THIS IS A TEST INSTANCE ONLY! REPOSITORIES CAN BE DELETED AT ANY TIME!

Git Source Code Mirror - This is a publish-only repository and all pull requests are ignored. Please follow Documentation/SubmittingPatches procedure for any of your improvements.
git
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

452 lines
12KB

  1. #ifndef PACK_OBJECTS_H
  2. #define PACK_OBJECTS_H
  3. #include "object-store.h"
  4. #include "thread-utils.h"
  5. #include "pack.h"
  6. struct repository;
  7. #define DEFAULT_DELTA_CACHE_SIZE (256 * 1024 * 1024)
  8. #define OE_DFS_STATE_BITS 2
  9. #define OE_DEPTH_BITS 12
  10. #define OE_IN_PACK_BITS 10
  11. #define OE_Z_DELTA_BITS 20
  12. /*
  13. * Note that oe_set_size() becomes expensive when the given size is
  14. * above this limit. Don't lower it too much.
  15. */
  16. #define OE_SIZE_BITS 31
  17. #define OE_DELTA_SIZE_BITS 23
  18. /*
  19. * State flags for depth-first search used for analyzing delta cycles.
  20. *
  21. * The depth is measured in delta-links to the base (so if A is a delta
  22. * against B, then A has a depth of 1, and B a depth of 0).
  23. */
  24. enum dfs_state {
  25. DFS_NONE = 0,
  26. DFS_ACTIVE,
  27. DFS_DONE,
  28. DFS_NUM_STATES
  29. };
  30. /*
  31. * The size of struct nearly determines pack-objects's memory
  32. * consumption. This struct is packed tight for that reason. When you
  33. * add or reorder something in this struct, think a bit about this.
  34. *
  35. * basic object info
  36. * -----------------
  37. * idx.oid is filled up before delta searching starts. idx.crc32 is
  38. * only valid after the object is written out and will be used for
  39. * generating the index. idx.offset will be both gradually set and
  40. * used in writing phase (base objects get offset first, then deltas
  41. * refer to them)
  42. *
  43. * "size" is the uncompressed object size. Compressed size of the raw
  44. * data for an object in a pack is not stored anywhere but is computed
  45. * and made available when reverse .idx is made. Note that when a
  46. * delta is reused, "size" is the uncompressed _delta_ size, not the
  47. * canonical one after the delta has been applied.
  48. *
  49. * "hash" contains a path name hash which is used for sorting the
  50. * delta list and also during delta searching. Once prepare_pack()
  51. * returns it's no longer needed.
  52. *
  53. * source pack info
  54. * ----------------
  55. * The (in_pack, in_pack_offset) tuple contains the location of the
  56. * object in the source pack. in_pack_header_size allows quickly
  57. * skipping the header and going straight to the zlib stream.
  58. *
  59. * "type" and "in_pack_type" both describe object type. in_pack_type
  60. * may contain a delta type, while type is always the canonical type.
  61. *
  62. * deltas
  63. * ------
  64. * Delta links (delta, delta_child and delta_sibling) are created to
  65. * reflect that delta graph from the source pack then updated or added
  66. * during delta searching phase when we find better deltas.
  67. *
  68. * delta_child and delta_sibling are last needed in
  69. * compute_write_order(). "delta" and "delta_size" must remain valid
  70. * at object writing phase in case the delta is not cached.
  71. *
  72. * If a delta is cached in memory and is compressed, delta_data points
  73. * to the data and z_delta_size contains the compressed size. If it's
  74. * uncompressed [1], z_delta_size must be zero. delta_size is always
  75. * the uncompressed size and must be valid even if the delta is not
  76. * cached.
  77. *
  78. * [1] during try_delta phase we don't bother with compressing because
  79. * the delta could be quickly replaced with a better one.
