include "streaming.h" |
93749194 | 21 | #include "thread-utils.h" |
6b8fda2d | 22 | #include "pack-bitmap.h" |
abcb8655 JK |
23 | #include "reachable.h" |
24 | #include "sha1-array.h" | |
edfbb2aa | 25 | #include "argv-array.h" |
8ecce684 | 26 | |
99fb6e04 | 27 | static const char *pack_usage[] = { |
b8c1d275 AH |
28 | N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), |
29 | N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), | |
99fb6e04 NTND |
30 | NULL |
31 | }; | |
c323ac7d | 32 | |
3f9ac8d2 | 33 | /* |
2834bc27 VM |
34 | * Objects we are going to pack are collected in the `to_pack` structure. |
35 | * It contains an array (dynamically expanded) of the object data, and a map | |
36 | * that can resolve SHA1s to their position in the array. | |
3f9ac8d2 | 37 | */ |
2834bc27 VM |
38 | static struct packing_data to_pack; |
39 | ||
79814f42 | 40 | static struct pack_idx_entry **written_list; |
2834bc27 | 41 | static uint32_t nr_result, nr_written; |
3f9ac8d2 | 42 | |
96f1e58f | 43 | static int non_empty; |
a7de7130 | 44 | static int reuse_delta = 1, reuse_object = 1; |
e5e9714a | 45 | static int keep_unreachable, unpack_unreachable, include_tag; |
7e52f566 | 46 | static unsigned long unpack_unreachable_expiration; |
e26a8c47 | 47 | static int pack_loose_unreachable; |
96f1e58f | 48 | static int local; |
56dfeb62 | 49 | static int have_non_local_packs; |
96f1e58f | 50 | static int incremental; |
e96fb9b8 | 51 | static int ignore_packed_keep; |
be6b1914 | 52 | static int allow_ofs_delta; |
ebcfb379 | 53 | static struct pack_idx_option pack_idx_opts; |
d01fb92f | 54 | static const char *base_name; |
024701f1 | 55 | static int progress = 1; |
4812a93a | 56 | static int window = 10; |
568508e7 | 57 | static unsigned long pack_size_limit; |
618e613a | 58 | static int depth = 50; |
43cc2b42 | 59 | static int delta_search_threads; |
df6d6101 | 60 | static int pack_to_stdout; |
8d1d8f83 | 61 | static int num_preferred_base; |
dc6a0757 | 62 | static struct progress *progress_state; |
960ccca6 DH |
63 | static int pack_compression_level = Z_DEFAULT_COMPRESSION; |
64 | static int pack_compression_seen; | |
c323ac7d | 65 | |
6b8fda2d VM |
66 | static struct packed_git *reuse_packfile; |
67 | static uint32_t reuse_packfile_objects; | |
68 | static off_t reuse_packfile_offset; | |
69 | ||
70 | static int use_bitmap_index = 1; | |
7cc8f971 | 71 | static int write_bitmap_index; |
ae4f07fb | 72 | static uint16_t write_bitmap_options; |
6b8fda2d | 73 | |
074b2eea | 74 | static unsigned long delta_cache_size = 0; |
5749b0b2 | 75 | static unsigned long max_delta_cache_size = 256 * 1024 * 1024; |
e3dfddb3 | 76 | static unsigned long cache_max_small_delta_size = 1000; |
074b2eea | 77 | |
a97773ce BD |
78 | static unsigned long window_memory_limit = 0; |
79 | ||
3f9ac8d2 JH |
80 | /* |
81 | * stats | |
82 | */ | |
7cadf491 SP |
83 | static uint32_t written, written_delta; |
84 | static uint32_t reused, reused_delta; | |
3f9ac8d2 | 85 | |
7cc8f971 VM |
86 | /* |
87 | * Indexed commits | |
88 | */ | |
89 | static struct commit **indexed_commits; | |
90 | static unsigned int indexed_commits_nr; | |
91 | static unsigned int indexed_commits_alloc; | |
92 | ||
93 | static void index_commit_for_bitmap(struct commit *commit) | |
94 | { | |
95 | if (indexed_commits_nr >= indexed_commits_alloc) { | |
96 | indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; | |
2756ca43 | 97 | REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); |
7cc8f971 VM |
98 | } |
99 | ||
100 | indexed_commits[indexed_commits_nr++] = commit; | |
101 | } | |
780e6e73 | 102 | |
3613f9b4 | 103 | static void *get_delta(struct object_entry *entry) |
c323ac7d | 104 | { |
3613f9b4 NP |
105 | unsigned long size, base_size, delta_size; |
106 | void *buf, *base_buf, *delta_buf; | |
21666f1a | 107 | enum object_type type; |
c323ac7d | 108 | |
3613f9b4 NP |
109 | buf = read_sha1_file(entry->idx.sha1, &type, &size); |
110 | if (!buf) | |
111 | die("unable to read %s", sha1_to_hex(entry->idx.sha1)); | |
112 | base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size); | |
113 | if (!base_buf) | |
aa7e44bf | 114 | die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1)); |
3613f9b4 | 115 | delta_buf = diff_delta(base_buf, base_size, |
dcde55bc | 116 | buf, size, &delta_size, 0); |
3613f9b4 | 117 | if (!delta_buf || delta_size != entry->delta_size) |
a6080a0a | 118 | die("delta size changed"); |
3613f9b4 NP |
119 | free(buf); |
120 | free(base_buf); | |
c323ac7d LT |
121 | return delta_buf; |
122 | } | |
123 | ||
30ebb40a NP |
124 | static unsigned long do_compress(void **pptr, unsigned long size) |
125 | { | |
ef49a7a0 | 126 | git_zstream stream; |
30ebb40a NP |
127 | void *in, *out; |
128 | unsigned long maxsize; | |
129 | ||
55bb5c91 | 130 | git_deflate_init(&stream, pack_compression_level); |
225a6f10 | 131 | maxsize = git_deflate_bound(&stream, size); |
30ebb40a NP |
132 | |
133 | in = *pptr; | |
134 | out = xmalloc(maxsize); | |
135 | *pptr = out; | |
136 | ||
137 | stream.next_in = in; | |
138 | stream.avail_in = size; | |
139 | stream.next_out = out; | |
140 | stream.avail_out = maxsize; | |
55bb5c91 | 141 | while (git_deflate(&stream, Z_FINISH) == Z_OK) |
30ebb40a | 142 | ; /* nothing */ |
55bb5c91 | 143 | git_deflate_end(&stream); |
30ebb40a NP |
144 | |
145 | free(in); | |
146 | return stream.total_out; | |
147 | } | |
148 | ||
cf2ba13a NTND |
149 | static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f, |
150 | const unsigned char *sha1) | |
151 | { | |
152 | git_zstream stream; | |
153 | unsigned char ibuf[1024 * 16]; | |
154 | unsigned char obuf[1024 * 16]; | |
155 | unsigned long olen = 0; | |
156 | ||
cf2ba13a NTND |
157 | git_deflate_init(&stream, pack_compression_level); |
158 | ||
159 | for (;;) { | |
160 | ssize_t readlen; | |
161 | int zret = Z_OK; | |
162 | readlen = read_istream(st, ibuf, sizeof(ibuf)); | |
163 | if (readlen == -1) | |
164 | die(_("unable to read %s"), sha1_to_hex(sha1)); | |
165 | ||
166 | stream.next_in = ibuf; | |
167 | stream.avail_in = readlen; | |
168 | while ((stream.avail_in || readlen == 0) && | |
169 | (zret == Z_OK || zret == Z_BUF_ERROR)) { | |
170 | stream.next_out = obuf; | |
171 | stream.avail_out = sizeof(obuf); | |
172 | zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); | |
173 | sha1write(f, obuf, stream.next_out - obuf); | |
174 | olen += stream.next_out - obuf; | |
175 | } | |
176 | if (stream.avail_in) | |
177 | die(_("deflate error (%d)"), zret); | |
178 | if (readlen == 0) { | |
179 | if (zret != Z_STREAM_END) | |
180 | die(_("deflate error (%d)"), zret); | |
181 | break; | |
182 | } | |
183 | } | |
184 | git_deflate_end(&stream); | |
185 | return olen; | |
186 | } | |
187 | ||
780e6e73 NP |
188 | /* |
189 | * we are going to reuse the existing object data as is. make | |
190 | * sure it is not corrupt. | |
191 | */ | |
079afb18 SP |
192 | static int check_pack_inflate(struct packed_git *p, |
193 | struct pack_window **w_curs, | |
6777a59f SP |
194 | off_t offset, |
195 | off_t len, | |
079afb18 SP |
196 | unsigned long expect) |
197 | { | |
ef49a7a0 | 198 | git_zstream stream; |
079afb18 SP |
199 | unsigned char fakebuf[4096], *in; |
200 | int st; | |
201 | ||
202 | memset(&stream, 0, sizeof(stream)); | |
39c68542 | 203 | git_inflate_init(&stream); |
079afb18 SP |
204 | do { |
205 | in = use_pack(p, w_curs, offset, &stream.avail_in); | |
206 | stream.next_in = in; | |
207 | stream.next_out = fakebuf; | |
208 | stream.avail_out = sizeof(fakebuf); | |
39c68542 | 209 | st = git_inflate(&stream, Z_FINISH); |
079afb18 SP |
210 | offset += stream.next_in - in; |
211 | } while (st == Z_OK || st == Z_BUF_ERROR); | |
39c68542 | 212 | git_inflate_end(&stream); |
079afb18 SP |
213 | return (st == Z_STREAM_END && |
214 | stream.total_out == expect && | |
215 | stream.total_in == len) ? 0 : -1; | |
216 | } | |
217 | ||
218 | static void copy_pack_data(struct sha1file *f, | |
219 | struct packed_git *p, | |
220 | struct pack_window **w_curs, | |
6777a59f SP |
221 | off_t offset, |
222 | off_t len) | |
079afb18 SP |
223 | { |
224 | unsigned char *in; | |
ef49a7a0 | 225 | unsigned long avail; |
079afb18 SP |
226 | |
227 | while (len) { | |
228 | in = use_pack(p, w_curs, offset, &avail); | |
229 | if (avail > len) | |
ef49a7a0 | 230 | avail = (unsigned long)len; |
079afb18 SP |
231 | sha1write(f, in, avail); |
232 | offset += avail; | |
233 | len -= avail; | |
234 | } | |
235 | } | |
236 | ||
1b4bb16b | 237 | /* Return 0 if we will bust the pack-size limit */ |
c9018b03 NTND |
238 | static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry, |
239 | unsigned long limit, int usable_delta) | |
c323ac7d | 240 | { |
c9018b03 | 241 | unsigned long size, datalen; |
2c5ef824 | 242 | unsigned char header[10], dheader[10]; |
6777a59f | 243 | unsigned hdrlen; |
2c5ef824 | 244 | enum object_type type; |
c9018b03 | 245 | void *buf; |
cf2ba13a | 246 | struct git_istream *st = NULL; |
c9018b03 NTND |
247 | |
248 | if (!usable_delta) { | |
cf2ba13a NTND |
249 | if (entry->type == OBJ_BLOB && |
250 | entry->size > big_file_threshold && | |
251 | (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL) | |
252 | buf = NULL; | |
253 | else { | |
254 | buf = read_sha1_file(entry->idx.sha1, &type, &size); | |
255 | if (!buf) | |
256 | die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1)); | |
257 | } | |
c9018b03 NTND |
258 | /* |
259 | * make sure no cached delta data remains from a | |
260 | * previous attempt before a pack split occurred. | |
261 | */ | |
262 | free(entry->delta_data); | |
263 | entry->delta_data = NULL; | |
264 | entry->z_delta_size = 0; | |
265 | } else if (entry->delta_data) { | |
266 | size = entry->delta_size; | |
267 | buf = entry->delta_data; | |
268 | entry->delta_data = NULL; | |
269 | type = (allow_ofs_delta && entry->delta->idx.offset) ? | |
270 | OBJ_OFS_DELTA : OBJ_REF_DELTA; | |
271 | } else { | |
272 | buf = get_delta(entry); | |
273 | size = entry->delta_size; | |
274 | type = (allow_ofs_delta && entry->delta->idx.offset) ? | |
275 | OBJ_OFS_DELTA : OBJ_REF_DELTA; | |
276 | } | |
277 | ||
cf2ba13a NTND |
278 | if (st) /* large blob case, just assume we don't compress well */ |
279 | datalen = size; | |
280 | else if (entry->z_delta_size) | |
c9018b03 NTND |
281 | datalen = entry->z_delta_size; |
282 | else | |
283 | datalen = do_compress(&buf, size); | |
284 | ||
285 | /* | |
286 | * The object header is a byte of 'type' followed by zero or | |
287 | * more bytes of length. | |
288 | */ | |
289 | hdrlen = encode_in_pack_object_header(type, size, header); | |
290 | ||
291 | if (type == OBJ_OFS_DELTA) { | |
292 | /* | |
293 | * Deltas with relative base contain an additional | |
294 | * encoding of the relative offset for the delta | |
295 | * base from this object's position in the pack. | |
296 | */ | |
297 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; | |
298 | unsigned pos = sizeof(dheader) - 1; | |
299 | dheader[pos] = ofs & 127; | |
300 | while (ofs >>= 7) | |
301 | dheader[--pos] = 128 | (--ofs & 127); | |
302 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) { | |
cf2ba13a NTND |
303 | if (st) |
304 | close_istream(st); | |
c9018b03 NTND |
305 | free(buf); |
306 | return 0; | |
307 | } | |
308 | sha1write(f, header, hdrlen); | |
309 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
310 | hdrlen += sizeof(dheader) - pos; | |
311 | } else if (type == OBJ_REF_DELTA) { | |
312 | /* | |
313 | * Deltas with a base reference contain | |
314 | * an additional 20 bytes for the base sha1. | |
315 | */ | |
316 | if (limit && hdrlen + 20 + datalen + 20 >= limit) { | |
cf2ba13a NTND |
317 | if (st) |
318 | close_istream(st); | |
c9018b03 NTND |
319 | free(buf); |
320 | return 0; | |
321 | } | |
322 | sha1write(f, header, hdrlen); | |
323 | sha1write(f, entry->delta->idx.sha1, 20); | |
324 | hdrlen += 20; | |
325 | } else { | |
326 | if (limit && hdrlen + datalen + 20 >= limit) { | |
cf2ba13a NTND |
327 | if (st) |
328 | close_istream(st); | |
c9018b03 NTND |
329 | free(buf); |
330 | return 0; | |
331 | } | |
332 | sha1write(f, header, hdrlen); | |
333 | } | |
cf2ba13a NTND |
334 | if (st) { |
335 | datalen = write_large_blob_data(st, f, entry->idx.sha1); | |
336 | close_istream(st); | |
337 | } else { | |
338 | sha1write(f, buf, datalen); | |
339 | free(buf); | |
340 | } | |
c9018b03 NTND |
341 | |
342 | return hdrlen + datalen; | |
343 | } | |
344 | ||
345 | /* Return 0 if we will bust the pack-size limit */ | |
346 | static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry, | |
347 | unsigned long limit, int usable_delta) | |
348 | { | |
349 | struct packed_git *p = entry->in_pack; | |
350 | struct pack_window *w_curs = NULL; | |
351 | struct revindex_entry *revidx; | |
352 | off_t offset; | |
353 | enum object_type type = entry->type; | |
354 | unsigned long datalen; | |
355 | unsigned char header[10], dheader[10]; | |
356 | unsigned hdrlen; | |
357 | ||
358 | if (entry->delta) | |
359 | type = (allow_ofs_delta && entry->delta->idx.offset) ? | |
360 | OBJ_OFS_DELTA : OBJ_REF_DELTA; | |
361 | hdrlen = encode_in_pack_object_header(type, entry->size, header); | |
362 | ||
363 | offset = entry->in_pack_offset; | |
364 | revidx = find_pack_revindex(p, offset); | |
365 | datalen = revidx[1].offset - offset; | |
366 | if (!pack_to_stdout && p->index_version > 1 && | |
367 | check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { | |
368 | error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1)); | |
369 | unuse_pack(&w_curs); | |
370 | return write_no_reuse_object(f, entry, limit, usable_delta); | |
371 | } | |
372 | ||
373 | offset += entry->in_pack_header_size; | |
374 | datalen -= entry->in_pack_header_size; | |
375 | ||
376 | if (!pack_to_stdout && p->index_version == 1 && | |
