1#ifndef PACK_OBJECTS_H 2#define PACK_OBJECTS_H 3 4#include"object-store.h" 5#include"thread-utils.h" 6#include"pack.h" 7 8#define DEFAULT_DELTA_CACHE_SIZE (256 * 1024 * 1024) 9 10#define OE_DFS_STATE_BITS 2 11#define OE_DEPTH_BITS 12 12#define OE_IN_PACK_BITS 10 13#define OE_Z_DELTA_BITS 20 14/* 15 * Note that oe_set_size() becomes expensive when the given size is 16 * above this limit. Don't lower it too much. 17 */ 18#define OE_SIZE_BITS 31 19#define OE_DELTA_SIZE_BITS 23 20 21/* 22 * State flags for depth-first search used for analyzing delta cycles. 23 * 24 * The depth is measured in delta-links to the base (so if A is a delta 25 * against B, then A has a depth of 1, and B a depth of 0). 26 */ 27enum dfs_state { 28 DFS_NONE =0, 29 DFS_ACTIVE, 30 DFS_DONE, 31 DFS_NUM_STATES 32}; 33 34/* 35 * The size of struct nearly determines pack-objects's memory 36 * consumption. This struct is packed tight for that reason. When you 37 * add or reorder something in this struct, think a bit about this. 38 * 39 * basic object info 40 * ----------------- 41 * idx.oid is filled up before delta searching starts. idx.crc32 is 42 * only valid after the object is written out and will be used for 43 * generating the index. idx.offset will be both gradually set and 44 * used in writing phase (base objects get offset first, then deltas 45 * refer to them) 46 * 47 * "size" is the uncompressed object size. Compressed size of the raw 48 * data for an object in a pack is not stored anywhere but is computed 49 * and made available when reverse .idx is made. Note that when a 50 * delta is reused, "size" is the uncompressed _delta_ size, not the 51 * canonical one after the delta has been applied. 52 * 53 * "hash" contains a path name hash which is used for sorting the 54 * delta list and also during delta searching. Once prepare_pack() 55 * returns it's no longer needed. 56 * 57 * source pack info 58 * ---------------- 59 * The (in_pack, in_pack_offset) tuple contains the location of the 60 * object in the source pack. in_pack_header_size allows quickly 61 * skipping the header and going straight to the zlib stream. 62 * 63 * "type" and "in_pack_type" both describe object type. in_pack_type 64 * may contain a delta type, while type is always the canonical type. 65 * 66 * deltas 67 * ------ 68 * Delta links (delta, delta_child and delta_sibling) are created to 69 * reflect that delta graph from the source pack then updated or added 70 * during delta searching phase when we find better deltas. 71 * 72 * delta_child and delta_sibling are last needed in 73 * compute_write_order(). "delta" and "delta_size" must remain valid 74 * at object writing phase in case the delta is not cached. 75 * 76 * If a delta is cached in memory and is compressed, delta_data points 77 * to the data and z_delta_size contains the compressed size. If it's 78 * uncompressed [1], z_delta_size must be zero. delta_size is always 79 * the uncompressed size and must be valid even if the delta is not 80 * cached. 81 * 82 * [1] during try_delta phase we don't bother with compressing because 83 * the delta could be quickly replaced with a better one. 84 */ 85struct object_entry { 86struct pack_idx_entry idx; 87void*delta_data;/* cached delta (uncompressed) */ 88 off_t in_pack_offset; 89uint32_t hash;/* name hint hash */ 90unsigned size_:OE_SIZE_BITS; 91unsigned size_valid:1; 92uint32_t delta_idx;/* delta base object */ 93uint32_t delta_child_idx;/* deltified objects who bases me */ 94uint32_t delta_sibling_idx;/* other deltified objects who 95 * uses the same base as me 96 */ 97unsigned delta_size_:OE_DELTA_SIZE_BITS;/* delta data size (uncompressed) */ 98unsigned delta_size_valid:1; 99unsigned char in_pack_header_size; 100unsigned in_pack_idx:OE_IN_PACK_BITS;/* already in pack */ 101unsigned z_delta_size:OE_Z_DELTA_BITS; 102unsigned type_valid:1; 103unsigned no_try_delta:1; 104unsigned type_:TYPE_BITS; 105unsigned in_pack_type:TYPE_BITS;/* could be delta */ 106unsigned preferred_base:1;/* 107 * we do not pack this, but is available 108 * to be used as the base object to delta 109 * objects against. 