  80. */
  81. struct object_entry {
  82. struct pack_idx_entry idx;
  83. void *delta_data; /* cached delta (uncompressed) */
  84. off_t in_pack_offset;
  85. uint32_t hash; /* name hint hash */
  86. unsigned size_:OE_SIZE_BITS;
  87. unsigned size_valid:1;
  88. uint32_t delta_idx; /* delta base object */
  89. uint32_t delta_child_idx; /* deltified objects who bases me */
  90. uint32_t delta_sibling_idx; /* other deltified objects who
  91. * uses the same base as me
  92. */
  93. unsigned delta_size_:OE_DELTA_SIZE_BITS; /* delta data size (uncompressed) */
  94. unsigned delta_size_valid:1;
  95. unsigned char in_pack_header_size;
  96. unsigned in_pack_idx:OE_IN_PACK_BITS; /* already in pack */
  97. unsigned z_delta_size:OE_Z_DELTA_BITS;
  98. unsigned type_valid:1;
  99. unsigned no_try_delta:1;
  100. unsigned type_:TYPE_BITS;
  101. unsigned in_pack_type:TYPE_BITS; /* could be delta */
  102. unsigned preferred_base:1; /*
  103. * we do not pack this, but is available
  104. * to be used as the base object to delta
  105. * objects against.
  106. */
  107. unsigned tagged:1; /* near the very tip of refs */
  108. unsigned filled:1; /* assigned write-order */
  109. unsigned dfs_state:OE_DFS_STATE_BITS;
  110. unsigned depth:OE_DEPTH_BITS;
  111. unsigned ext_base:1; /* delta_idx points outside packlist */
  112. /*
  113. * pahole results on 64-bit linux (gcc and clang)
  114. *
  115. * size: 80, bit_padding: 9 bits
  116. *
  117. * and on 32-bit (gcc)
  118. *
  119. * size: 76, bit_padding: 9 bits
  120. */
  121. };
  122. struct packing_data {
  123. struct repository *repo;
  124. struct object_entry *objects;
  125. uint32_t nr_objects, nr_alloc;
  126. int32_t *index;
  127. uint32_t index_size;
  128. unsigned int *in_pack_pos;
  129. unsigned long *delta_size;
  130. /*
  131. * Only one of these can be non-NULL and they have different
  132. * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns
  133. * the pack of an object using in_pack_idx field. If not,
  134. * in_pack[] array is used the same way as in_pack_pos[]
  135. */
  136. struct packed_git **in_pack_by_idx;
  137. struct packed_git **in_pack;
  138. /*
  139. * During packing with multiple threads, protect the in-core
  140. * object database from concurrent accesses.
  141. */
  142. pthread_mutex_t odb_lock;
  143. /*
  144. * This list contains entries for bases which we know the other side
  145. * has (e.g., via reachability bitmaps), but which aren't in our
  146. * "objects" list.
  147. */
  148. struct object_entry *ext_bases;
  149. uint32_t nr_ext, alloc_ext;
  150. uintmax_t oe_size_limit;
  151. uintmax_t oe_delta_size_limit;
  152. /* delta islands */
  153. unsigned int *tree_depth;
  154. unsigned char *layer;
  155. };
  156. void prepare_packing_data(struct repository *r, struct packing_data *pdata);
  157. /* Protect access to object database */
  158. static inline void packing_data_lock(struct packing_data *pdata)
  159. {
  160. pthread_mutex_lock(&pdata->odb_lock);
  161. }
  162. static inline void packing_data_unlock(struct packing_data *pdata)
  163. {
  164. pthread_mutex_unlock(&pdata->odb_lock);
  165. }
  166. struct object_entry *packlist_alloc(struct packing_data *pdata,
  167. const struct object_id *oid);
  168. struct object_entry *packlist_find(struct packing_data *pdata,
  169. const struct object_id *oid);
  170. static inline uint32_t pack_name_hash(const char *name)
  171. {
  172. uint32_t c, hash = 0;
  173. if (!name)
  174. return 0;
  175. /*
  176. * This effectively just creates a sortable number from the
  177. * last sixteen non-whitespace characters. Last characters
  178. * count "most", so things that end in ".c" sort together.