377 | check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) { | |
378 | error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1)); | |
379 | unuse_pack(&w_curs); | |
380 | return write_no_reuse_object(f, entry, limit, usable_delta); | |
381 | } | |
382 | ||
383 | if (type == OBJ_OFS_DELTA) { | |
384 | off_t ofs = entry->idx.offset - entry->delta->idx.offset; | |
385 | unsigned pos = sizeof(dheader) - 1; | |
386 | dheader[pos] = ofs & 127; | |
387 | while (ofs >>= 7) | |
388 | dheader[--pos] = 128 | (--ofs & 127); | |
389 | if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) { | |
390 | unuse_pack(&w_curs); | |
391 | return 0; | |
392 | } | |
393 | sha1write(f, header, hdrlen); | |
394 | sha1write(f, dheader + pos, sizeof(dheader) - pos); | |
395 | hdrlen += sizeof(dheader) - pos; | |
396 | reused_delta++; | |
397 | } else if (type == OBJ_REF_DELTA) { | |
398 | if (limit && hdrlen + 20 + datalen + 20 >= limit) { | |
399 | unuse_pack(&w_curs); | |
400 | return 0; | |
401 | } | |
402 | sha1write(f, header, hdrlen); | |
403 | sha1write(f, entry->delta->idx.sha1, 20); | |
404 | hdrlen += 20; | |
405 | reused_delta++; | |
406 | } else { | |
407 | if (limit && hdrlen + datalen + 20 >= limit) { | |
408 | unuse_pack(&w_curs); | |
409 | return 0; | |
410 | } | |
411 | sha1write(f, header, hdrlen); | |
412 | } | |
413 | copy_pack_data(f, p, &w_curs, offset, datalen); | |
414 | unuse_pack(&w_curs); | |
415 | reused++; | |
416 | return hdrlen + datalen; | |
417 | } | |
418 | ||
419 | /* Return 0 if we will bust the pack-size limit */ | |
420 | static unsigned long write_object(struct sha1file *f, | |
421 | struct object_entry *entry, | |
422 | off_t write_offset) | |
423 | { | |
424 | unsigned long limit, len; | |
2c5ef824 | 425 | int usable_delta, to_reuse; |
c323ac7d | 426 | |
78d1e84f NP |
427 | if (!pack_to_stdout) |
428 | crc32_begin(f); | |
429 | ||
a2430dde | 430 | /* apply size limit if limited packsize and not first object */ |
a1e4760f NP |
431 | if (!pack_size_limit || !nr_written) |
432 | limit = 0; | |
433 | else if (pack_size_limit <= write_offset) | |
434 | /* | |
435 | * the earlier object did not fit the limit; avoid | |
436 | * mistaking this with unlimited (i.e. limit = 0). | |
437 | */ | |
438 | limit = 1; | |
439 | else | |
440 | limit = pack_size_limit - write_offset; | |
2c5ef824 NP |
441 | |
442 | if (!entry->delta) | |
443 | usable_delta = 0; /* no delta */ | |
444 | else if (!pack_size_limit) | |
445 | usable_delta = 1; /* unlimited packfile */ | |
446 | else if (entry->delta->idx.offset == (off_t)-1) | |
447 | usable_delta = 0; /* base was written to another pack */ | |
448 | else if (entry->delta->idx.offset) | |
449 | usable_delta = 1; /* base already exists in this pack */ | |
450 | else | |
451 | usable_delta = 0; /* base could end up in another pack */ | |
452 | ||
a7de7130 | 453 | if (!reuse_object) |
fa736f72 NP |
454 | to_reuse = 0; /* explicit */ |
455 | else if (!entry->in_pack) | |
ab7cd7bb | 456 | to_reuse = 0; /* can't reuse what we don't have */ |
c9018b03 | 457 | else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA) |
17b08f2c DH |
458 | /* check_object() decided it for us ... */ |
459 | to_reuse = usable_delta; | |
460 | /* ... but pack split may override that */ | |
c9018b03 | 461 | else if (entry->type != entry->in_pack_type) |
ab7cd7bb JH |
462 | to_reuse = 0; /* pack has delta which is unusable */ |
463 | else if (entry->delta) | |
464 | to_reuse = 0; /* we want to pack afresh */ | |
465 | else | |
466 | to_reuse = 1; /* we have it in-pack undeltified, | |
467 | * and we do not need to deltify it. | |
468 | */ | |
469 | ||
c9018b03 NTND |
470 | if (!to_reuse) |
471 | len = write_no_reuse_object(f, entry, limit, usable_delta); | |
472 | else | |
473 | len = write_reuse_object(f, entry, limit, usable_delta); | |
474 | if (!len) | |
475 | return 0; | |
64bd76b1 | 476 | |
17b08f2c | 477 | if (usable_delta) |
ab7cd7bb | 478 | written_delta++; |
3f9ac8d2 | 479 | written++; |
78d1e84f | 480 | if (!pack_to_stdout) |
aa7e44bf | 481 | entry->idx.crc32 = crc32_end(f); |
c9018b03 | 482 | return len; |
c323ac7d LT |
483 | } |
484 | ||
f63c79db JH |
485 | enum write_one_status { |
486 | WRITE_ONE_SKIP = -1, /* already written */ | |
487 | WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ | |
488 | WRITE_ONE_WRITTEN = 1, /* normal */ | |
489 | WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ | |
490 | }; | |
491 | ||
492 | static enum write_one_status write_one(struct sha1file *f, | |
493 | struct object_entry *e, | |
494 | off_t *offset) | |
9d5ab962 | 495 | { |
d7dd0223 | 496 | unsigned long size; |
f63c79db | 497 | int recursing; |
d7dd0223 | 498 | |
f63c79db JH |
499 | /* |
500 | * we set offset to 1 (which is an impossible value) to mark | |
501 | * the fact that this object is involved in "write its base | |
502 | * first before writing a deltified object" recursion. | |
503 | */ | |
504 | recursing = (e->idx.offset == 1); | |
505 | if (recursing) { | |
506 | warning("recursive delta detected for object %s", | |
507 | sha1_to_hex(e->idx.sha1)); | |
508 | return WRITE_ONE_RECURSIVE; | |
509 | } else if (e->idx.offset || e->preferred_base) { | |
510 | /* offset is non zero if object is written already. */ | |
511 | return WRITE_ONE_SKIP; | |
512 | } | |
d7dd0223 | 513 | |
720c9f7b | 514 | /* if we are deltified, write out base object first. */ |
f63c79db JH |
515 | if (e->delta) { |
516 | e->idx.offset = 1; /* now recurse */ | |
517 | switch (write_one(f, e->delta, offset)) { | |
518 | case WRITE_ONE_RECURSIVE: | |
519 | /* we cannot depend on this one */ | |
520 | e->delta = NULL; | |
521 | break; | |
522 | default: | |
523 | break; | |
524 | case WRITE_ONE_BREAK: | |
525 | e->idx.offset = recursing; | |
526 | return WRITE_ONE_BREAK; | |
527 | } | |
528 | } | |
d7dd0223 | 529 | |
6ed7f25e NP |
530 | e->idx.offset = *offset; |
531 | size = write_object(f, e, *offset); | |
17b08f2c | 532 | if (!size) { |
f63c79db JH |
533 | e->idx.offset = recursing; |
534 | return WRITE_ONE_BREAK; | |
17b08f2c | 535 | } |
79814f42 | 536 | written_list[nr_written++] = &e->idx; |
d7dd0223 NP |
537 | |
538 | /* make sure off_t is sufficiently large not to wrap */ | |
c03c8315 | 539 | if (signed_add_overflows(*offset, size)) |
d7dd0223 | 540 | die("pack too large for current definition of off_t"); |
6ed7f25e | 541 | *offset += size; |
f63c79db | 542 | return WRITE_ONE_WRITTEN; |
9d5ab962 JH |
543 | } |
544 | ||
d155254c | 545 | static int mark_tagged(const char *path, const struct object_id *oid, int flag, |
1b4bb16b JH |
546 | void *cb_data) |
547 | { | |
548 | unsigned char peeled[20]; | |
d155254c | 549 | struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL); |
1b4bb16b JH |
550 | |
551 | if (entry) | |
552 | entry->tagged = 1; | |
553 | if (!peel_ref(path, peeled)) { | |
2834bc27 | 554 | entry = packlist_find(&to_pack, peeled, NULL); |
1b4bb16b JH |
555 | if (entry) |
556 | entry->tagged = 1; | |
557 | } | |
558 | return 0; | |
559 | } | |
560 | ||
be126818 | 561 | static inline void add_to_write_order(struct object_entry **wo, |
92bef1a1 | 562 | unsigned int *endp, |
1b4bb16b JH |
563 | struct object_entry *e) |
564 | { | |
565 | if (e->filled) | |
566 | return; | |
567 | wo[(*endp)++] = e; | |
568 | e->filled = 1; | |
569 | } | |
570 | ||
571 | static void add_descendants_to_write_order(struct object_entry **wo, | |
92bef1a1 | 572 | unsigned int *endp, |
1b4bb16b JH |
573 | struct object_entry *e) |
574 | { | |
f380872f DM |
575 | int add_to_order = 1; |
576 | while (e) { | |
577 | if (add_to_order) { | |
578 | struct object_entry *s; | |
579 | /* add this node... */ | |
580 | add_to_write_order(wo, endp, e); | |
581 | /* all its siblings... */ | |
582 | for (s = e->delta_sibling; s; s = s->delta_sibling) { | |
583 | add_to_write_order(wo, endp, s); | |
584 | } | |
585 | } | |
586 | /* drop down a level to add left subtree nodes if possible */ | |
587 | if (e->delta_child) { | |
588 | add_to_order = 1; | |
589 | e = e->delta_child; | |
590 | } else { | |
591 | add_to_order = 0; | |
592 | /* our sibling might have some children, it is next */ | |
593 | if (e->delta_sibling) { | |
594 | e = e->delta_sibling; | |
595 | continue; | |
596 | } | |
597 | /* go back to our parent node */ | |
598 | e = e->delta; | |
599 | while (e && !e->delta_sibling) { | |
600 | /* we're on the right side of a subtree, keep | |
601 | * going up until we can go right again */ | |
602 | e = e->delta; | |
603 | } | |
604 | if (!e) { | |
605 | /* done- we hit our original root node */ | |
606 | return; | |
607 | } | |
608 | /* pass it off to sibling at this level */ | |
609 | e = e->delta_sibling; | |
610 | } | |
611 | }; | |
1b4bb16b JH |
612 | } |
613 | ||
614 | static void add_family_to_write_order(struct object_entry **wo, | |
92bef1a1 | 615 | unsigned int *endp, |
1b4bb16b JH |
616 | struct object_entry *e) |
617 | { | |
618 | struct object_entry *root; | |
619 | ||
620 | for (root = e; root->delta; root = root->delta) | |
621 | ; /* nothing */ | |
1b4bb16b JH |
622 | add_descendants_to_write_order(wo, endp, root); |
623 | } | |
624 | ||
625 | static struct object_entry **compute_write_order(void) | |
626 | { | |
38d4debb | 627 | unsigned int i, wo_end, last_untagged; |
1b4bb16b | 628 | |
b32fa95f | 629 | struct object_entry **wo; |
2834bc27 | 630 | struct object_entry *objects = to_pack.objects; |
1b4bb16b | 631 | |
2834bc27 | 632 | for (i = 0; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
633 | objects[i].tagged = 0; |
634 | objects[i].filled = 0; | |
635 | objects[i].delta_child = NULL; | |
636 | objects[i].delta_sibling = NULL; | |
637 | } | |
638 | ||
639 | /* | |
640 | * Fully connect delta_child/delta_sibling network. | |
641 | * Make sure delta_sibling is sorted in the original | |
642 | * recency order. | |
643 | */ | |
2834bc27 | 644 | for (i = to_pack.nr_objects; i > 0;) { |
92bef1a1 | 645 | struct object_entry *e = &objects[--i]; |
1b4bb16b JH |
646 | if (!e->delta) |
647 | continue; | |
648 | /* Mark me as the first child */ | |
649 | e->delta_sibling = e->delta->delta_child; | |
650 | e->delta->delta_child = e; | |
651 | } | |
652 | ||
653 | /* | |
654 | * Mark objects that are at the tip of tags. | |
655 | */ | |
d155254c | 656 | for_each_tag_ref(mark_tagged, NULL); |
1b4bb16b JH |
657 | |
658 | /* | |
38d4debb | 659 | * Give the objects in the original recency order until |
1b4bb16b JH |
660 | * we see a tagged tip. |
661 | */ | |
b32fa95f | 662 | ALLOC_ARRAY(wo, to_pack.nr_objects); |
2834bc27 | 663 | for (i = wo_end = 0; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
664 | if (objects[i].tagged) |
665 | break; | |
666 | add_to_write_order(wo, &wo_end, &objects[i]); | |
667 | } | |
38d4debb | 668 | last_untagged = i; |
1b4bb16b JH |
669 | |
670 | /* | |
671 | * Then fill all the tagged tips. | |
672 | */ | |
2834bc27 | 673 | for (; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
674 | if (objects[i].tagged) |
675 | add_to_write_order(wo, &wo_end, &objects[i]); | |
676 | } | |
677 | ||
678 | /* | |
679 | * And then all remaining commits and tags. | |
680 | */ | |
2834bc27 | 681 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
682 | if (objects[i].type != OBJ_COMMIT && |
683 | objects[i].type != OBJ_TAG) | |
684 | continue; | |
685 | add_to_write_order(wo, &wo_end, &objects[i]); | |
686 | } | |
687 | ||
688 | /* | |
689 | * And then all the trees. | |
690 | */ | |
2834bc27 | 691 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
1b4bb16b JH |
692 | if (objects[i].type != OBJ_TREE) |
693 | continue; | |
694 | add_to_write_order(wo, &wo_end, &objects[i]); | |
695 | } | |
696 | ||
697 | /* | |
698 | * Finally all the rest in really tight order | |
699 | */ | |
2834bc27 | 700 | for (i = last_untagged; i < to_pack.nr_objects; i++) { |
38d4debb DM |
701 | if (!objects[i].filled) |
702 | add_family_to_write_order(wo, &wo_end, &objects[i]); | |
703 | } | |
704 | ||
2834bc27 VM |
705 | if (wo_end != to_pack.nr_objects) |
706 | die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects); | |
1b4bb16b JH |
707 | |
708 | return wo; | |
709 | } | |
710 | ||
6b8fda2d VM |
711 | static off_t write_reused_pack(struct sha1file *f) |
712 | { | |
713 | unsigned char buffer[8192]; | |
657673f1 | 714 | off_t to_write, total; |
6b8fda2d VM |
715 | int fd; |
716 | ||
717 | if (!is_pack_valid(reuse_packfile)) | |
718 | die("packfile is invalid: %s", reuse_packfile->pack_name); | |
719 | ||
720 | fd = git_open_noatime(reuse_packfile->pack_name); | |
721 | if (fd < 0) | |
722 | die_errno("unable to open packfile for reuse: %s", | |
723 | reuse_packfile->pack_name); | |
724 | ||
725 | if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) | |
726 | die_errno("unable to seek in reused packfile"); | |
727 | ||
728 | if (reuse_packfile_offset < 0) | |
729 | reuse_packfile_offset = reuse_packfile->pack_size - 20; | |
730 | ||
657673f1 | 731 | total = to_write = reuse_packfile_offset - sizeof(struct pack_header); |
6b8fda2d VM |
732 | |
733 | while (to_write) { | |
734 | int read_pack = xread(fd, buffer, sizeof(buffer)); | |
735 | ||
736 | if (read_pack <= 0) | |
737 | die_errno("unable to read from reused packfile"); | |
738 | ||
739 | if (read_pack > to_write) | |
740 | read_pack = to_write; | |
741 | ||
742 | sha1write(f, buffer, read_pack); | |
743 | to_write -= read_pack; | |
657673f1 JK |
744 | |
745 | /* | |
746 | * We don't know the actual number of objects written, | |
747 | * only how many bytes written, how many bytes total, and | |
748 | * how many objects total. So we can fake it by pretending all | |
749 | * objects we are writing are the same size. This gives us a | |
750 | * smooth progress meter, and at the end it matches the true | |
751 | * answer. | |
