Merge branch 'jk/t9001-deflake'
[git/gitster.git] / pack-objects.h
blobb9898a4e64b8b4d53b21ea776c16f79d9794efef
1 #ifndef PACK_OBJECTS_H
2 #define PACK_OBJECTS_H
4 #include "object-store-ll.h"
5 #include "thread-utils.h"
6 #include "pack.h"
8 struct repository;
10 #define DEFAULT_DELTA_CACHE_SIZE (256 * 1024 * 1024)
12 #define OE_DFS_STATE_BITS 2
13 #define OE_DEPTH_BITS 12
14 #define OE_IN_PACK_BITS 10
15 #define OE_Z_DELTA_BITS 20
17 * Note that oe_set_size() becomes expensive when the given size is
18 * above this limit. Don't lower it too much.
20 #define OE_SIZE_BITS 31
21 #define OE_DELTA_SIZE_BITS 23
24 * State flags for depth-first search used for analyzing delta cycles.
26 * The depth is measured in delta-links to the base (so if A is a delta
27 * against B, then A has a depth of 1, and B a depth of 0).
29 enum dfs_state {
30 DFS_NONE = 0,
31 DFS_ACTIVE,
32 DFS_DONE,
33 DFS_NUM_STATES
37 * The size of struct nearly determines pack-objects's memory
38 * consumption. This struct is packed tight for that reason. When you
39 * add or reorder something in this struct, think a bit about this.
41 * basic object info
42 * -----------------
43 * idx.oid is filled up before delta searching starts. idx.crc32 is
44 * only valid after the object is written out and will be used for
45 * generating the index. idx.offset will be both gradually set and
46 * used in writing phase (base objects get offset first, then deltas
47 * refer to them)
49 * "size" is the uncompressed object size. Compressed size of the raw
50 * data for an object in a pack is not stored anywhere but is computed
51 * and made available when reverse .idx is made. Note that when a
52 * delta is reused, "size" is the uncompressed _delta_ size, not the
53 * canonical one after the delta has been applied.
55 * "hash" contains a path name hash which is used for sorting the
56 * delta list and also during delta searching. Once prepare_pack()
57 * returns it's no longer needed.
59 * source pack info
60 * ----------------
61 * The (in_pack, in_pack_offset) tuple contains the location of the
62 * object in the source pack. in_pack_header_size allows quickly
63 * skipping the header and going straight to the zlib stream.
65 * "type" and "in_pack_type" both describe object type. in_pack_type
66 * may contain a delta type, while type is always the canonical type.
68 * deltas
69 * ------
70 * Delta links (delta, delta_child and delta_sibling) are created to
71 * reflect that delta graph from the source pack then updated or added
72 * during delta searching phase when we find better deltas.
74 * delta_child and delta_sibling are last needed in
75 * compute_write_order(). "delta" and "delta_size" must remain valid
76 * at object writing phase in case the delta is not cached.
78 * If a delta is cached in memory and is compressed, delta_data points
79 * to the data and z_delta_size contains the compressed size. If it's
80 * uncompressed [1], z_delta_size must be zero. delta_size is always
81 * the uncompressed size and must be valid even if the delta is not
82 * cached.
84 * [1] during try_delta phase we don't bother with compressing because
85 * the delta could be quickly replaced with a better one.
87 struct object_entry {
88 struct pack_idx_entry idx;
89 void *delta_data; /* cached delta (uncompressed) */
90 off_t in_pack_offset;
91 uint32_t hash; /* name hint hash */
92 unsigned size_:OE_SIZE_BITS;
93 unsigned size_valid:1;
94 uint32_t delta_idx; /* delta base object */
95 uint32_t delta_child_idx; /* deltified objects who bases me */
96 uint32_t delta_sibling_idx; /* other deltified objects who
97 * uses the same base as me
99 unsigned delta_size_:OE_DELTA_SIZE_BITS; /* delta data size (uncompressed) */
100 unsigned delta_size_valid:1;
101 unsigned char in_pack_header_size;
102 unsigned in_pack_idx:OE_IN_PACK_BITS; /* already in pack */
103 unsigned z_delta_size:OE_Z_DELTA_BITS;
104 unsigned type_valid:1;
105 unsigned no_try_delta:1;
106 unsigned type_:TYPE_BITS;
107 unsigned in_pack_type:TYPE_BITS; /* could be delta */
109 unsigned preferred_base:1; /*
110 * we do not pack this, but is available
111 * to be used as the base object to delta
112 * objects against.