  179. */
  180. while ((c = *name++) != 0) {
  181. if (isspace(c))
  182. continue;
  183. hash = (hash >> 2) + (c << 24);
  184. }
  185. return hash;
  186. }
  187. static inline enum object_type oe_type(const struct object_entry *e)
  188. {
  189. return e->type_valid ? e->type_ : OBJ_BAD;
  190. }
  191. static inline void oe_set_type(struct object_entry *e,
  192. enum object_type type)
  193. {
  194. if (type >= OBJ_ANY)
  195. BUG("OBJ_ANY cannot be set in pack-objects code");
  196. e->type_valid = type >= OBJ_NONE;
  197. e->type_ = (unsigned)type;
  198. }
  199. static inline unsigned int oe_in_pack_pos(const struct packing_data *pack,
  200. const struct object_entry *e)
  201. {
  202. return pack->in_pack_pos[e - pack->objects];
  203. }
  204. static inline void oe_set_in_pack_pos(const struct packing_data *pack,
  205. const struct object_entry *e,
  206. unsigned int pos)
  207. {
  208. pack->in_pack_pos[e - pack->objects] = pos;
  209. }
  210. static inline struct packed_git *oe_in_pack(const struct packing_data *pack,
  211. const struct object_entry *e)
  212. {
  213. if (pack->in_pack_by_idx)
  214. return pack->in_pack_by_idx[e->in_pack_idx];
  215. else
  216. return pack->in_pack[e - pack->objects];
  217. }
  218. void oe_map_new_pack(struct packing_data *pack);
  219. static inline void oe_set_in_pack(struct packing_data *pack,
  220. struct object_entry *e,
  221. struct packed_git *p)
  222. {
  223. if (pack->in_pack_by_idx) {
  224. if (p->index) {
  225. e->in_pack_idx = p->index;
  226. return;
  227. }
  228. /*
  229. * We're accessing packs by index, but this pack doesn't have
  230. * an index (e.g., because it was added since we created the
  231. * in_pack_by_idx array). Bail to oe_map_new_pack(), which
  232. * will convert us to using the full in_pack array, and then
  233. * fall through to our in_pack handling.
  234. */
  235. oe_map_new_pack(pack);
  236. }
  237. pack->in_pack[e - pack->objects] = p;
  238. }
  239. static inline struct object_entry *oe_delta(
  240. const struct packing_data *pack,
  241. const struct object_entry *e)
  242. {
  243. if (!e->delta_idx)
  244. return NULL;
  245. if (e->ext_base)
  246. return &pack->ext_bases[e->delta_idx - 1];
  247. else
  248. return &pack->objects[e->delta_idx - 1];
  249. }
  250. static inline void oe_set_delta(struct packing_data *pack,
  251. struct object_entry *e,
  252. struct object_entry *delta)
  253. {
  254. if (delta)
  255. e->delta_idx = (delta - pack->objects) + 1;
  256. else
  257. e->delta_idx = 0;
  258. }
  259. void oe_set_delta_ext(struct packing_data *pack,
  260. struct object_entry *e,
  261. const unsigned char *sha1);
  262. static inline struct object_entry *oe_delta_child(
  263. const struct packing_data *pack,
  264. const struct object_entry *e)
  265. {
  266. if (e->delta_child_idx)
  267. return &pack->objects[e->delta_child_idx - 1];
  268. return NULL;
  269. }
  270. static inline void oe_set_delta_child(struct packing_data *pack,
  271. struct object_entry *e,
  272. struct object_entry *delta)
  273. {
  274. if (delta)
  275. e->delta_child_idx = (delta - pack->objects) + 1;
  276. else
  277. e->delta_child_idx = 0;
  278. }
  279. static inline struct object_entry *oe_delta_sibling(
  280. const struct packing_data *pack,
  281. const struct object_entry *e)
  282. {
  283. if (e->delta_sibling_idx)
  284. return &pack->objects[e->delta_sibling_idx - 1];
  285. return NULL;
  286. }
  287. static inline void oe_set_delta_sibling(struct packing_data *pack,
  288. struct object_entry *e,
  289. struct object_entry *delta)