752 | */ | |
753 | written = reuse_packfile_objects * | |
754 | (((double)(total - to_write)) / total); | |
755 | display_progress(progress_state, written); | |
6b8fda2d VM |
756 | } |
757 | ||
758 | close(fd); | |
657673f1 JK |
759 | written = reuse_packfile_objects; |
760 | display_progress(progress_state, written); | |
6b8fda2d VM |
761 | return reuse_packfile_offset - sizeof(struct pack_header); |
762 | } | |
763 | ||
9cea46cd EW |
764 | static const char no_split_warning[] = N_( |
765 | "disabling bitmap writing, packs are split due to pack.packSizeLimit" | |
766 | ); | |
767 | ||
d01fb92f | 768 | static void write_pack_file(void) |
c323ac7d | 769 | { |
ebe27b13 | 770 | uint32_t i = 0, j; |
d22b9290 | 771 | struct sha1file *f; |
6ed7f25e | 772 | off_t offset; |
ebe27b13 | 773 | uint32_t nr_remaining = nr_result; |
f746bae8 | 774 | time_t last_mtime = 0; |
1b4bb16b | 775 | struct object_entry **write_order; |
81a216a5 | 776 | |
bcd7954e | 777 | if (progress > pack_to_stdout) |
754dbc43 | 778 | progress_state = start_progress(_("Writing objects"), nr_result); |
b32fa95f | 779 | ALLOC_ARRAY(written_list, to_pack.nr_objects); |
1b4bb16b | 780 | write_order = compute_write_order(); |
183bdb2c | 781 | |
ebe27b13 | 782 | do { |
aa7e44bf | 783 | unsigned char sha1[20]; |
7ba502c4 | 784 | char *pack_tmp_name = NULL; |
aa7e44bf | 785 | |
cdf9db3c | 786 | if (pack_to_stdout) |
2a128d63 | 787 | f = sha1fd_throughput(1, "<stdout>", progress_state); |
cdf9db3c JH |
788 | else |
789 | f = create_tmp_packfile(&pack_tmp_name); | |
ebe27b13 | 790 | |
c0ad4657 | 791 | offset = write_pack_header(f, nr_remaining); |
6b8fda2d VM |
792 | |
793 | if (reuse_packfile) { | |
794 | off_t packfile_size; | |
795 | assert(pack_to_stdout); | |
796 | ||
797 | packfile_size = write_reused_pack(f); | |
798 | offset += packfile_size; | |
799 | } | |
800 | ||
ebe27b13 | 801 | nr_written = 0; |
2834bc27 | 802 | for (; i < to_pack.nr_objects; i++) { |
1b4bb16b | 803 | struct object_entry *e = write_order[i]; |
cddec4f8 | 804 | if (write_one(f, e, &offset) == WRITE_ONE_BREAK) |
720c9f7b NP |
805 | break; |
806 | display_progress(progress_state, written); | |
807 | } | |
c553ca25 | 808 | |
ebe27b13 DH |
809 | /* |
810 | * Did we write the wrong # entries in the header? | |
811 | * If so, rewrite it like in fast-import | |
812 | */ | |
54352bb2 LT |
813 | if (pack_to_stdout) { |
814 | sha1close(f, sha1, CSUM_CLOSE); | |
815 | } else if (nr_written == nr_remaining) { | |
816 | sha1close(f, sha1, CSUM_FSYNC); | |
ebe27b13 | 817 | } else { |
ac0463ed | 818 | int fd = sha1close(f, sha1, 0); |
abeb40e5 | 819 | fixup_pack_header_footer(fd, sha1, pack_tmp_name, |
ac0463ed | 820 | nr_written, sha1, offset); |
7ba502c4 | 821 | close(fd); |
9cea46cd EW |
822 | if (write_bitmap_index) { |
823 | warning(_(no_split_warning)); | |
824 | write_bitmap_index = 0; | |
825 | } | |
ebe27b13 DH |
826 | } |
827 | ||
828 | if (!pack_to_stdout) { | |
f746bae8 | 829 | struct stat st; |
58892711 | 830 | struct strbuf tmpname = STRBUF_INIT; |
d01fb92f | 831 | |
f746bae8 NP |
832 | /* |
833 | * Packs are runtime accessed in their mtime | |
834 | * order since newer packs are more likely to contain | |
835 | * younger objects. So if we are creating multiple | |
836 | * packs then we should modify the mtime of later ones | |
837 | * to preserve this property. | |
838 | */ | |
0e990530 | 839 | if (stat(pack_tmp_name, &st) < 0) { |
54d47394 | 840 | warning_errno("failed to stat %s", pack_tmp_name); |
f746bae8 NP |
841 | } else if (!last_mtime) { |
842 | last_mtime = st.st_mtime; | |
843 | } else { | |
844 | struct utimbuf utb; | |
845 | utb.actime = st.st_atime; | |
846 | utb.modtime = --last_mtime; | |
0e990530 | 847 | if (utime(pack_tmp_name, &utb) < 0) |
54d47394 | 848 | warning_errno("failed utime() on %s", pack_tmp_name); |
f746bae8 NP |
849 | } |
850 | ||
58892711 | 851 | strbuf_addf(&tmpname, "%s-", base_name); |
7cc8f971 VM |
852 | |
853 | if (write_bitmap_index) { | |
854 | bitmap_writer_set_checksum(sha1); | |
855 | bitmap_writer_build_type_index(written_list, nr_written); | |
856 | } | |
857 | ||
58892711 | 858 | finish_tmp_packfile(&tmpname, pack_tmp_name, |
0e990530 JH |
859 | written_list, nr_written, |
860 | &pack_idx_opts, sha1); | |
7cc8f971 VM |
861 | |
862 | if (write_bitmap_index) { | |
58892711 | 863 | strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1)); |
7cc8f971 VM |
864 | |
865 | stop_progress(&progress_state); | |
866 | ||
867 | bitmap_writer_show_progress(progress); | |
868 | bitmap_writer_reuse_bitmaps(&to_pack); | |
869 | bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); | |
870 | bitmap_writer_build(&to_pack); | |
ae4f07fb | 871 | bitmap_writer_finish(written_list, nr_written, |
58892711 | 872 | tmpname.buf, write_bitmap_options); |
7cc8f971 VM |
873 | write_bitmap_index = 0; |
874 | } | |
875 | ||
58892711 | 876 | strbuf_release(&tmpname); |
7ba502c4 | 877 | free(pack_tmp_name); |
aa7e44bf | 878 | puts(sha1_to_hex(sha1)); |
ebe27b13 DH |
879 | } |
880 | ||
881 | /* mark written objects as written to previous pack */ | |
882 | for (j = 0; j < nr_written; j++) { | |
79814f42 | 883 | written_list[j]->offset = (off_t)-1; |
ebe27b13 DH |
884 | } |
885 | nr_remaining -= nr_written; | |
2834bc27 | 886 | } while (nr_remaining && i < to_pack.nr_objects); |
ebe27b13 DH |
887 | |
888 | free(written_list); | |
1b4bb16b | 889 | free(write_order); |
4d4fcc54 | 890 | stop_progress(&progress_state); |
67c08ce1 | 891 | if (written != nr_result) |
6e1c2344 RJ |
892 | die("wrote %"PRIu32" objects while expecting %"PRIu32, |
893 | written, nr_result); | |
c323ac7d LT |
894 | } |
895 | ||
a74db82e JH |
896 | static void setup_delta_attr_check(struct git_attr_check *check) |
897 | { | |
898 | static struct git_attr *attr_delta; | |
899 | ||
900 | if (!attr_delta) | |
7fb0eaa2 | 901 | attr_delta = git_attr("delta"); |
a74db82e JH |
902 | |
903 | check[0].attr = attr_delta; | |
904 | } | |
905 | ||
906 | static int no_try_delta(const char *path) | |
907 | { | |
908 | struct git_attr_check check[1]; | |
909 | ||
910 | setup_delta_attr_check(check); | |
d932f4eb | 911 | if (git_check_attr(path, ARRAY_SIZE(check), check)) |
a74db82e JH |
912 | return 0; |
913 | if (ATTR_FALSE(check->value)) | |
914 | return 1; | |
915 | return 0; | |
916 | } | |
917 | ||
ce2bc424 JK |
918 | /* |
919 | * When adding an object, check whether we have already added it | |
920 | * to our packing list. If so, we can skip. However, if we are | |
921 | * being asked to excludei t, but the previous mention was to include | |
922 | * it, make sure to adjust its flags and tweak our numbers accordingly. | |
923 | * | |
924 | * As an optimization, we pass out the index position where we would have | |
925 | * found the item, since that saves us from having to look it up again a | |
926 | * few lines later when we want to add the new entry. | |
927 | */ | |
928 | static int have_duplicate_entry(const unsigned char *sha1, | |
929 | int exclude, | |
930 | uint32_t *index_pos) | |
c323ac7d | 931 | { |
c323ac7d | 932 | struct object_entry *entry; |
29b734e4 | 933 | |
ce2bc424 JK |
934 | entry = packlist_find(&to_pack, sha1, index_pos); |
935 | if (!entry) | |
29b734e4 | 936 | return 0; |
ce2bc424 JK |
937 | |
938 | if (exclude) { | |
939 | if (!entry->preferred_base) | |
940 | nr_result--; | |
941 | entry->preferred_base = 1; | |
29b734e4 | 942 | } |
c323ac7d | 943 | |
ce2bc424 JK |
944 | return 1; |
945 | } | |
946 | ||
947 | /* | |
948 | * Check whether we want the object in the pack (e.g., we do not want | |
949 | * objects found in non-local stores if the "--local" option was used). | |
950 | * | |
951 | * As a side effect of this check, we will find the packed version of this | |
952 | * object, if any. We therefore pass out the pack information to avoid having | |
953 | * to look it up again later. | |
954 | */ | |
955 | static int want_object_in_pack(const unsigned char *sha1, | |
956 | int exclude, | |
957 | struct packed_git **found_pack, | |
958 | off_t *found_offset) | |
959 | { | |
960 | struct packed_git *p; | |
961 | ||
daae0625 BC |
962 | if (!exclude && local && has_loose_object_nonlocal(sha1)) |
963 | return 0; | |
964 | ||
ce2bc424 JK |
965 | *found_pack = NULL; |
966 | *found_offset = 0; | |
967 | ||
5c49c116 NP |
968 | for (p = packed_git; p; p = p->next) { |
969 | off_t offset = find_pack_entry_one(sha1, p); | |
970 | if (offset) { | |
ce2bc424 | 971 | if (!*found_pack) { |
319b678a | 972 | if (!is_pack_valid(p)) |
4c080182 | 973 | continue; |
ce2bc424 JK |
974 | *found_offset = offset; |
975 | *found_pack = p; | |
64560374 | 976 | } |
5c49c116 | 977 | if (exclude) |
ce2bc424 | 978 | return 1; |
5c49c116 NP |
979 | if (incremental) |
980 | return 0; | |
cd379967 JK |
981 | |
982 | /* | |
983 | * When asked to do --local (do not include an | |
984 | * object that appears in a pack we borrow | |
985 | * from elsewhere) or --honor-pack-keep (do not | |
986 | * include an object that appears in a pack marked | |
987 | * with .keep), we need to make sure no copy of this | |
988 | * object come from in _any_ pack that causes us to | |
989 | * omit it, and need to complete this loop. When | |
990 | * neither option is in effect, we know the object | |
991 | * we just found is going to be packed, so break | |
992 | * out of the loop to return 1 now. | |
993 | */ | |
56dfeb62 JK |
994 | if (!ignore_packed_keep && |
995 | (!local || !have_non_local_packs)) | |
cd379967 JK |
996 | break; |
997 | ||
5c49c116 NP |
998 | if (local && !p->pack_local) |
999 | return 0; | |
e96fb9b8 BC |
1000 | if (ignore_packed_keep && p->pack_local && p->pack_keep) |
1001 | return 0; | |
64560374 LT |
1002 | } |
1003 | } | |
eb019375 | 1004 | |
ce2bc424 JK |
1005 | return 1; |
1006 | } | |
1007 | ||
1008 | static void create_object_entry(const unsigned char *sha1, | |
1009 | enum object_type type, | |
1010 | uint32_t hash, | |
1011 | int exclude, | |
1012 | int no_try_delta, | |
1013 | uint32_t index_pos, | |
1014 | struct packed_git *found_pack, | |
1015 | off_t found_offset) | |
1016 | { | |
1017 | struct object_entry *entry; | |
29b734e4 | 1018 | |
2834bc27 | 1019 | entry = packlist_alloc(&to_pack, sha1, index_pos); |
27225f2e | 1020 | entry->hash = hash; |
5c49c116 NP |
1021 | if (type) |
1022 | entry->type = type; | |
29b734e4 NP |
1023 | if (exclude) |
1024 | entry->preferred_base = 1; | |
81a216a5 NP |
1025 | else |
1026 | nr_result++; | |
29b734e4 NP |
1027 | if (found_pack) { |
1028 | entry->in_pack = found_pack; | |
1029 | entry->in_pack_offset = found_offset; | |
1030 | } | |
7a979d99 | 1031 | |
ce2bc424 JK |
1032 | entry->no_try_delta = no_try_delta; |
1033 | } | |
29b734e4 | 1034 | |
373c67da JK |
1035 | static const char no_closure_warning[] = N_( |
1036 | "disabling bitmap writing, as some objects are not being packed" | |
1037 | ); | |
1038 | ||
ce2bc424 JK |
1039 | static int add_object_entry(const unsigned char *sha1, enum object_type type, |
1040 | const char *name, int exclude) | |
1041 | { | |
1042 | struct packed_git *found_pack; | |
1043 | off_t found_offset; | |
1044 | uint32_t index_pos; | |
7a979d99 | 1045 | |
ce2bc424 JK |
1046 | if (have_duplicate_entry(sha1, exclude, &index_pos)) |
1047 | return 0; | |
a74db82e | 1048 | |
373c67da JK |
1049 | if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) { |
1050 | /* The pack is missing an object, so it will not have closure */ | |
1051 | if (write_bitmap_index) { | |
1052 | warning(_(no_closure_warning)); | |
1053 | write_bitmap_index = 0; | |
1054 | } | |
ce2bc424 | 1055 | return 0; |
373c67da | 1056 | } |
29b734e4 | 1057 | |
ce2bc424 JK |
1058 | create_object_entry(sha1, type, pack_name_hash(name), |
1059 | exclude, name && no_try_delta(name), | |
1060 | index_pos, found_pack, found_offset); | |
a74db82e | 1061 | |
78d2214e | 1062 | display_progress(progress_state, nr_result); |
29b734e4 | 1063 | return 1; |
c323ac7d LT |
1064 | } |
1065 | ||
6b8fda2d VM |
1066 | static int add_object_entry_from_bitmap(const unsigned char *sha1, |
1067 | enum object_type type, | |
1068 | int flags, uint32_t name_hash, | |
1069 | struct packed_git *pack, off_t offset) | |
1070 | { | |
1071 | uint32_t index_pos; | |
1072 | ||
1073 | if (have_duplicate_entry(sha1, 0, &index_pos)) | |
1074 | return 0; | |
1075 | ||
1076 | create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset); | |
1077 | ||
78d2214e | 1078 | display_progress(progress_state, nr_result); |
29b734e4 | 1079 | return 1; |
c323ac7d LT |
1080 | } |
1081 | ||
5379a5c5 JH |
1082 | struct pbase_tree_cache { |
1083 | unsigned char sha1[20]; | |
1084 | int ref; | |
1085 | int temporary; | |
1086 | void *tree_data; | |
1087 | unsigned long tree_size; | |
1088 | }; | |
1089 | ||
1090 | static struct pbase_tree_cache *(pbase_tree_cache[256]); | |
1091 | static int pbase_tree_cache_ix(const unsigned char *sha1) | |
1092 | { | |
1093 | return sha1[0] % ARRAY_SIZE(pbase_tree_cache); | |
1094 | } | |
1095 | static int pbase_tree_cache_ix_incr(int ix) | |
1096 | { | |
1097 | return (ix+1) % ARRAY_SIZE(pbase_tree_cache); | |
1098 | } | |
1099 | ||
1100 | static struct pbase_tree { | |
1101 | struct pbase_tree *next; | |
1102 | /* This is a phony "cache" entry; we are not | |
01689909 | 1103 | * going to evict it or find it through _get() |
5379a5c5 JH |
1104 | * mechanism -- this is for the toplevel node that |
1105 | * would almost always change with any commit. | |
1106 | */ | |
1107 | struct pbase_tree_cache pcache; | |
1108 | } *pbase_tree; | |
1109 | ||
1110 | static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1) | |
1111 | { | |
1112 | struct pbase_tree_cache *ent, *nent; | |
1113 | void *data; | |
1114 | unsigned long size; | |
21666f1a | 1115 | enum object_type type; |
5379a5c5 JH |
1116 | int neigh; |