110 */ 111unsigned tagged:1;/* near the very tip of refs */ 112unsigned filled:1;/* assigned write-order */ 113unsigned dfs_state:OE_DFS_STATE_BITS; 114unsigned depth:OE_DEPTH_BITS; 115 116/* 117 * pahole results on 64-bit linux (gcc and clang) 118 * 119 * size: 80, bit_padding: 9 bits 120 * 121 * and on 32-bit (gcc) 122 * 123 * size: 76, bit_padding: 9 bits 124 */ 125}; 126 127struct packing_data { 128struct object_entry *objects; 129uint32_t nr_objects, nr_alloc; 130 131int32_t*index; 132uint32_t index_size; 133 134unsigned int*in_pack_pos; 135unsigned long*delta_size; 136 137/* 138 * Only one of these can be non-NULL and they have different 139 * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns 140 * the pack of an object using in_pack_idx field. If not, 141 * in_pack[] array is used the same way as in_pack_pos[] 142 */ 143struct packed_git **in_pack_by_idx; 144struct packed_git **in_pack; 145 146#ifndef NO_PTHREADS 147 pthread_mutex_t lock; 148#endif 149 150uintmax_t oe_size_limit; 151uintmax_t oe_delta_size_limit; 152}; 153 154voidprepare_packing_data(struct packing_data *pdata); 155 156staticinlinevoidpacking_data_lock(struct packing_data *pdata) 157{ 158#ifndef NO_PTHREADS 159pthread_mutex_lock(&pdata->lock); 160#endif 161} 162staticinlinevoidpacking_data_unlock(struct packing_data *pdata) 163{ 164#ifndef NO_PTHREADS 165pthread_mutex_unlock(&pdata->lock); 166#endif 167} 168 169struct object_entry *packlist_alloc(struct packing_data *pdata, 170const unsigned char*sha1, 171uint32_t index_pos); 172 173struct object_entry *packlist_find(struct packing_data *pdata, 174const unsigned char*sha1, 175uint32_t*index_pos); 176 177staticinlineuint32_tpack_name_hash(const char*name) 178{ 179uint32_t c, hash =0; 180 181if(!name) 182return0; 183 184/* 185 * This effectively just creates a sortable number from the 186 * last sixteen non-whitespace characters. Last characters 187 * count "most", so things that end in ".c" sort together. 188 */ 189while((c = *name++) !=0) { 190if(isspace(c)) 191continue; 192 hash = (hash >>2) + (c <<24); 193} 194return hash; 195} 196 197staticinlineenum object_type oe_type(const struct object_entry *e) 198{ 199return e->type_valid ? e->type_ : OBJ_BAD; 200} 201 202staticinlinevoidoe_set_type(struct object_entry *e, 203enum object_type type) 204{ 205if(type >= OBJ_ANY) 206BUG("OBJ_ANY cannot be set in pack-objects code"); 207 208 e->type_valid = type >= OBJ_NONE; 209 e->type_ = (unsigned)type; 210} 211 212staticinlineunsigned intoe_in_pack_pos(const struct packing_data *pack, 213const struct object_entry *e) 214{ 215return pack->in_pack_pos[e - pack->objects]; 216} 217 218staticinlinevoidoe_set_in_pack_pos(const struct packing_data *pack, 219const struct object_entry *e, 220unsigned int pos) 221{ 222 pack->in_pack_pos[e - pack->objects] = pos; 223} 224 225staticinlinestruct packed_git *oe_in_pack(const struct packing_data *pack, 226const struct object_entry *e) 227{ 228if(pack->in_pack_by_idx) 229return pack->in_pack_by_idx[e->in_pack_idx]; 230else 231return pack->in_pack[e - pack->objects]; 232} 233 234voidoe_map_new_pack(struct packing_data *pack, 235struct packed_git *p); 236staticinlinevoidoe_set_in_pack(struct packing_data *pack, 237struct object_entry *e, 238struct packed_git *p) 239{ 240if(!p->index) 241oe_map_new_pack(pack, p); 