  290. {
  291. if (delta)
  292. e->delta_sibling_idx = (delta - pack->objects) + 1;
  293. else
  294. e->delta_sibling_idx = 0;
  295. }
  296. unsigned long oe_get_size_slow(struct packing_data *pack,
  297. const struct object_entry *e);
  298. static inline unsigned long oe_size(struct packing_data *pack,
  299. const struct object_entry *e)
  300. {
  301. if (e->size_valid)
  302. return e->size_;
  303. return oe_get_size_slow(pack, e);
  304. }
  305. static inline int oe_size_less_than(struct packing_data *pack,
  306. const struct object_entry *lhs,
  307. unsigned long rhs)
  308. {
  309. if (lhs->size_valid)
  310. return lhs->size_ < rhs;
  311. if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
  312. return 0;
  313. return oe_get_size_slow(pack, lhs) < rhs;
  314. }
  315. static inline int oe_size_greater_than(struct packing_data *pack,
  316. const struct object_entry *lhs,
  317. unsigned long rhs)
  318. {
  319. if (lhs->size_valid)
  320. return lhs->size_ > rhs;
  321. if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
  322. return 1;
  323. return oe_get_size_slow(pack, lhs) > rhs;
  324. }
  325. static inline void oe_set_size(struct packing_data *pack,
  326. struct object_entry *e,
  327. unsigned long size)
  328. {
  329. if (size < pack->oe_size_limit) {
  330. e->size_ = size;
  331. e->size_valid = 1;
  332. } else {
  333. e->size_valid = 0;
  334. if (oe_get_size_slow(pack, e) != size)
  335. BUG("'size' is supposed to be the object size!");
  336. }
  337. }
  338. static inline unsigned long oe_delta_size(struct packing_data *pack,
  339. const struct object_entry *e)
  340. {
  341. if (e->delta_size_valid)
  342. return e->delta_size_;
  343. /*
  344. * pack->delta_size[] can't be NULL because oe_set_delta_size()
  345. * must have been called when a new delta is saved with
  346. * oe_set_delta().
  347. * If oe_delta() returns NULL (i.e. default state, which means
  348. * delta_size_valid is also false), then the caller must never
  349. * call oe_delta_size().
  350. */
  351. return pack->delta_size[e - pack->objects];
  352. }
  353. static inline void oe_set_delta_size(struct packing_data *pack,
  354. struct object_entry *e,
  355. unsigned long size)
  356. {
  357. if (size < pack->oe_delta_size_limit) {
  358. e->delta_size_ = size;
  359. e->delta_size_valid = 1;
  360. } else {
  361. packing_data_lock(pack);
  362. if (!pack->delta_size)
  363. ALLOC_ARRAY(pack->delta_size, pack->nr_alloc);
  364. packing_data_unlock(pack);
  365. pack->delta_size[e - pack->objects] = size;
  366. e->delta_size_valid = 0;
  367. }
  368. }
  369. static inline unsigned int oe_tree_depth(struct packing_data *pack,
  370. struct object_entry *e)
  371. {
  372. if (!pack->tree_depth)
  373. return 0;
  374. return pack->tree_depth[e - pack->objects];
  375. }
  376. static inline void oe_set_tree_depth(struct packing_data *pack,
  377. struct object_entry *e,
  378. unsigned int tree_depth)
  379. {
  380. if (!pack->tree_depth)
  381. CALLOC_ARRAY(pack->tree_depth, pack->nr_alloc);
  382. pack->tree_depth[e - pack->objects] = tree_depth;
  383. }
  384. static inline unsigned char oe_layer(struct packing_data *pack,
  385. struct object_entry *e)
  386. {
  387. if (!pack->layer)
  388. return 0;
  389. return pack->layer[e - pack->objects];
  390. }
  391. static inline void oe_set_layer(struct packing_data *pack,
  392. struct object_entry *e,
  393. unsigned char layer)
  394. {
  395. if (!pack->layer)
  396. CALLOC_ARRAY(pack->layer, pack->nr_alloc);
  397. pack->layer[e - pack->objects] = layer;
  398. }
  399. #endif