1117 | int my_ix = pbase_tree_cache_ix(sha1); | |
1118 | int available_ix = -1; | |
1119 | ||
1120 | /* pbase-tree-cache acts as a limited hashtable. | |
1121 | * your object will be found at your index or within a few | |
1122 | * slots after that slot if it is cached. | |
1123 | */ | |
1124 | for (neigh = 0; neigh < 8; neigh++) { | |
1125 | ent = pbase_tree_cache[my_ix]; | |
a89fccd2 | 1126 | if (ent && !hashcmp(ent->sha1, sha1)) { |
5379a5c5 JH |
1127 | ent->ref++; |
1128 | return ent; | |
1129 | } | |
1130 | else if (((available_ix < 0) && (!ent || !ent->ref)) || | |
1131 | ((0 <= available_ix) && | |
1132 | (!ent && pbase_tree_cache[available_ix]))) | |
1133 | available_ix = my_ix; | |
1134 | if (!ent) | |
1135 | break; | |
1136 | my_ix = pbase_tree_cache_ix_incr(my_ix); | |
1137 | } | |
1138 | ||
1139 | /* Did not find one. Either we got a bogus request or | |
1140 | * we need to read and perhaps cache. | |
1141 | */ | |
21666f1a | 1142 | data = read_sha1_file(sha1, &type, &size); |
5379a5c5 JH |
1143 | if (!data) |
1144 | return NULL; | |
21666f1a | 1145 | if (type != OBJ_TREE) { |
5379a5c5 JH |
1146 | free(data); |
1147 | return NULL; | |
1148 | } | |
1149 | ||
1150 | /* We need to either cache or return a throwaway copy */ | |
1151 | ||
1152 | if (available_ix < 0) | |
1153 | ent = NULL; | |
1154 | else { | |
1155 | ent = pbase_tree_cache[available_ix]; | |
1156 | my_ix = available_ix; | |
1157 | } | |
1158 | ||
1159 | if (!ent) { | |
1160 | nent = xmalloc(sizeof(*nent)); | |
1161 | nent->temporary = (available_ix < 0); | |
1162 | } | |
1163 | else { | |
1164 | /* evict and reuse */ | |
1165 | free(ent->tree_data); | |
1166 | nent = ent; | |
1167 | } | |
e702496e | 1168 | hashcpy(nent->sha1, sha1); |
5379a5c5 JH |
1169 | nent->tree_data = data; |
1170 | nent->tree_size = size; | |
1171 | nent->ref = 1; | |
1172 | if (!nent->temporary) | |
1173 | pbase_tree_cache[my_ix] = nent; | |
1174 | return nent; | |
1175 | } | |
1176 | ||
1177 | static void pbase_tree_put(struct pbase_tree_cache *cache) | |
1178 | { | |
1179 | if (!cache->temporary) { | |
1180 | cache->ref--; | |
1181 | return; | |
1182 | } | |
1183 | free(cache->tree_data); | |
1184 | free(cache); | |
1185 | } | |
1186 | ||
1187 | static int name_cmp_len(const char *name) | |
1188 | { | |
1189 | int i; | |
1190 | for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++) | |
1191 | ; | |
1192 | return i; | |
1193 | } | |
1194 | ||
1195 | static void add_pbase_object(struct tree_desc *tree, | |
5379a5c5 | 1196 | const char *name, |
ce0bd642 LT |
1197 | int cmplen, |
1198 | const char *fullname) | |
3f9ac8d2 | 1199 | { |
4c068a98 | 1200 | struct name_entry entry; |
8a5a8d6c | 1201 | int cmp; |
4c068a98 LT |
1202 | |
1203 | while (tree_entry(tree,&entry)) { | |
1211be6b LT |
1204 | if (S_ISGITLINK(entry.mode)) |
1205 | continue; | |
0de16337 | 1206 | cmp = tree_entry_len(&entry) != cmplen ? 1 : |
8a5a8d6c NP |
1207 | memcmp(name, entry.path, cmplen); |
1208 | if (cmp > 0) | |
7a979d99 | 1209 | continue; |
8a5a8d6c NP |
1210 | if (cmp < 0) |
1211 | return; | |
5379a5c5 | 1212 | if (name[cmplen] != '/') { |
7d924c91 | 1213 | add_object_entry(entry.oid->hash, |
4d1012c3 | 1214 | object_type(entry.mode), |
bc32fed5 | 1215 | fullname, 1); |
5379a5c5 JH |
1216 | return; |
1217 | } | |
8a5a8d6c | 1218 | if (S_ISDIR(entry.mode)) { |
7a979d99 | 1219 | struct tree_desc sub; |
5379a5c5 JH |
1220 | struct pbase_tree_cache *tree; |
1221 | const char *down = name+cmplen+1; | |
1222 | int downlen = name_cmp_len(down); | |
1223 | ||
7d924c91 | 1224 | tree = pbase_tree_get(entry.oid->hash); |
5379a5c5 JH |
1225 | if (!tree) |
1226 | return; | |
6fda5e51 | 1227 | init_tree_desc(&sub, tree->tree_data, tree->tree_size); |
5379a5c5 | 1228 | |
ce0bd642 | 1229 | add_pbase_object(&sub, down, downlen, fullname); |
5379a5c5 JH |
1230 | pbase_tree_put(tree); |
1231 | } | |
1232 | } | |
1233 | } | |
1d6b38cc | 1234 | |
5379a5c5 JH |
1235 | static unsigned *done_pbase_paths; |
1236 | static int done_pbase_paths_num; | |
1237 | static int done_pbase_paths_alloc; | |
1238 | static int done_pbase_path_pos(unsigned hash) | |
1239 | { | |
1240 | int lo = 0; | |
1241 | int hi = done_pbase_paths_num; | |
1242 | while (lo < hi) { | |
1243 | int mi = (hi + lo) / 2; | |
1244 | if (done_pbase_paths[mi] == hash) | |
1245 | return mi; | |
1246 | if (done_pbase_paths[mi] < hash) | |
1247 | hi = mi; | |
1248 | else | |
1249 | lo = mi + 1; | |
1250 | } | |
1251 | return -lo-1; | |
1252 | } | |
1253 | ||
1254 | static int check_pbase_path(unsigned hash) | |
1255 | { | |
1256 | int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash); | |
1257 | if (0 <= pos) | |
1258 | return 1; | |
1259 | pos = -pos - 1; | |
25e19407 DD |
1260 | ALLOC_GROW(done_pbase_paths, |
1261 | done_pbase_paths_num + 1, | |
1262 | done_pbase_paths_alloc); | |
5379a5c5 JH |
1263 | done_pbase_paths_num++; |
1264 | if (pos < done_pbase_paths_num) | |
1265 | memmove(done_pbase_paths + pos + 1, | |
1266 | done_pbase_paths + pos, | |
1267 | (done_pbase_paths_num - pos - 1) * sizeof(unsigned)); | |
1268 | done_pbase_paths[pos] = hash; | |
1269 | return 0; | |
1270 | } | |
1271 | ||
bc32fed5 | 1272 | static void add_preferred_base_object(const char *name) |
5379a5c5 JH |
1273 | { |
1274 | struct pbase_tree *it; | |
8a5a8d6c | 1275 | int cmplen; |
68fb36eb | 1276 | unsigned hash = pack_name_hash(name); |
5379a5c5 | 1277 | |
8a5a8d6c | 1278 | if (!num_preferred_base || check_pbase_path(hash)) |
5379a5c5 JH |
1279 | return; |
1280 | ||
8a5a8d6c | 1281 | cmplen = name_cmp_len(name); |
5379a5c5 JH |
1282 | for (it = pbase_tree; it; it = it->next) { |
1283 | if (cmplen == 0) { | |
bc32fed5 | 1284 | add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1); |
5379a5c5 JH |
1285 | } |
1286 | else { | |
1287 | struct tree_desc tree; | |
6fda5e51 | 1288 | init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size); |
ce0bd642 | 1289 | add_pbase_object(&tree, name, cmplen, name); |
7a979d99 | 1290 | } |
3f9ac8d2 | 1291 | } |
3f9ac8d2 JH |
1292 | } |
1293 | ||
7a979d99 | 1294 | static void add_preferred_base(unsigned char *sha1) |
3f9ac8d2 | 1295 | { |
5379a5c5 JH |
1296 | struct pbase_tree *it; |
1297 | void *data; | |
1298 | unsigned long size; | |
1299 | unsigned char tree_sha1[20]; | |
1d6b38cc | 1300 | |
8d1d8f83 JH |
1301 | if (window <= num_preferred_base++) |
1302 | return; | |
1303 | ||
5379a5c5 JH |
1304 | data = read_object_with_reference(sha1, tree_type, &size, tree_sha1); |
1305 | if (!data) | |
7a979d99 | 1306 | return; |
5379a5c5 JH |
1307 | |
1308 | for (it = pbase_tree; it; it = it->next) { | |
a89fccd2 | 1309 | if (!hashcmp(it->pcache.sha1, tree_sha1)) { |
5379a5c5 JH |
1310 | free(data); |
1311 | return; | |
1312 | } | |
1313 | } | |
1314 | ||
1315 | it = xcalloc(1, sizeof(*it)); | |
1316 | it->next = pbase_tree; | |
1317 | pbase_tree = it; | |
1318 | ||
e702496e | 1319 | hashcpy(it->pcache.sha1, tree_sha1); |
5379a5c5 JH |
1320 | it->pcache.tree_data = data; |
1321 | it->pcache.tree_size = size; | |
3f9ac8d2 JH |
1322 | } |
1323 | ||
0ef95f72 NP |
1324 | static void cleanup_preferred_base(void) |
1325 | { | |
1326 | struct pbase_tree *it; | |
1327 | unsigned i; | |
1328 | ||
1329 | it = pbase_tree; | |
1330 | pbase_tree = NULL; | |
1331 | while (it) { | |
1332 | struct pbase_tree *this = it; | |
1333 | it = this->next; | |
1334 | free(this->pcache.tree_data); | |
1335 | free(this); | |
1336 | } | |
1337 | ||
1338 | for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) { | |
1339 | if (!pbase_tree_cache[i]) | |
1340 | continue; | |
1341 | free(pbase_tree_cache[i]->tree_data); | |
1342 | free(pbase_tree_cache[i]); | |
1343 | pbase_tree_cache[i] = NULL; | |
1344 | } | |
1345 | ||
1346 | free(done_pbase_paths); | |
1347 | done_pbase_paths = NULL; | |
1348 | done_pbase_paths_num = done_pbase_paths_alloc = 0; | |
1349 | } | |
1350 | ||
c323ac7d LT |
1351 | static void check_object(struct object_entry *entry) |
1352 | { | |
5c49c116 | 1353 | if (entry->in_pack) { |
780e6e73 | 1354 | struct packed_git *p = entry->in_pack; |
03e79c88 | 1355 | struct pack_window *w_curs = NULL; |
5c49c116 NP |
1356 | const unsigned char *base_ref = NULL; |
1357 | struct object_entry *base_entry; | |
1358 | unsigned long used, used_0; | |
ef49a7a0 | 1359 | unsigned long avail; |
5c49c116 NP |
1360 | off_t ofs; |
1361 | unsigned char *buf, c; | |
780e6e73 | 1362 | |
6777a59f | 1363 | buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail); |
ab7cd7bb | 1364 | |
5c49c116 | 1365 | /* |
fa736f72 NP |
1366 | * We want in_pack_type even if we do not reuse delta |
1367 | * since non-delta representations could still be reused. | |
ab7cd7bb | 1368 | */ |
09ded04b | 1369 | used = unpack_object_header_buffer(buf, avail, |
5c49c116 NP |
1370 | &entry->in_pack_type, |
1371 | &entry->size); | |
03d66015 NP |
1372 | if (used == 0) |
1373 | goto give_up; | |
ab7cd7bb | 1374 | |
5c49c116 NP |
1375 | /* |
1376 | * Determine if this is a delta and if so whether we can | |
1377 | * reuse it or not. Otherwise let's find out as cheaply as | |
1378 | * possible what the actual type and size for this object is. | |
3f9ac8d2 | 1379 | */ |
5c49c116 NP |
1380 | switch (entry->in_pack_type) { |
1381 | default: | |
1382 | /* Not a delta hence we've already got all we need. */ | |
1383 | entry->type = entry->in_pack_type; | |
1384 | entry->in_pack_header_size = used; | |
03d66015 NP |
1385 | if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB) |
1386 | goto give_up; | |
5c49c116 NP |
1387 | unuse_pack(&w_curs); |
1388 | return; | |
1389 | case OBJ_REF_DELTA: | |
a7de7130 | 1390 | if (reuse_delta && !entry->preferred_base) |
5c49c116 NP |
1391 | base_ref = use_pack(p, &w_curs, |
1392 | entry->in_pack_offset + used, NULL); | |
1393 | entry->in_pack_header_size = used + 20; | |
1394 | break; | |
1395 | case OBJ_OFS_DELTA: | |
1396 | buf = use_pack(p, &w_curs, | |
1397 | entry->in_pack_offset + used, NULL); | |
1398 | used_0 = 0; | |
1399 | c = buf[used_0++]; | |
1400 | ofs = c & 127; | |
1401 | while (c & 128) { | |
1402 | ofs += 1; | |
03d66015 NP |
1403 | if (!ofs || MSB(ofs, 7)) { |
1404 | error("delta base offset overflow in pack for %s", | |
1405 | sha1_to_hex(entry->idx.sha1)); | |
1406 | goto give_up; | |
1407 | } | |
5c49c116 NP |
1408 | c = buf[used_0++]; |
1409 | ofs = (ofs << 7) + (c & 127); | |
780e6e73 | 1410 | } |
5c49c116 | 1411 | ofs = entry->in_pack_offset - ofs; |
03d66015 NP |
1412 | if (ofs <= 0 || ofs >= entry->in_pack_offset) { |
1413 | error("delta base offset out of bound for %s", | |
1414 | sha1_to_hex(entry->idx.sha1)); | |
1415 | goto give_up; | |
1416 | } | |
a7de7130 | 1417 | if (reuse_delta && !entry->preferred_base) { |
3449f8c4 NP |
1418 | struct revindex_entry *revidx; |
1419 | revidx = find_pack_revindex(p, ofs); | |
08698b1e NP |
1420 | if (!revidx) |
1421 | goto give_up; | |
3449f8c4 NP |
1422 | base_ref = nth_packed_object_sha1(p, revidx->nr); |
1423 | } | |
5c49c116 NP |
1424 | entry->in_pack_header_size = used + used_0; |
1425 | break; | |
780e6e73 | 1426 | } |
780e6e73 | 1427 | |
2834bc27 | 1428 | if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) { |
5c49c116 NP |
1429 | /* |
1430 | * If base_ref was set above that means we wish to | |
1431 | * reuse delta data, and we even found that base | |
1432 | * in the list of objects we want to pack. Goodie! | |
1433 | * | |
1434 | * Depth value does not matter - find_deltas() will | |
1435 | * never consider reused delta as the base object to | |
1436 | * deltify other objects against, in order to avoid | |
1437 | * circular deltas. | |
3f9ac8d2 | 1438 | */ |
780e6e73 | 1439 | entry->type = entry->in_pack_type; |
5c49c116 | 1440 | entry->delta = base_entry; |
64bd76b1 | 1441 | entry->delta_size = entry->size; |
15b4d577 JH |
1442 | entry->delta_sibling = base_entry->delta_child; |
1443 | base_entry->delta_child = entry; | |
5c49c116 NP |
1444 | unuse_pack(&w_curs); |
1445 | return; | |
1446 | } | |
ab7cd7bb | 1447 | |
5c49c116 NP |
1448 | if (entry->type) { |
1449 | /* | |
1450 | * This must be a delta and we already know what the | |
1451 | * final object type is. Let's extract the actual | |
1452 | * object size from the delta header. | |
1453 | */ | |
1454 | entry->size = get_size_from_delta(p, &w_curs, | |
1455 | entry->in_pack_offset + entry->in_pack_header_size); | |
03d66015 NP |
1456 | if (entry->size == 0) |
1457 | goto give_up; | |
5c49c116 | 1458 | unuse_pack(&w_curs); |
3f9ac8d2 JH |
1459 | return; |
1460 | } | |
5c49c116 NP |
1461 | |
1462 | /* | |
1463 | * No choice but to fall back to the recursive delta walk | |
1464 | * with sha1_object_info() to find about the object type | |
1465 | * at this point... | |
1466 | */ | |
03d66015 | 1467 | give_up: |
5c49c116 | 1468 | unuse_pack(&w_curs); |
36e4d74a | 1469 | } |
3f9ac8d2 | 1470 | |
aa7e44bf | 1471 | entry->type = sha1_object_info(entry->idx.sha1, &entry->size); |
6d6f9cdd SP |
1472 | /* |
1473 | * The error condition is checked in prepare_pack(). This is | |
1474 | * to permit a missing preferred base object to be ignored | |
1475 | * as a preferred base. Doing so can result in a larger | |
1476 | * pack file, but the transfer will still take place. | |
1477 | */ | |
3f9ac8d2 JH |
1478 | } |
1479 | ||
5c49c116 NP |
1480 | static int pack_offset_sort(const void *_a, const void *_b) |
1481 | { | |
1482 | const struct object_entry *a = *(struct object_entry **)_a; | |
1483 | const struct object_entry *b = *(struct object_entry **)_b; | |
1484 | ||
1485 | /* avoid filesystem trashing with loose objects */ | |
1486 | if (!a->in_pack && !b->in_pack) | |
aa7e44bf | 1487 | return hashcmp(a->idx.sha1, b->idx.sha1); |
5c49c116 NP |
1488 | |
1489 | if (a->in_pack < b->in_pack) | |
1490 | return -1; | |
1491 | if (a->in_pack > b->in_pack) | |