242if(pack->in_pack_by_idx) 243 e->in_pack_idx = p->index; 244else 245 pack->in_pack[e - pack->objects] = p; 246} 247 248staticinlinestruct object_entry *oe_delta( 249const struct packing_data *pack, 250const struct object_entry *e) 251{ 252if(e->delta_idx) 253return&pack->objects[e->delta_idx -1]; 254return NULL; 255} 256 257staticinlinevoidoe_set_delta(struct packing_data *pack, 258struct object_entry *e, 259struct object_entry *delta) 260{ 261if(delta) 262 e->delta_idx = (delta - pack->objects) +1; 263else 264 e->delta_idx =0; 265} 266 267staticinlinestruct object_entry *oe_delta_child( 268const struct packing_data *pack, 269const struct object_entry *e) 270{ 271if(e->delta_child_idx) 272return&pack->objects[e->delta_child_idx -1]; 273return NULL; 274} 275 276staticinlinevoidoe_set_delta_child(struct packing_data *pack, 277struct object_entry *e, 278struct object_entry *delta) 279{ 280if(delta) 281 e->delta_child_idx = (delta - pack->objects) +1; 282else 283 e->delta_child_idx =0; 284} 285 286staticinlinestruct object_entry *oe_delta_sibling( 287const struct packing_data *pack, 288const struct object_entry *e) 289{ 290if(e->delta_sibling_idx) 291return&pack->objects[e->delta_sibling_idx -1]; 292return NULL; 293} 294 295staticinlinevoidoe_set_delta_sibling(struct packing_data *pack, 296struct object_entry *e, 297struct object_entry *delta) 298{ 299if(delta) 300 e->delta_sibling_idx = (delta - pack->objects) +1; 301else 302 e->delta_sibling_idx =0; 303} 304 305unsigned longoe_get_size_slow(struct packing_data *pack, 306const struct object_entry *e); 307staticinlineunsigned longoe_size(struct packing_data *pack, 308const struct object_entry *e) 309{ 310if(e->size_valid) 311return e->size_; 312 313returnoe_get_size_slow(pack, e); 314} 315 316staticinlineintoe_size_less_than(struct packing_data *pack, 317const struct object_entry *lhs, 318unsigned long rhs) 319{ 320if(lhs->size_valid) 321return lhs->size_ < rhs; 322if(rhs < pack->oe_size_limit)/* rhs < 2^x <= lhs ? */ 323return0; 324returnoe_get_size_slow(pack, lhs) < rhs; 325} 326 327staticinlineintoe_size_greater_than(struct packing_data *pack, 328const struct object_entry *lhs, 329unsigned long rhs) 330{ 331if(lhs->size_valid) 332return lhs->size_ > rhs; 333if(rhs < pack->oe_size_limit)/* rhs < 2^x <= lhs ? */ 334return1; 335returnoe_get_size_slow(pack, lhs) > rhs; 336} 337 338staticinlinevoidoe_set_size(struct packing_data *pack, 339struct object_entry *e, 340unsigned long size) 341{ 342if(size < pack->oe_size_limit) { 343 e->size_ = size; 344 e->size_valid =1; 345}else{ 346 e->size_valid =0; 347if(oe_get_size_slow(pack, e) != size) 348BUG("'size' is supposed to be the object size!"); 349} 350} 351 352staticinlineunsigned longoe_delta_size(struct packing_data *pack, 353const struct object_entry *e) 354{ 355if(e->delta_size_valid) 356return e->delta_size_; 357 358/* 359 * pack->detla_size[] can't be NULL because oe_set_delta_size() 360 * must have been called when a new delta is saved with 361 * oe_set_delta(). 362 * If oe_delta() returns NULL (i.e. default state, which means 363 * delta_size_valid is also false), then the caller must never 364 * call oe_delta_size(). 365 */ 366return pack->delta_size[e - pack->objects]; 367} 368 369staticinlinevoidoe_set_delta_size(struct packing_data *pack, 370struct object_entry *e, 371unsigned long size) 372{ 373if(size < pack->oe_delta_size_limit) { 374 e->delta_size_ = size; 375 e->delta_size_valid =1; 376}else{ 377packing_data_lock(pack); 378if(!pack->delta_size) 379ALLOC_ARRAY(pack->delta_size, pack->nr_alloc); 380packing_data_unlock(pack); 381 382 pack->delta_size[e - pack->objects] = size; 383 e->delta_size_valid =0; 384} 385} 386 387#endif