1492 | return 1; | |
1493 | return a->in_pack_offset < b->in_pack_offset ? -1 : | |
1494 | (a->in_pack_offset > b->in_pack_offset); | |
1495 | } | |
1496 | ||
c323ac7d LT |
1497 | static void get_object_details(void) |
1498 | { | |
7cadf491 | 1499 | uint32_t i; |
5c49c116 NP |
1500 | struct object_entry **sorted_by_offset; |
1501 | ||
2834bc27 VM |
1502 | sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *)); |
1503 | for (i = 0; i < to_pack.nr_objects; i++) | |
1504 | sorted_by_offset[i] = to_pack.objects + i; | |
1505 | qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort); | |
c323ac7d | 1506 | |
2834bc27 | 1507 | for (i = 0; i < to_pack.nr_objects; i++) { |
15366280 JH |
1508 | struct object_entry *entry = sorted_by_offset[i]; |
1509 | check_object(entry); | |
754980d0 | 1510 | if (big_file_threshold < entry->size) |
15366280 JH |
1511 | entry->no_try_delta = 1; |
1512 | } | |
3449f8c4 | 1513 | |
5c49c116 | 1514 | free(sorted_by_offset); |
c323ac7d LT |
1515 | } |
1516 | ||
b904166c NP |
1517 | /* |
1518 | * We search for deltas in a list sorted by type, by filename hash, and then | |
1519 | * by size, so that we see progressively smaller and smaller files. | |
1520 | * That's because we prefer deltas to be from the bigger file | |
1521 | * to the smaller -- deletes are potentially cheaper, but perhaps | |
1522 | * more importantly, the bigger file is likely the more recent | |
1523 | * one. The deepest deltas are therefore the oldest objects which are | |
1524 | * less susceptible to be accessed often. | |
1525 | */ | |
9668cf59 | 1526 | static int type_size_sort(const void *_a, const void *_b) |
c323ac7d | 1527 | { |
9668cf59 NP |
1528 | const struct object_entry *a = *(struct object_entry **)_a; |
1529 | const struct object_entry *b = *(struct object_entry **)_b; | |
1530 | ||
c323ac7d | 1531 | if (a->type > b->type) |
27225f2e | 1532 | return -1; |
b904166c | 1533 | if (a->type < b->type) |
27225f2e | 1534 | return 1; |
b904166c | 1535 | if (a->hash > b->hash) |
7a979d99 | 1536 | return -1; |
b904166c | 1537 | if (a->hash < b->hash) |
7a979d99 | 1538 | return 1; |
b904166c | 1539 | if (a->preferred_base > b->preferred_base) |
c323ac7d | 1540 | return -1; |
b904166c NP |
1541 | if (a->preferred_base < b->preferred_base) |
1542 | return 1; | |
c323ac7d | 1543 | if (a->size > b->size) |
b904166c NP |
1544 | return -1; |
1545 | if (a->size < b->size) | |
c323ac7d | 1546 | return 1; |
b904166c | 1547 | return a < b ? -1 : (a > b); /* newest first */ |
c323ac7d LT |
1548 | } |
1549 | ||
1550 | struct unpacked { | |
1551 | struct object_entry *entry; | |
1552 | void *data; | |
f6c7081a | 1553 | struct delta_index *index; |
5a235b5e | 1554 | unsigned depth; |
c323ac7d LT |
1555 | }; |
1556 | ||
d250626c NP |
1557 | static int delta_cacheable(unsigned long src_size, unsigned long trg_size, |
1558 | unsigned long delta_size) | |
074b2eea MK |
1559 | { |
1560 | if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size) | |
1561 | return 0; | |
1562 | ||
e3dfddb3 MK |
1563 | if (delta_size < cache_max_small_delta_size) |
1564 | return 1; | |
1565 | ||
074b2eea MK |
1566 | /* cache delta, if objects are large enough compared to delta size */ |
1567 | if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10)) | |
1568 | return 1; | |
1569 | ||
1570 | return 0; | |
1571 | } | |
1572 | ||
7eb151d6 | 1573 | #ifndef NO_PTHREADS |
8ecce684 | 1574 | |
44626dc7 | 1575 | static pthread_mutex_t read_mutex; |
8ecce684 NP |
1576 | #define read_lock() pthread_mutex_lock(&read_mutex) |
1577 | #define read_unlock() pthread_mutex_unlock(&read_mutex) | |
1578 | ||
44626dc7 | 1579 | static pthread_mutex_t cache_mutex; |
3c701839 NP |
1580 | #define cache_lock() pthread_mutex_lock(&cache_mutex) |
1581 | #define cache_unlock() pthread_mutex_unlock(&cache_mutex) | |
1582 | ||
44626dc7 | 1583 | static pthread_mutex_t progress_mutex; |
8ecce684 NP |
1584 | #define progress_lock() pthread_mutex_lock(&progress_mutex) |
1585 | #define progress_unlock() pthread_mutex_unlock(&progress_mutex) | |
1586 | ||
1587 | #else | |
1588 | ||
e1ef8673 JH |
1589 | #define read_lock() (void)0 |
1590 | #define read_unlock() (void)0 | |
1591 | #define cache_lock() (void)0 | |
1592 | #define cache_unlock() (void)0 | |
1593 | #define progress_lock() (void)0 | |
1594 | #define progress_unlock() (void)0 | |
8ecce684 NP |
1595 | |
1596 | #endif | |
1597 | ||
f6c7081a | 1598 | static int try_delta(struct unpacked *trg, struct unpacked *src, |
ef0316fc | 1599 | unsigned max_depth, unsigned long *mem_usage) |
c323ac7d | 1600 | { |
f6c7081a NP |
1601 | struct object_entry *trg_entry = trg->entry; |
1602 | struct object_entry *src_entry = src->entry; | |
560b25a8 | 1603 | unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz; |
c83f032e | 1604 | unsigned ref_depth; |
21666f1a | 1605 | enum object_type type; |
c323ac7d LT |
1606 | void *delta_buf; |
1607 | ||
1608 | /* Don't bother doing diffs between different types */ | |
f6c7081a | 1609 | if (trg_entry->type != src_entry->type) |
c323ac7d LT |
1610 | return -1; |
1611 | ||
51d1e83f | 1612 | /* |
15f07e06 JK |
1613 | * We do not bother to try a delta that we discarded on an |
1614 | * earlier try, but only when reusing delta data. Note that | |
1615 | * src_entry that is marked as the preferred_base should always | |
1616 | * be considered, as even if we produce a suboptimal delta against | |
1617 | * it, we will still save the transfer cost, as we already know | |
1618 | * the other side has it and we won't send src_entry at all. | |
51d1e83f | 1619 | */ |
a7de7130 | 1620 | if (reuse_delta && trg_entry->in_pack && |
e9195b58 | 1621 | trg_entry->in_pack == src_entry->in_pack && |
15f07e06 | 1622 | !src_entry->preferred_base && |
e9195b58 JH |
1623 | trg_entry->in_pack_type != OBJ_REF_DELTA && |
1624 | trg_entry->in_pack_type != OBJ_OFS_DELTA) | |
51d1e83f LT |
1625 | return 0; |
1626 | ||
898b14ce | 1627 | /* Let's not bust the allowed depth. */ |
5a235b5e | 1628 | if (src->depth >= max_depth) |
d116a45a | 1629 | return 0; |
c323ac7d | 1630 | |
c3b06a69 | 1631 | /* Now some size filtering heuristics. */ |
560b25a8 | 1632 | trg_size = trg_entry->size; |
c83f032e NP |
1633 | if (!trg_entry->delta) { |
1634 | max_size = trg_size/2 - 20; | |
1635 | ref_depth = 1; | |
1636 | } else { | |
1637 | max_size = trg_entry->delta_size; | |
5a235b5e | 1638 | ref_depth = trg->depth; |
c83f032e | 1639 | } |
720fe22d | 1640 | max_size = (uint64_t)max_size * (max_depth - src->depth) / |
c83f032e | 1641 | (max_depth - ref_depth + 1); |
c3b06a69 NP |
1642 | if (max_size == 0) |
1643 | return 0; | |
f6c7081a | 1644 | src_size = src_entry->size; |
560b25a8 | 1645 | sizediff = src_size < trg_size ? trg_size - src_size : 0; |
27225f2e | 1646 | if (sizediff >= max_size) |
f527cb8c | 1647 | return 0; |
a1dab41a BD |
1648 | if (trg_size < src_size / 32) |
1649 | return 0; | |
f6c7081a | 1650 | |
560b25a8 NP |
1651 | /* Load data if not already done */ |
1652 | if (!trg->data) { | |
8ecce684 | 1653 | read_lock(); |
aa7e44bf | 1654 | trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz); |
8ecce684 | 1655 | read_unlock(); |
2e3404c3 JH |
1656 | if (!trg->data) |
1657 | die("object %s cannot be read", | |
1658 | sha1_to_hex(trg_entry->idx.sha1)); | |
560b25a8 NP |
1659 | if (sz != trg_size) |
1660 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1661 | sha1_to_hex(trg_entry->idx.sha1), sz, trg_size); |
ef0316fc | 1662 | *mem_usage += sz; |
560b25a8 NP |
1663 | } |
1664 | if (!src->data) { | |
8ecce684 | 1665 | read_lock(); |
aa7e44bf | 1666 | src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz); |
8ecce684 | 1667 | read_unlock(); |
71064a95 NP |
1668 | if (!src->data) { |
1669 | if (src_entry->preferred_base) { | |
1670 | static int warned = 0; | |
1671 | if (!warned++) | |
1672 | warning("object %s cannot be read", | |
1673 | sha1_to_hex(src_entry->idx.sha1)); | |
1674 | /* | |
1675 | * Those objects are not included in the | |
1676 | * resulting pack. Be resilient and ignore | |
1677 | * them if they can't be read, in case the | |
1678 | * pack could be created nevertheless. | |
1679 | */ | |
1680 | return 0; | |
1681 | } | |
2e3404c3 JH |
1682 | die("object %s cannot be read", |
1683 | sha1_to_hex(src_entry->idx.sha1)); | |
71064a95 | 1684 | } |
560b25a8 NP |
1685 | if (sz != src_size) |
1686 | die("object %s inconsistent object length (%lu vs %lu)", | |
aa7e44bf | 1687 | sha1_to_hex(src_entry->idx.sha1), sz, src_size); |
ef0316fc | 1688 | *mem_usage += sz; |
560b25a8 NP |
1689 | } |
1690 | if (!src->index) { | |
1691 | src->index = create_delta_index(src->data, src_size); | |
a588d88a MK |
1692 | if (!src->index) { |
1693 | static int warned = 0; | |
1694 | if (!warned++) | |
1695 | warning("suboptimal pack - out of memory"); | |
1696 | return 0; | |
1697 | } | |
ef0316fc | 1698 | *mem_usage += sizeof_delta_index(src->index); |
560b25a8 NP |
1699 | } |
1700 | ||
1701 | delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size); | |
c323ac7d | 1702 | if (!delta_buf) |
75c42d8c | 1703 | return 0; |
f6c7081a | 1704 | |
9e2d57a0 | 1705 | if (trg_entry->delta) { |
848d732c BD |
1706 | /* Prefer only shallower same-sized deltas. */ |
1707 | if (delta_size == trg_entry->delta_size && | |
5a235b5e | 1708 | src->depth + 1 >= trg->depth) { |
848d732c BD |
1709 | free(delta_buf); |
1710 | return 0; | |
1711 | } | |
074b2eea | 1712 | } |
9e2d57a0 | 1713 | |
3c701839 NP |
1714 | /* |
1715 | * Handle memory allocation outside of the cache | |
1716 | * accounting lock. Compiler will optimize the strangeness | |
7eb151d6 | 1717 | * away when NO_PTHREADS is defined. |
3c701839 | 1718 | */ |
8e0f7003 | 1719 | free(trg_entry->delta_data); |
3c701839 | 1720 | cache_lock(); |
9e2d57a0 NP |
1721 | if (trg_entry->delta_data) { |
1722 | delta_cache_size -= trg_entry->delta_size; | |
9e2d57a0 NP |
1723 | trg_entry->delta_data = NULL; |
1724 | } | |
d250626c | 1725 | if (delta_cacheable(src_size, trg_size, delta_size)) { |
b7a28f78 | 1726 | delta_cache_size += delta_size; |
3c701839 NP |
1727 | cache_unlock(); |
1728 | trg_entry->delta_data = xrealloc(delta_buf, delta_size); | |
1729 | } else { | |
1730 | cache_unlock(); | |
074b2eea | 1731 | free(delta_buf); |
3c701839 NP |
1732 | } |
1733 | ||
b7a28f78 NP |
1734 | trg_entry->delta = src_entry; |
1735 | trg_entry->delta_size = delta_size; | |
1736 | trg->depth = src->depth + 1; | |
1737 | ||
f6c7081a | 1738 | return 1; |
c323ac7d LT |
1739 | } |
1740 | ||
898b14ce | 1741 | static unsigned int check_delta_limit(struct object_entry *me, unsigned int n) |
b2504a0d | 1742 | { |
898b14ce NP |
1743 | struct object_entry *child = me->delta_child; |
1744 | unsigned int m = n; | |
1745 | while (child) { | |
1746 | unsigned int c = check_delta_limit(child, n + 1); | |
1747 | if (m < c) | |
1748 | m = c; | |
1749 | child = child->delta_sibling; | |
1750 | } | |
1751 | return m; | |
b2504a0d NP |
1752 | } |
1753 | ||
75ad235c | 1754 | static unsigned long free_unpacked(struct unpacked *n) |
a97773ce | 1755 | { |
ef0316fc | 1756 | unsigned long freed_mem = sizeof_delta_index(n->index); |
a97773ce BD |
1757 | free_delta_index(n->index); |
1758 | n->index = NULL; | |
1759 | if (n->data) { | |
ef0316fc | 1760 | freed_mem += n->entry->size; |
a97773ce BD |
1761 | free(n->data); |
1762 | n->data = NULL; | |
a97773ce BD |
1763 | } |
1764 | n->entry = NULL; | |
7d7baa5e | 1765 | n->depth = 0; |
ef0316fc | 1766 | return freed_mem; |
a97773ce BD |
1767 | } |
1768 | ||
384b32c0 | 1769 | static void find_deltas(struct object_entry **list, unsigned *list_size, |
e334977d | 1770 | int window, int depth, unsigned *processed) |
c323ac7d | 1771 | { |
384b32c0 | 1772 | uint32_t i, idx = 0, count = 0; |
7cadf491 | 1773 | struct unpacked *array; |
ef0316fc | 1774 | unsigned long mem_usage = 0; |
c323ac7d | 1775 | |
19d4b416 | 1776 | array = xcalloc(window, sizeof(struct unpacked)); |
21fcd1bd | 1777 | |
384b32c0 | 1778 | for (;;) { |
421b488a | 1779 | struct object_entry *entry; |
c323ac7d | 1780 | struct unpacked *n = array + idx; |
ef0316fc | 1781 | int j, max_depth, best_base = -1; |
c323ac7d | 1782 | |
384b32c0 NP |
1783 | progress_lock(); |
1784 | if (!*list_size) { | |
1785 | progress_unlock(); | |
1786 | break; | |
1787 | } | |
421b488a | 1788 | entry = *list++; |
384b32c0 NP |
1789 | (*list_size)--; |
1790 | if (!entry->preferred_base) { | |
1791 | (*processed)++; | |
1792 | display_progress(progress_state, *processed); | |
1793 | } | |
1794 | progress_unlock(); | |
1795 | ||
ef0316fc | 1796 | mem_usage -= free_unpacked(n); |
c323ac7d | 1797 | n->entry = entry; |
ab7cd7bb | 1798 | |
a97773ce | 1799 | while (window_memory_limit && |
ef0316fc | 1800 | mem_usage > window_memory_limit && |
a97773ce BD |
1801 | count > 1) { |
1802 | uint32_t tail = (idx + window - count) % window; | |
75ad235c | 1803 | mem_usage -= free_unpacked(array + tail); |
a97773ce BD |
1804 | count--; |
1805 | } | |
1806 | ||
75d39853 NP |
1807 | /* We do not compute delta to *create* objects we are not |
1808 | * going to pack. | |
1809 | */ | |
1810 | if (entry->preferred_base) | |
1811 | goto next; | |
1812 | ||
898b14ce NP |
1813 | /* |
1814 | * If the current object is at pack edge, take the depth the | |
1815 | * objects that depend on the current object into account | |
1816 | * otherwise they would become too deep. | |
1817 | */ | |
1818 | max_depth = depth; | |
1819 | if (entry->delta_child) { | |
1820 | max_depth -= check_delta_limit(entry, 0); | |
1821 | if (max_depth <= 0) | |
1822 | goto next; | |
1823 | } | |
1824 | ||
78817c15 LT |
1825 | j = window; |
1826 | while (--j > 0) { | |
77639870 | 1827 | int ret; |
7cadf491 | 1828 | uint32_t other_idx = idx + j; |
c323ac7d | 1829 | struct unpacked *m; |
78817c15 LT |
1830 | if (other_idx >= window) |
1831 | other_idx -= window; | |
c323ac7d LT |
1832 | m = array + other_idx; |
1833 | if (!m->entry) | |
1834 | break; | |
ef0316fc | 1835 | ret = try_delta(n, m, max_depth, &mem_usage); |
77639870 | 1836 | if (ret < 0) |
c323ac7d | 1837 | break; |
77639870 JH |
1838 | else if (ret > 0) |
1839 | best_base = other_idx; | |
c323ac7d | 1840 | } |
898b14ce | 1841 | |
ed4a9031 NP |
1842 | /* |
1843 | * If we decided to cache the delta data, then it is best | |
1844 | * to compress it right away. First because we have to do | |
1845 | * it anyway, and doing it here while we're threaded will | |
1846 | * save a lot of time in the non threaded write phase, | |
1847 | * as well as allow for caching more deltas within | |
1848 | * the same cache size limit. | |
1849 | * ... | |
1850 | * But only if not writing to stdout, since in that case | |
1851 | * the network is most likely throttling writes anyway, | |
1852 | * and therefore it is best to go to the write phase ASAP | |
1853 | * instead, as we can afford spending more time compressing | |
1854 | * between writes at that moment. | |
1855 | */ | |
1856 | if (entry->delta_data && !pack_to_stdout) { | |
1857 | entry->z_delta_size = do_compress(&entry->delta_data, | |
1858 | entry->delta_size); | |
1859 | cache_lock(); | |
1860 | delta_cache_size -= entry->delta_size; | |
1861 | delta_cache_size += entry->z_delta_size; | |
1862 | cache_unlock(); | |
1863 | } | |
1864 | ||
70ca1a3f JH |
1865 | /* if we made n a delta, and if n is already at max |
1866 | * depth, leaving it in the window is pointless. we | |
1867 | * should evict it first. | |
70ca1a3f | 1868 | */ |
70baf5d4 | 1869 | if (entry->delta && max_depth <= n->depth) |
70ca1a3f | 1870 | continue; |
ff45715c | 1871 | |
77639870 JH |
1872 | /* |
1873 | * Move the best delta base up in the window, after the | |
1874 | * currently deltified object, to keep it longer. It will | |
1875 | * be the first base object to be attempted next. | |
1876 | */ | |
1877 | if (entry->delta) { | |
1878 | struct unpacked swap = array[best_base]; | |
1879 | int dist = (window + idx - best_base) % window; | |
1880 | int dst = best_base; | |
1881 | while (dist--) { | |
1882 | int src = (dst + 1) % window; | |
1883 | array[dst] = array[src]; | |
1884 | dst = src; | |
1885 | } | |
1886 | array[dst] = swap; | |
1887 | } | |
1888 | ||
898b14ce | 1889 | next: |
521a4f4c | 1890 | idx++; |
a97773ce BD |
1891 | if (count + 1 < window) |
1892 | count++; | |
521a4f4c LT |
1893 | if (idx >= window) |
1894 | idx = 0; | |
384b32c0 | 1895 | } |
adee7bdf | 1896 | |
f6c7081a | 1897 | for (i = 0; i < window; ++i) { |
ff45715c | 1898 | free_delta_index(array[i].index); |
adee7bdf | 1899 | free(array[i].data); |
f6c7081a | 1900 | } |
adee7bdf | 1901 | free(array); |
c323ac7d LT |
1902 | } |
1903 | ||
7eb151d6 | 1904 | #ifndef NO_PTHREADS |
8ecce684 | 1905 | |
a9a74636 NP |
1906 | static void try_to_free_from_threads(size_t size) |
1907 | { | |
1908 | read_lock(); | |
7c3ecb32 | 1909 | release_pack_memory(size); |
a9a74636 NP |
1910 | read_unlock(); |
1911 | } | |
1912 | ||
bc9b2175 | 1913 | static try_to_free_t old_try_to_free_routine; |
851c34b0 | 1914 | |
50f22ada JS |
1915 | /* |
1916 | * The main thread waits on the condition that (at least) one of the workers | |
1917 | * has stopped working (which is indicated in the .working member of | |
1918 | * struct thread_params). | |
1919 | * When a work thread has completed its work, it sets .working to 0 and | |
1920 | * signals the main thread and waits on the condition that .data_ready | |
1921 | * becomes 1. | |
1922 | */ | |
1923 | ||
8ecce684 NP |
1924 | struct thread_params { |
1925 | pthread_t thread; | |
1926 | struct object_entry **list; | |
1927 | unsigned list_size; | |
384b32c0 | 1928 | unsigned remaining; |
8ecce684 NP |
1929 | int window; |
1930 | int depth; | |
50f22ada JS |
1931 | int working; |
1932 | int data_ready; | |
1933 | pthread_mutex_t mutex; | |
1934 | pthread_cond_t cond; | |
8ecce684 NP |
1935 | unsigned *processed; |
1936 | }; | |
1937 | ||
44626dc7 AH |
1938 | static pthread_cond_t progress_cond; |
1939 | ||
1940 | /* | |
1941 | * Mutex and conditional variable can't be statically-initialized on Windows. | |
1942 | */ | |
1943 | static void init_threaded_search(void) | |
1944 | { | |
93749194 | 1945 | init_recursive_mutex(&read_mutex); |
44626dc7 AH |
1946 | pthread_mutex_init(&cache_mutex, NULL); |
1947 | pthread_mutex_init(&progress_mutex, NULL); | |
1948 | pthread_cond_init(&progress_cond, NULL); | |
851c34b0 | 1949 | old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads); |
44626dc7 AH |
1950 | } |
1951 | ||
1952 | static void cleanup_threaded_search(void) | |
1953 | { | |
851c34b0 | 1954 | set_try_to_free_routine(old_try_to_free_routine); |
44626dc7 AH |
1955 | pthread_cond_destroy(&progress_cond); |
1956 | pthread_mutex_destroy(&read_mutex); | |
1957 | pthread_mutex_destroy(&cache_mutex); | |
1958 | pthread_mutex_destroy(&progress_mutex); | |
1959 | } | |
c2a33679 | 1960 | |
8ecce684 NP |
1961 | static void *threaded_find_deltas(void *arg) |
1962 | { | |
c2a33679 NP |
1963 | struct thread_params *me = arg; |
1964 | ||
50f22ada | 1965 | while (me->remaining) { |
384b32c0 | 1966 | find_deltas(me->list, &me->remaining, |
c2a33679 | 1967 | me->window, me->depth, me->processed); |
50f22ada JS |
1968 | |
1969 | progress_lock(); | |
1970 | me->working = 0; | |
1971 | pthread_cond_signal(&progress_cond); | |
1972 | progress_unlock(); | |
1973 | ||
1974 | /* | |
1975 | * We must not set ->data_ready before we wait on the | |
1976 | * condition because the main thread may have set it to 1 | |
1977 | * before we get here. In order to be sure that new | |
1978 | * work is available if we see 1 in ->data_ready, it | |
1979 | * was initialized to 0 before this thread was spawned | |
1980 | * and we reset it to 0 right away. | |
1981 | */ | |
1982 | pthread_mutex_lock(&me->mutex); | |
1983 | while (!me->data_ready) | |
1984 | pthread_cond_wait(&me->cond, &me->mutex); | |
1985 | me->data_ready = 0; | |
1986 | pthread_mutex_unlock(&me->mutex); | |
c2a33679 | 1987 | } |
50f22ada JS |
1988 | /* leave ->working 1 so that this doesn't get more work assigned */ |
1989 | return NULL; | |
8ecce684 NP |
1990 | } |
1991 | ||
8ecce684 NP |
1992 | static void ll_find_deltas(struct object_entry **list, unsigned list_size, |
1993 | int window, int depth, unsigned *processed) | |
1994 | { | |
dcda3614 | 1995 | struct thread_params *p; |
384b32c0 | 1996 | int i, ret, active_threads = 0; |
c2a33679 | 1997 | |
44626dc7 AH |
1998 | init_threaded_search(); |
1999 | ||
367f4a43 | 2000 | if (delta_search_threads <= 1) { |
384b32c0 | 2001 | find_deltas(list, &list_size, window, depth, processed); |
44626dc7 | 2002 | cleanup_threaded_search(); |
367f4a43 NP |
2003 | return; |
2004 | } | |
43cc2b42 | 2005 | if (progress > pack_to_stdout) |
b6c29915 | 2006 | fprintf(stderr, "Delta compression using up to %d threads.\n", |
43cc2b42 | 2007 | delta_search_threads); |
dcda3614 | 2008 | p = xcalloc(delta_search_threads, sizeof(*p)); |
367f4a43 | 2009 | |
50f22ada | 2010 | /* Partition the work amongst work threads. */ |
367f4a43 | 2011 | for (i = 0; i < delta_search_threads; i++) { |
50f22ada JS |
2012 | unsigned sub_size = list_size / (delta_search_threads - i); |
2013 | ||
bf874896 NP |
2014 | /* don't use too small segments or no deltas will be found */ |
2015 | if (sub_size < 2*window && i+1 < delta_search_threads) | |
2016 | sub_size = 0; | |
2017 | ||
8ecce684 NP |
2018 | p[i].window = window; |
2019 | p[i].depth = depth; | |
2020 | p[i].processed = processed; | |
50f22ada JS |
2021 | p[i].working = 1; |
2022 | p[i].data_ready = 0; | |
c2a33679 | 2023 | |
59921b4b | 2024 | /* try to split chunks on "path" boundaries */ |
6fc74703 NP |
2025 | while (sub_size && sub_size < list_size && |
2026 | list[sub_size]->hash && | |
384b32c0 NP |
2027 | list[sub_size]->hash == list[sub_size-1]->hash) |
2028 | sub_size++; | |
2029 | ||
50f22ada JS |
2030 | p[i].list = list; |
2031 | p[i].list_size = sub_size; | |
2032 | p[i].remaining = sub_size; | |
59921b4b | 2033 | |
384b32c0 NP |
2034 | list += sub_size; |
2035 | list_size -= sub_size; | |
2036 | } | |
2037 | ||
50f22ada JS |
2038 | /* Start work threads. */ |
2039 | for (i = 0; i < delta_search_threads; i++) { | |
2040 | if (!p[i].list_size) | |
2041 | continue; | |
68e6a4f8 JS |
2042 | pthread_mutex_init(&p[i].mutex, NULL); |
2043 | pthread_cond_init(&p[i].cond, NULL); | |
50f22ada JS |
2044 | ret = pthread_create(&p[i].thread, NULL, |
2045 | threaded_find_deltas, &p[i]); | |
2046 | if (ret) | |
2047 | die("unable to create thread: %s", strerror(ret)); | |
2048 | active_threads++; | |
2049 | } | |
2050 | ||
384b32c0 NP |
2051 | /* |
2052 | * Now let's wait for work completion. Each time a thread is done | |
2053 | * with its work, we steal half of the remaining work from the | |
2054 | * thread with the largest number of unprocessed objects and give | |
2055 | * it to that newly idle thread. This ensure good load balancing | |
2056 | * until the remaining object list segments are simply too short | |
2057 | * to be worth splitting anymore. | |
2058 | */ | |
50f22ada JS |
2059 | while (active_threads) { |
2060 | struct thread_params *target = NULL; | |
384b32c0 NP |
2061 | struct thread_params *victim = NULL; |
2062 | unsigned sub_size = 0; | |
384b32c0 NP |
2063 | |
2064 | progress_lock(); | |
50f22ada JS |
2065 | for (;;) { |
2066 | for (i = 0; !target && i < delta_search_threads; i++) | |
2067 | if (!p[i].working) | |
2068 | target = &p[i]; | |
2069 | if (target) | |
2070 | break; | |
2071 | pthread_cond_wait(&progress_cond, &progress_mutex); | |
2072 | } | |
2073 | ||
384b32c0 NP |
2074 | for (i = 0; i < delta_search_threads; i++) |
2075 | if (p[i].remaining > 2*window && | |
2076 | (!victim || victim->remaining < p[i].remaining)) | |
2077 | victim = &p[i]; | |
2078 | if (victim) { | |
2079 | sub_size = victim->remaining / 2; | |
2080 | list = victim->list + victim->list_size - sub_size; | |
2081 | while (sub_size && list[0]->hash && | |
2082 | list[0]->hash == list[-1]->hash) { | |
2083 | list++; | |
2084 | sub_size--; | |
2085 | } | |
eb9688ff NP |
2086 | if (!sub_size) { |
2087 | /* | |
2088 | * It is possible for some "paths" to have | |
2089 | * so many objects that no hash boundary | |
2090 | * might be found. Let's just steal the | |
2091 | * exact half in that case. | |
2092 | */ | |
2093 | sub_size = victim->remaining / 2; | |
2094 | list -= sub_size; | |
2095 | } | |
384b32c0 NP |
2096 | target->list = list; |
2097 | victim->list_size -= sub_size; | |
2098 | victim->remaining -= sub_size; | |
2099 | } | |
384b32c0 NP |
2100 | target->list_size = sub_size; |
2101 | target->remaining = sub_size; | |
50f22ada JS |
2102 | target->working = 1; |
2103 | progress_unlock(); | |
2104 | ||
2105 | pthread_mutex_lock(&target->mutex); | |
2106 | target->data_ready = 1; | |
2107 | pthread_cond_signal(&target->cond); | |
2108 | pthread_mutex_unlock(&target->mutex); | |
c2a33679 | 2109 | |
384b32c0 | 2110 | if (!sub_size) { |
b81d9af7 | 2111 | pthread_join(target->thread, NULL); |
50f22ada JS |
2112 | pthread_cond_destroy(&target->cond); |
2113 | pthread_mutex_destroy(&target->mutex); | |
384b32c0 | 2114 | active_threads--; |
c2a33679 | 2115 | } |
50f22ada | 2116 | } |
44626dc7 | 2117 | cleanup_threaded_search(); |
dcda3614 | 2118 | free(p); |
8ecce684 NP |
2119 | } |
2120 | ||
2121 | #else | |
384b32c0 | 2122 | #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p) |
8ecce684 NP |
2123 | #endif |
2124 | ||
d155254c | 2125 | static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data) |
f0a24aa5 | 2126 | { |
d155254c | 2127 | struct object_id peeled; |
f0a24aa5 | 2128 | |
59556548 | 2129 | if (starts_with(path, "refs/tags/") && /* is a tag? */ |
d155254c MH |
2130 | !peel_ref(path, peeled.hash) && /* peelable? */ |
2131 | packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */ | |
2132 | add_object_entry(oid->hash, OBJ_TAG, NULL, 0); | |
f0a24aa5 SP |
2133 | return 0; |
2134 | } | |
2135 | ||
f3123c4a JH |
2136 | static void prepare_pack(int window, int depth) |
2137 | { | |
9668cf59 | 2138 | struct object_entry **delta_list; |
6e1c2344 RJ |
2139 | uint32_t i, nr_deltas; |
2140 | unsigned n; | |
9668cf59 | 2141 | |
3f9ac8d2 | 2142 | get_object_details(); |
9668cf59 | 2143 | |
0e8189e2 NP |
2144 | /* |
2145 | * If we're locally repacking then we need to be doubly careful | |
2146 | * from now on in order to make sure no stealth corruption gets | |
2147 | * propagated to the new pack. Clients receiving streamed packs | |
2148 | * should validate everything they get anyway so no need to incur | |
2149 | * the additional cost here in that case. | |
2150 | */ | |
2151 | if (!pack_to_stdout) | |
2152 | do_check_packed_object_crc = 1; | |
2153 | ||
2834bc27 | 2154 | if (!to_pack.nr_objects || !window || !depth) |
9668cf59 NP |
2155 | return; |
2156 | ||
b32fa95f | 2157 | ALLOC_ARRAY(delta_list, to_pack.nr_objects); |
75d39853 NP |
2158 | nr_deltas = n = 0; |
2159 | ||
2834bc27 VM |
2160 | for (i = 0; i < to_pack.nr_objects; i++) { |
2161 | struct object_entry *entry = to_pack.objects + i; | |
75d39853 NP |
2162 | |
2163 | if (entry->delta) | |
2164 | /* This happens if we decided to reuse existing | |
a7de7130 | 2165 | * delta from a pack. "reuse_delta &&" is implied. |
75d39853 NP |
2166 | */ |
2167 | continue; | |
2168 | ||
2169 | if (entry->size < 50) | |
2170 | continue; | |
2171 | ||
2172 | if (entry->no_try_delta) | |
2173 | continue; | |
2174 | ||
6d6f9cdd | 2175 | if (!entry->preferred_base) { |
75d39853 | 2176 | nr_deltas++; |
6d6f9cdd SP |
2177 | if (entry->type < 0) |
2178 | die("unable to get type of object %s", | |
2179 | sha1_to_hex(entry->idx.sha1)); | |
eede9f42 NP |
2180 | } else { |
2181 | if (entry->type < 0) { | |
2182 | /* | |
2183 | * This object is not found, but we | |
2184 | * don't have to include it anyway. | |
2185 | */ | |
2186 | continue; | |
2187 | } | |
6d6f9cdd | 2188 | } |
75d39853 NP |
2189 | |
2190 | delta_list[n++] = entry; | |
2191 | } | |
2192 | ||
2f8b8947 | 2193 | if (nr_deltas && n > 1) { |
e334977d NP |
2194 | unsigned nr_done = 0; |
2195 | if (progress) | |
754dbc43 | 2196 | progress_state = start_progress(_("Compressing objects"), |
dc6a0757 | 2197 | nr_deltas); |
75d39853 | 2198 | qsort(delta_list, n, sizeof(*delta_list), type_size_sort); |
8ecce684 | 2199 | ll_find_deltas(delta_list, n, window+1, depth, &nr_done); |
4d4fcc54 | 2200 | stop_progress(&progress_state); |
e334977d NP |
2201 | if (nr_done != nr_deltas) |
2202 | die("inconsistency with delta count"); | |
75d39853 | 2203 | } |
9668cf59 | 2204 | free(delta_list); |
f3123c4a JH |
2205 | } |
2206 | ||
ef90d6d4 | 2207 | static int git_pack_config(const char *k, const char *v, void *cb) |
4812a93a | 2208 | { |
eeefa7c9 | 2209 | if (!strcmp(k, "pack.window")) { |
4812a93a JK |
2210 | window = git_config_int(k, v); |
2211 | return 0; | |
2212 | } | |
a97773ce BD |
2213 | if (!strcmp(k, "pack.windowmemory")) { |
2214 | window_memory_limit = git_config_ulong(k, v); | |
2215 | return 0; | |
2216 | } | |
2217 | if (!strcmp(k, "pack.depth")) { | |
842aaf93 TT |
2218 | depth = git_config_int(k, v); |
2219 | return 0; | |
2220 | } | |
960ccca6 DH |
2221 | if (!strcmp(k, "pack.compression")) { |
2222 | int level = git_config_int(k, v); | |
2223 | if (level == -1) | |
2224 | level = Z_DEFAULT_COMPRESSION; | |
2225 | else if (level < 0 || level > Z_BEST_COMPRESSION) | |
2226 | die("bad pack compression level %d", level); | |
2227 | pack_compression_level = level; | |
2228 | pack_compression_seen = 1; | |
2229 | return 0; | |
2230 | } | |
074b2eea MK |
2231 | if (!strcmp(k, "pack.deltacachesize")) { |
2232 | max_delta_cache_size = git_config_int(k, v); | |
2233 | return 0; | |
2234 | } | |
e3dfddb3 MK |
2235 | if (!strcmp(k, "pack.deltacachelimit")) { |
2236 | cache_max_small_delta_size = git_config_int(k, v); | |
2237 | return 0; | |
2238 | } | |
ae4f07fb VM |
2239 | if (!strcmp(k, "pack.writebitmaphashcache")) { |
2240 | if (git_config_bool(k, v)) | |
2241 | write_bitmap_options |= BITMAP_OPT_HASH_CACHE; | |
2242 | else | |
2243 | write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE; | |
2244 | } | |
6b8fda2d VM |
2245 | if (!strcmp(k, "pack.usebitmaps")) { |
2246 | use_bitmap_index = git_config_bool(k, v); | |
2247 | return 0; | |
2248 | } | |
693b86ff NP |
2249 | if (!strcmp(k, "pack.threads")) { |
2250 | delta_search_threads = git_config_int(k, v); | |
833e3df1 | 2251 | if (delta_search_threads < 0) |
693b86ff NP |
2252 | die("invalid number of threads specified (%d)", |
2253 | delta_search_threads); | |
7eb151d6 | 2254 | #ifdef NO_PTHREADS |
833e3df1 | 2255 | if (delta_search_threads != 1) |
693b86ff NP |
2256 | warning("no threads support, ignoring %s", k); |
2257 | #endif | |
2258 | return 0; | |
2259 | } | |
4d00bda2 | 2260 | if (!strcmp(k, "pack.indexversion")) { |
ebcfb379 JH |
2261 | pack_idx_opts.version = git_config_int(k, v); |
2262 | if (pack_idx_opts.version > 2) | |
6e1c2344 | 2263 | die("bad pack.indexversion=%"PRIu32, |
ebcfb379 | 2264 | pack_idx_opts.version); |
4d00bda2 NP |
2265 | return 0; |
2266 | } | |
ef90d6d4 | 2267 | return git_default_config(k, v, cb); |
4812a93a JK |
2268 | } |
2269 | ||
b5d97e6b | 2270 | static void read_object_list_from_stdin(void) |
c323ac7d | 2271 | { |
b5d97e6b JH |
2272 | char line[40 + 1 + PATH_MAX + 2]; |
2273 | unsigned char sha1[20]; | |
b2504a0d | 2274 | |
da93d12b | 2275 | for (;;) { |
da93d12b LT |
2276 | if (!fgets(line, sizeof(line), stdin)) { |
2277 | if (feof(stdin)) | |
2278 | break; | |
2279 | if (!ferror(stdin)) | |
2280 | die("fgets returned NULL, not EOF, not error!"); | |
687dd75c | 2281 | if (errno != EINTR) |
d824cbba | 2282 | die_errno("fgets"); |
687dd75c JH |
2283 | clearerr(stdin); |
2284 | continue; | |
da93d12b | 2285 | } |
7a979d99 JH |
2286 | if (line[0] == '-') { |
2287 | if (get_sha1_hex(line+1, sha1)) | |
2288 | die("expected edge sha1, got garbage:\n %s", | |
b5d97e6b | 2289 | line); |
8d1d8f83 | 2290 | add_preferred_base(sha1); |
7a979d99 | 2291 | continue; |
21fcd1bd | 2292 | } |
c323ac7d | 2293 | if (get_sha1_hex(line, sha1)) |
ef07618f | 2294 | die("expected sha1, got garbage:\n %s", line); |
b5d97e6b | 2295 | |
bc32fed5 JH |
2296 | add_preferred_base_object(line+41); |
2297 | add_object_entry(sha1, 0, line+41, 0); | |
c323ac7d | 2298 | } |
b5d97e6b JH |
2299 | } |
2300 | ||
08cdfb13 JH |
2301 | #define OBJECT_ADDED (1u<<20) |
2302 | ||
11c211fa | 2303 | static void show_commit(struct commit *commit, void *data) |
b5d97e6b | 2304 | { |
ed1c9977 | 2305 | add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0); |
08cdfb13 | 2306 | commit->object.flags |= OBJECT_ADDED; |
7cc8f971 VM |
2307 | |
2308 | if (write_bitmap_index) | |
2309 | index_commit_for_bitmap(commit); | |
b5d97e6b JH |
2310 | } |
2311 | ||
de1e67d0 | 2312 | static void show_object(struct object *obj, const char *name, void *data) |
b5d97e6b | 2313 | { |
8d2dfc49 | 2314 | add_preferred_base_object(name); |
ed1c9977 | 2315 | add_object_entry(obj->oid.hash, obj->type, name, 0); |
8d2dfc49 | 2316 | obj->flags |= OBJECT_ADDED; |
b5d97e6b JH |
2317 | } |
2318 | ||
8d1d8f83 JH |
2319 | static void show_edge(struct commit *commit) |
2320 | { | |
ed1c9977 | 2321 | add_preferred_base(commit->object.oid.hash); |
8d1d8f83 JH |
2322 | } |
2323 | ||
08cdfb13 JH |
2324 | struct in_pack_object { |
2325 | off_t offset; | |
2326 | struct object *object; | |
2327 | }; | |
2328 | ||
2329 | struct in_pack { | |
2330 | int alloc; | |
2331 | int nr; | |
2332 | struct in_pack_object *array; | |
2333 | }; | |
2334 | ||
2335 | static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack) | |
2336 | { | |
ed1c9977 | 2337 | in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p); |
08cdfb13 JH |
2338 | in_pack->array[in_pack->nr].object = object; |
2339 | in_pack->nr++; | |
2340 | } | |
2341 | ||
2342 | /* | |
2343 | * Compare the objects in the offset order, in order to emulate the | |
f18d244a | 2344 | * "git rev-list --objects" output that produced the pack originally. |
08cdfb13 JH |
2345 | */ |
2346 | static int ofscmp(const void *a_, const void *b_) | |
2347 | { | |
2348 | struct in_pack_object *a = (struct in_pack_object *)a_; | |
2349 | struct in_pack_object *b = (struct in_pack_object *)b_; | |
2350 | ||
2351 | if (a->offset < b->offset) | |
2352 | return -1; | |
2353 | else if (a->offset > b->offset) | |
2354 | return 1; | |
2355 | else | |
f2fd0760 | 2356 | return oidcmp(&a->object->oid, &b->object->oid); |
08cdfb13 JH |
2357 | } |
2358 | ||
2359 | static void add_objects_in_unpacked_packs(struct rev_info *revs) | |
2360 | { | |
2361 | struct packed_git *p; | |
2362 | struct in_pack in_pack; | |
2363 | uint32_t i; | |
2364 | ||
2365 | memset(&in_pack, 0, sizeof(in_pack)); | |
2366 | ||
2367 | for (p = packed_git; p; p = p->next) { | |
2368 | const unsigned char *sha1; | |
2369 | struct object *o; | |
2370 | ||
79bc4c71 | 2371 | if (!p->pack_local || p->pack_keep) |
08cdfb13 JH |
2372 | continue; |
2373 | if (open_pack_index(p)) | |
2374 | die("cannot open pack index"); | |
2375 | ||
2376 | ALLOC_GROW(in_pack.array, | |
2377 | in_pack.nr + p->num_objects, | |
2378 | in_pack.alloc); | |
2379 | ||
2380 | for (i = 0; i < p->num_objects; i++) { | |
2381 | sha1 = nth_packed_object_sha1(p, i); | |
2382 | o = lookup_unknown_object(sha1); | |
2383 | if (!(o->flags & OBJECT_ADDED)) | |
2384 | mark_in_pack_object(o, p, &in_pack); | |
2385 | o->flags |= OBJECT_ADDED; | |
2386 | } | |
2387 | } | |
2388 | ||
2389 | if (in_pack.nr) { | |
2390 | qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]), | |
2391 | ofscmp); | |
2392 | for (i = 0; i < in_pack.nr; i++) { | |
2393 | struct object *o = in_pack.array[i].object; | |
ed1c9977 | 2394 | add_object_entry(o->oid.hash, o->type, "", 0); |
08cdfb13 JH |
2395 | } |
2396 | } | |
2397 | free(in_pack.array); | |
2398 | } | |
2399 | ||
e26a8c47 JK |
2400 | static int add_loose_object(const unsigned char *sha1, const char *path, |
2401 | void *data) | |
2402 | { | |
2403 | enum object_type type = sha1_object_info(sha1, NULL); | |
2404 | ||
2405 | if (type < 0) { | |
2406 | warning("loose object at %s could not be examined", path); | |
2407 | return 0; | |
2408 | } | |
2409 | ||
2410 | add_object_entry(sha1, type, "", 0); | |
2411 | return 0; | |
2412 | } | |
2413 | ||
2414 | /* | |
2415 | * We actually don't even have to worry about reachability here. | |
2416 | * add_object_entry will weed out duplicates, so we just add every | |
2417 | * loose object we find. | |
2418 | */ | |
2419 | static void add_unreachable_loose_objects(void) | |
2420 | { | |
2421 | for_each_loose_file_in_objdir(get_object_directory(), | |
2422 | add_loose_object, | |
2423 | NULL, NULL, NULL); | |
2424 | } | |
2425 | ||
094085e3 BC |
2426 | static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1) |
2427 | { | |
2428 | static struct packed_git *last_found = (void *)1; | |
2429 | struct packed_git *p; | |
2430 | ||
2431 | p = (last_found != (void *)1) ? last_found : packed_git; | |
2432 | ||
2433 | while (p) { | |
2434 | if ((!p->pack_local || p->pack_keep) && | |
2435 | find_pack_entry_one(sha1, p)) { | |
2436 | last_found = p; | |
2437 | return 1; | |
2438 | } | |
2439 | if (p == last_found) | |
2440 | p = packed_git; | |
2441 | else | |
2442 | p = p->next; | |
2443 | if (p == last_found) | |
2444 | p = p->next; | |
2445 | } | |
2446 | return 0; | |
2447 | } | |
2448 | ||
abcb8655 JK |
2449 | /* |
2450 | * Store a list of sha1s that are should not be discarded | |
2451 | * because they are either written too recently, or are | |
2452 | * reachable from another object that was. | |
2453 | * | |
2454 | * This is filled by get_object_list. | |
2455 | */ | |
2456 | static struct sha1_array recent_objects; | |
2457 | ||
d0d46abc JK |
2458 | static int loosened_object_can_be_discarded(const unsigned char *sha1, |
2459 | unsigned long mtime) | |
2460 | { | |
2461 | if (!unpack_unreachable_expiration) | |
2462 | return 0; | |
2463 | if (mtime > unpack_unreachable_expiration) | |
2464 | return 0; | |
abcb8655 JK |
2465 | if (sha1_array_lookup(&recent_objects, sha1) >= 0) |
2466 | return 0; | |
d0d46abc JK |
2467 | return 1; |
2468 | } | |
2469 | ||
ca11b212 NP |
2470 | static void loosen_unused_packed_objects(struct rev_info *revs) |
2471 | { | |
2472 | struct packed_git *p; | |
2473 | uint32_t i; | |
2474 | const unsigned char *sha1; | |
2475 | ||
2476 | for (p = packed_git; p; p = p->next) { | |
79bc4c71 | 2477 | if (!p->pack_local || p->pack_keep) |
ca11b212 NP |
2478 | continue; |
2479 | ||
2480 | if (open_pack_index(p)) | |
2481 | die("cannot open pack index"); | |
2482 | ||
2483 | for (i = 0; i < p->num_objects; i++) { | |
2484 | sha1 = nth_packed_object_sha1(p, i); | |
2834bc27 | 2485 | if (!packlist_find(&to_pack, sha1, NULL) && |
d0d46abc JK |
2486 | !has_sha1_pack_kept_or_nonlocal(sha1) && |
2487 | !loosened_object_can_be_discarded(sha1, p->mtime)) | |
ca11b212 NP |
2488 | if (force_object_loose(sha1, p->mtime)) |
2489 | die("unable to force loose object"); | |
2490 | } | |
2491 | } | |
2492 | } | |
2493 | ||
69e4b342 JK |
2494 | /* |
2495 | * This tracks any options which a reader of the pack might | |
2496 | * not understand, and which would therefore prevent blind reuse | |
2497 | * of what we have on disk. | |
2498 | */ | |
2499 | static int pack_options_allow_reuse(void) | |
2500 | { | |
2501 | return allow_ofs_delta; | |
2502 | } | |
2503 | ||
6b8fda2d VM |
2504 | static int get_object_list_from_bitmap(struct rev_info *revs) |
2505 | { | |
2506 | if (prepare_bitmap_walk(revs) < 0) | |
2507 | return -1; | |
2508 | ||
69e4b342 JK |
2509 | if (pack_options_allow_reuse() && |
2510 | !reuse_partial_packfile_from_bitmap( | |
6b8fda2d VM |
2511 | &reuse_packfile, |
2512 | &reuse_packfile_objects, | |
2513 | &reuse_packfile_offset)) { | |
2514 | assert(reuse_packfile_objects); | |
2515 | nr_result += reuse_packfile_objects; | |
78d2214e | 2516 | display_progress(progress_state, nr_result); |
6b8fda2d VM |
2517 | } |
2518 | ||
2519 | traverse_bitmap_commit_list(&add_object_entry_from_bitmap); | |
2520 | return 0; | |
2521 | } | |
2522 | ||
abcb8655 | 2523 | static void record_recent_object(struct object *obj, |
de1e67d0 | 2524 | const char *name, |
abcb8655 JK |
2525 | void *data) |
2526 | { | |
ed1c9977 | 2527 | sha1_array_append(&recent_objects, obj->oid.hash); |
abcb8655 JK |
2528 | } |
2529 | ||
2530 | static void record_recent_commit(struct commit *commit, void *data) | |
2531 | { | |
ed1c9977 | 2532 | sha1_array_append(&recent_objects, commit->object.oid.hash); |
abcb8655 JK |
2533 | } |
2534 | ||
8d1d8f83 | 2535 | static void get_object_list(int ac, const char **av) |
b5d97e6b JH |
2536 | { |
2537 | struct rev_info revs; | |
2538 | char line[1000]; | |
b5d97e6b JH |
2539 | int flags = 0; |
2540 | ||
b5d97e6b JH |
2541 | init_revisions(&revs, NULL); |
2542 | save_commit_buffer = 0; | |
b5d97e6b JH |
2543 | setup_revisions(ac, av, &revs, NULL); |
2544 | ||
b790e0f6 NTND |
2545 | /* make sure shallows are read */ |
2546 | is_repository_shallow(); | |
2547 | ||
b5d97e6b JH |
2548 | while (fgets(line, sizeof(line), stdin) != NULL) { |
2549 | int len = strlen(line); | |
872c930d | 2550 | if (len && line[len - 1] == '\n') |
b5d97e6b JH |
2551 | line[--len] = 0; |
2552 | if (!len) | |
2553 | break; | |
2554 | if (*line == '-') { | |
2555 | if (!strcmp(line, "--not")) { | |
2556 | flags ^= UNINTERESTING; | |
7cc8f971 | 2557 | write_bitmap_index = 0; |
b5d97e6b JH |
2558 | continue; |
2559 | } | |
b790e0f6 NTND |
2560 | if (starts_with(line, "--shallow ")) { |
2561 | unsigned char sha1[20]; | |
2562 | if (get_sha1_hex(line + 10, sha1)) | |
2563 | die("not an SHA-1 '%s'", line + 10); | |
2564 | register_shallow(sha1); | |
f7f91086 | 2565 | use_bitmap_index = 0; |
b790e0f6 NTND |
2566 | continue; |
2567 | } | |
b5d97e6b JH |
2568 | die("not a rev '%s'", line); |
2569 | } | |
8e676e8b | 2570 | if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME)) |
b5d97e6b JH |
2571 | die("bad revision '%s'", line); |
2572 | } | |
2573 | ||
6b8fda2d VM |
2574 | if (use_bitmap_index && !get_object_list_from_bitmap(&revs)) |
2575 | return; | |
2576 | ||
3d51e1b5 MK |
2577 | if (prepare_revision_walk(&revs)) |
2578 | die("revision walk setup failed"); | |
e76a5fb4 | 2579 | mark_edges_uninteresting(&revs, show_edge); |
11c211fa | 2580 | traverse_commit_list(&revs, show_commit, show_object, NULL); |
08cdfb13 | 2581 | |
abcb8655 JK |
2582 | if (unpack_unreachable_expiration) { |
2583 | revs.ignore_missing_links = 1; | |
2584 | if (add_unseen_recent_objects_to_traversal(&revs, | |
2585 | unpack_unreachable_expiration)) | |
2586 | die("unable to add recent objects"); | |
2587 | if (prepare_revision_walk(&revs)) | |
2588 | die("revision walk setup failed"); | |
2589 | traverse_commit_list(&revs, record_recent_commit, | |
2590 | record_recent_object, NULL); | |
2591 | } | |
2592 | ||
08cdfb13 JH |
2593 | if (keep_unreachable) |
2594 | add_objects_in_unpacked_packs(&revs); | |
e26a8c47 JK |
2595 | if (pack_loose_unreachable) |
2596 | add_unreachable_loose_objects(); | |
ca11b212 NP |
2597 | if (unpack_unreachable) |
2598 | loosen_unused_packed_objects(&revs); | |
abcb8655 JK |
2599 | |
2600 | sha1_array_clear(&recent_objects); | |
b5d97e6b JH |
2601 | } |
2602 | ||
99fb6e04 NTND |
2603 | static int option_parse_index_version(const struct option *opt, |
2604 | const char *arg, int unset) | |
2605 | { | |
2606 | char *c; | |
2607 | const char *val = arg; | |
2608 | pack_idx_opts.version = strtoul(val, &c, 10); | |
2609 | if (pack_idx_opts.version > 2) | |
2610 | die(_("unsupported index version %s"), val); | |
2611 | if (*c == ',' && c[1]) | |
2612 | pack_idx_opts.off32_limit = strtoul(c+1, &c, 0); | |
2613 | if (*c || pack_idx_opts.off32_limit & 0x80000000) | |
2614 | die(_("bad index version '%s'"), val); | |
2615 | return 0; | |
2616 | } | |
2617 | ||
7e52f566 JK |
2618 | static int option_parse_unpack_unreachable(const struct option *opt, |
2619 | const char *arg, int unset) | |
2620 | { | |
2621 | if (unset) { | |
2622 | unpack_unreachable = 0; | |
2623 | unpack_unreachable_expiration = 0; | |
2624 | } | |
2625 | else { | |
2626 | unpack_unreachable = 1; | |
2627 | if (arg) | |
2628 | unpack_unreachable_expiration = approxidate(arg); | |
2629 | } | |
2630 | return 0; | |
2631 | } | |
2632 | ||
b5d97e6b JH |
2633 | int cmd_pack_objects(int argc, const char **argv, const char *prefix) |
2634 | { | |
b5d97e6b | 2635 | int use_internal_rev_list = 0; |
8d1d8f83 | 2636 | int thin = 0; |
2dacf26d | 2637 | int shallow = 0; |
4f366275 | 2638 | int all_progress_implied = 0; |
edfbb2aa | 2639 | struct argv_array rp = ARGV_ARRAY_INIT; |
99fb6e04 | 2640 | int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0; |
c90f9e13 | 2641 | int rev_list_index = 0; |
99fb6e04 NTND |
2642 | struct option pack_objects_options[] = { |
2643 | OPT_SET_INT('q', "quiet", &progress, | |
4c688120 | 2644 | N_("do not show progress meter"), 0), |
99fb6e04 | 2645 | OPT_SET_INT(0, "progress", &progress, |
4c688120 | 2646 | N_("show progress meter"), 1), |
99fb6e04 | 2647 | OPT_SET_INT(0, "all-progress", &progress, |
4c688120 | 2648 | N_("show progress meter during object writing phase"), 2), |
99fb6e04 NTND |
2649 | OPT_BOOL(0, "all-progress-implied", |
2650 | &all_progress_implied, | |
4c688120 NTND |
2651 | N_("similar to --all-progress when progress meter is shown")), |
2652 | { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"), | |
2653 | N_("write the pack index file in the specified idx format version"), | |
99fb6e04 | 2654 | 0, option_parse_index_version }, |
2a514ed8 CB |
2655 | OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit, |
2656 | N_("maximum size of each output pack file")), | |
99fb6e04 | 2657 | OPT_BOOL(0, "local", &local, |
4c688120 | 2658 | N_("ignore borrowed objects from alternate object store")), |
99fb6e04 | 2659 | OPT_BOOL(0, "incremental", &incremental, |
4c688120 | 2660 | N_("ignore packed objects")), |
99fb6e04 | 2661 | OPT_INTEGER(0, "window", &window, |
4c688120 | 2662 | N_("limit pack window by objects")), |
2a514ed8 CB |
2663 | OPT_MAGNITUDE(0, "window-memory", &window_memory_limit, |
2664 | N_("limit pack window by memory in addition to object limit")), | |
99fb6e04 | 2665 | OPT_INTEGER(0, "depth", &depth, |
4c688120 | 2666 | N_("maximum length of delta chain allowed in the resulting pack")), |
99fb6e04 | 2667 | OPT_BOOL(0, "reuse-delta", &reuse_delta, |
4c688120 | 2668 | N_("reuse existing deltas")), |
99fb6e04 | 2669 | OPT_BOOL(0, "reuse-object", &reuse_object, |
4c688120 | 2670 | N_("reuse existing objects")), |
99fb6e04 | 2671 | OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta, |
4c688120 | 2672 | N_("use OFS_DELTA objects")), |
99fb6e04 | 2673 | OPT_INTEGER(0, "threads", &delta_search_threads, |
4c688120 | 2674 | N_("use threads when searching for best delta matches")), |
99fb6e04 | 2675 | OPT_BOOL(0, "non-empty", &non_empty, |
4c688120 | 2676 | N_("do not create an empty pack output")), |
99fb6e04 | 2677 | OPT_BOOL(0, "revs", &use_internal_rev_list, |
4c688120 | 2678 | N_("read revision arguments from standard input")), |
99fb6e04 | 2679 | { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL, |
4c688120 | 2680 | N_("limit the objects to those that are not yet packed"), |
99fb6e04 NTND |
2681 | PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 }, |
2682 | { OPTION_SET_INT, 0, "all", &rev_list_all, NULL, | |
4c688120 | 2683 | N_("include objects reachable from any reference"), |
99fb6e04 NTND |
2684 | PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 }, |
2685 | { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL, | |
4c688120 | 2686 | N_("include objects referred by reflog entries"), |
99fb6e04 | 2687 | PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 }, |
c90f9e13 JK |
2688 | { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL, |
2689 | N_("include objects referred to by the index"), | |
2690 | PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 }, | |
99fb6e04 | 2691 | OPT_BOOL(0, "stdout", &pack_to_stdout, |
4c688120 | 2692 | N_("output pack to stdout")), |
99fb6e04 | 2693 | OPT_BOOL(0, "include-tag", &include_tag, |
4c688120 | 2694 | N_("include tag objects that refer to objects to be packed")), |
99fb6e04 | 2695 | OPT_BOOL(0, "keep-unreachable", &keep_unreachable, |
4c688120 | 2696 | N_("keep unreachable objects")), |
e26a8c47 JK |
2697 | OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable, |
2698 | N_("pack loose unreachable objects")), | |
4c688120 NTND |
2699 | { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"), |
2700 | N_("unpack unreachable objects newer than <time>"), | |
7e52f566 | 2701 | PARSE_OPT_OPTARG, option_parse_unpack_unreachable }, |
99fb6e04 | 2702 | OPT_BOOL(0, "thin", &thin, |
4c688120 | 2703 | N_("create thin packs")), |
2dacf26d | 2704 | OPT_BOOL(0, "shallow", &shallow, |
2705 | N_("create packs suitable for shallow fetches")), | |
99fb6e04 | 2706 | OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep, |
4c688120 | 2707 | N_("ignore packs that have companion .keep file")), |
99fb6e04 | 2708 | OPT_INTEGER(0, "compression", &pack_compression_level, |
4c688120 | 2709 | N_("pack compression level")), |
99fb6e04 | 2710 | OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents, |
4c688120 | 2711 | N_("do not hide commits by grafts"), 0), |
6b8fda2d VM |
2712 | OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index, |
2713 | N_("use a bitmap index if available to speed up counting objects")), | |
7cc8f971 VM |
2714 | OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index, |
2715 | N_("write a bitmap index together with the pack index")), | |
99fb6e04 NTND |
2716 | OPT_END(), |
2717 | }; | |
8d1d8f83 | 2718 | |
afc711b8 | 2719 | check_replace_refs = 0; |
dae556bd | 2720 | |
ebcfb379 | 2721 | reset_pack_idx_option(&pack_idx_opts); |
ef90d6d4 | 2722 | git_config(git_pack_config, NULL); |
960ccca6 DH |
2723 | if (!pack_compression_seen && core_compression_seen) |
2724 | pack_compression_level = core_compression_level; | |
b5d97e6b JH |
2725 | |
2726 | progress = isatty(2); | |
99fb6e04 NTND |
2727 | argc = parse_options(argc, argv, prefix, pack_objects_options, |
2728 | pack_usage, 0); | |
b5d97e6b | 2729 | |
99fb6e04 NTND |
2730 | if (argc) { |
2731 | base_name = argv[0]; | |
2732 | argc--; | |
b5d97e6b | 2733 | } |
99fb6e04 NTND |
2734 | if (pack_to_stdout != !base_name || argc) |
2735 | usage_with_options(pack_usage, pack_objects_options); | |
b5d97e6b | 2736 | |
edfbb2aa | 2737 | argv_array_push(&rp, "pack-objects"); |
99fb6e04 NTND |
2738 | if (thin) { |
2739 | use_internal_rev_list = 1; | |
2dacf26d | 2740 | argv_array_push(&rp, shallow |
2741 | ? "--objects-edge-aggressive" | |
2742 | : "--objects-edge"); | |
99fb6e04 | 2743 | } else |
edfbb2aa | 2744 | argv_array_push(&rp, "--objects"); |
b5d97e6b | 2745 | |
99fb6e04 NTND |
2746 | if (rev_list_all) { |
2747 | use_internal_rev_list = 1; | |
edfbb2aa | 2748 | argv_array_push(&rp, "--all"); |
99fb6e04 NTND |
2749 | } |
2750 | if (rev_list_reflog) { | |
2751 | use_internal_rev_list = 1; | |
edfbb2aa | 2752 | argv_array_push(&rp, "--reflog"); |
99fb6e04 | 2753 | } |
c90f9e13 JK |
2754 | if (rev_list_index) { |
2755 | use_internal_rev_list = 1; | |
2756 | argv_array_push(&rp, "--indexed-objects"); | |
99fb6e04 NTND |
2757 | } |
2758 | if (rev_list_unpacked) { | |
2759 | use_internal_rev_list = 1; | |
edfbb2aa | 2760 | argv_array_push(&rp, "--unpacked"); |
99fb6e04 | 2761 | } |
b5d97e6b | 2762 | |
99fb6e04 NTND |
2763 | if (!reuse_object) |
2764 | reuse_delta = 0; | |
2765 | if (pack_compression_level == -1) | |
2766 | pack_compression_level = Z_DEFAULT_COMPRESSION; | |
2767 | else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION) | |
2768 | die("bad pack compression level %d", pack_compression_level); | |
0c45d258 JH |
2769 | |
2770 | if (!delta_search_threads) /* --threads=0 means autodetect */ | |
2771 | delta_search_threads = online_cpus(); | |
2772 | ||
99fb6e04 NTND |
2773 | #ifdef NO_PTHREADS |
2774 | if (delta_search_threads != 1) | |
2b34e486 | 2775 | warning("no threads support, ignoring --threads"); |
99fb6e04 | 2776 | #endif |
2b84b5a8 JS |
2777 | if (!pack_to_stdout && !pack_size_limit) |
2778 | pack_size_limit = pack_size_limit_cfg; | |
01c12a23 DH |
2779 | if (pack_to_stdout && pack_size_limit) |
2780 | die("--max-pack-size cannot be used to build a pack for transfer."); | |
07cf0f24 NP |
2781 | if (pack_size_limit && pack_size_limit < 1024*1024) { |
2782 | warning("minimum pack size limit is 1 MiB"); | |
2783 | pack_size_limit = 1024*1024; | |
2784 | } | |
01c12a23 | 2785 | |
8d1d8f83 JH |
2786 | if (!pack_to_stdout && thin) |
2787 | die("--thin cannot be used to build an indexable pack."); | |
b5d97e6b | 2788 | |
ca11b212 NP |
2789 | if (keep_unreachable && unpack_unreachable) |
2790 | die("--keep-unreachable and --unpack-unreachable are incompatible."); | |
b1e757f3 JK |
2791 | if (!rev_list_all || !rev_list_reflog || !rev_list_index) |
2792 | unpack_unreachable_expiration = 0; | |
ca11b212 | 2793 | |
6b8fda2d VM |
2794 | if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow()) |
2795 | use_bitmap_index = 0; | |
2796 | ||
7cc8f971 VM |
2797 | if (pack_to_stdout || !rev_list_all) |
2798 | write_bitmap_index = 0; | |
2799 | ||
4f366275 NP |
2800 | if (progress && all_progress_implied) |
2801 | progress = 2; | |
2802 | ||
b5d97e6b | 2803 | prepare_packed_git(); |
56dfeb62 JK |
2804 | if (ignore_packed_keep) { |
2805 | struct packed_git *p; | |
2806 | for (p = packed_git; p; p = p->next) | |
2807 | if (p->pack_local && p->pack_keep) | |
2808 | break; | |
2809 | if (!p) /* no keep-able packs found */ | |
2810 | ignore_packed_keep = 0; | |
2811 | } | |
2812 | if (local) { | |
2813 | /* | |
2814 | * unlike ignore_packed_keep above, we do not want to | |
2815 | * unset "local" based on looking at packs, as it | |
2816 | * also covers non-local objects | |
2817 | */ | |
2818 | struct packed_git *p; | |
2819 | for (p = packed_git; p; p = p->next) { | |
2820 | if (!p->pack_local) { | |
2821 | have_non_local_packs = 1; | |
2822 | break; | |
2823 | } | |
2824 | } | |
2825 | } | |
b5d97e6b | 2826 | |
13aaf148 | 2827 | if (progress) |
754dbc43 | 2828 | progress_state = start_progress(_("Counting objects"), 0); |
b5d97e6b JH |
2829 | if (!use_internal_rev_list) |
2830 | read_object_list_from_stdin(); | |
8d1d8f83 | 2831 | else { |
edfbb2aa JK |
2832 | get_object_list(rp.argc, rp.argv); |
2833 | argv_array_clear(&rp); | |
8d1d8f83 | 2834 | } |
0ef95f72 | 2835 | cleanup_preferred_base(); |
d155254c MH |
2836 | if (include_tag && nr_result) |
2837 | for_each_ref(add_ref_tag, NULL); | |
4d4fcc54 | 2838 | stop_progress(&progress_state); |
96a02f8f | 2839 | |
f0b0af1b | 2840 | if (non_empty && !nr_result) |
1c4a2912 | 2841 | return 0; |
f7ae6a93 NP |
2842 | if (nr_result) |
2843 | prepare_pack(window, depth); | |
d01fb92f | 2844 | write_pack_file(); |
ab7cd7bb | 2845 | if (progress) |
6e1c2344 RJ |
2846 | fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32")," |
2847 | " reused %"PRIu32" (delta %"PRIu32")\n", | |
67c08ce1 | 2848 | written, written_delta, reused, reused_delta); |
c323ac7d LT |
2849 | return 0; |
2850 | } |