114 unsigned tagged:1; /* near the very tip of refs */
115 unsigned filled:1; /* assigned write-order */
116 unsigned dfs_state:OE_DFS_STATE_BITS;
117 unsigned depth:OE_DEPTH_BITS;
118 unsigned ext_base:1; /* delta_idx points outside packlist */
121 struct packing_data {
122 struct repository *repo;
123 struct object_entry *objects;
124 uint32_t nr_objects, nr_alloc;
126 int32_t *index;
127 uint32_t index_size;
129 unsigned int *in_pack_pos;
130 unsigned long *delta_size;
133 * Only one of these can be non-NULL and they have different
134 * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns
135 * the pack of an object using in_pack_idx field. If not,
136 * in_pack[] array is used the same way as in_pack_pos[]
138 struct packed_git **in_pack_by_idx;
139 struct packed_git **in_pack;
142 * During packing with multiple threads, protect the in-core
143 * object database from concurrent accesses.
145 pthread_mutex_t odb_lock;
148 * This list contains entries for bases which we know the other side
149 * has (e.g., via reachability bitmaps), but which aren't in our
150 * "objects" list.
152 struct object_entry *ext_bases;
153 uint32_t nr_ext, alloc_ext;
155 uintmax_t oe_size_limit;
156 uintmax_t oe_delta_size_limit;
158 /* delta islands */
159 unsigned int *tree_depth;
160 unsigned char *layer;
163 * Used when writing cruft packs.
165 * Object mtimes are stored in pack order when writing, but
166 * written out in lexicographic (index) order.
168 uint32_t *cruft_mtime;
171 void prepare_packing_data(struct repository *r, struct packing_data *pdata);
172 void clear_packing_data(struct packing_data *pdata);
174 /* Protect access to object database */
175 static inline void packing_data_lock(struct packing_data *pdata)
177 pthread_mutex_lock(&pdata->odb_lock);
179 static inline void packing_data_unlock(struct packing_data *pdata)
181 pthread_mutex_unlock(&pdata->odb_lock);
184 struct object_entry *packlist_alloc(struct packing_data *pdata,
185 const struct object_id *oid);
187 struct object_entry *packlist_find(struct packing_data *pdata,
188 const struct object_id *oid);
190 static inline uint32_t pack_name_hash(const char *name)
192 uint32_t c, hash = 0;
194 if (!name)
195 return 0;
198 * This effectively just creates a sortable number from the
199 * last sixteen non-whitespace characters. Last characters
200 * count "most", so things that end in ".c" sort together.
202 while ((c = *name++) != 0) {
203 if (isspace(c))
204 continue;
205 hash = (hash >> 2) + (c << 24);
207 return hash;
210 static inline enum object_type oe_type(const struct object_entry *e)
212 return e->type_valid ? e->type_ : OBJ_BAD;
215 static inline void oe_set_type(struct object_entry *e,
216 enum object_type type)
218 if (type >= OBJ_ANY)
219 BUG("OBJ_ANY cannot be set in pack-objects code");
221 e->type_valid = type >= OBJ_NONE;
222 e->type_ = (unsigned)type;
225 static inline unsigned int oe_in_pack_pos(const struct packing_data *pack,
226 const struct object_entry *e)
228 return pack->in_pack_pos[e - pack->objects];
231 static inline void oe_set_in_pack_pos(const struct packing_data *pack,
232 const struct object_entry *e,
233 unsigned int pos)
235 pack->in_pack_pos[e - pack->objects] = pos;
238 static inline struct packed_git *oe_in_pack(const struct packing_data *pack,
239 const struct object_entry *e)
241 if (pack->in_pack_by_idx)
242 return pack->in_pack_by_idx[e->in_pack_idx];
243 else
244 return pack->in_pack[e - pack->objects];
247 void oe_map_new_pack(struct packing_data *pack);
249 static inline void oe_set_in_pack(struct packing_data *pack,
250 struct object_entry *e,
251 struct packed_git *p)
253 if (pack->in_pack_by_idx) {
254 if (p->index) {
255 e->in_pack_idx = p->index;
256 return;
259 * We're accessing packs by index, but this pack doesn't have
260 * an index (e.g., because it was added since we created the
261 * in_pack_by_idx array). Bail to oe_map_new_pack(), which
262 * will convert us to using the full in_pack array, and then
263 * fall through to our in_pack handling.
265 oe_map_new_pack(pack);
267 pack->in_pack[e - pack->objects] = p;
270 void oe_set_delta_ext(struct packing_data *pack,
271 struct object_entry *e,
272 const struct object_id *oid);
274 static inline unsigned int oe_tree_depth(struct packing_data *pack,
275 struct object_entry *e)
277 if (!pack->tree_depth)
278 return 0;
279 return pack->tree_depth[e - pack->objects];
282 static inline void oe_set_layer(struct packing_data *pack,
283 struct object_entry *e,
284 unsigned char layer)
286 if (!pack->layer)
287 CALLOC_ARRAY(pack->layer, pack->nr_alloc);
288 pack->layer[e - pack->objects] = layer;
291 static inline uint32_t oe_cruft_mtime(struct packing_data *pack,
292 struct object_entry *e)
294 if (!pack->cruft_mtime)
295 return 0;
296 return pack->cruft_mtime[e - pack->objects];
299 static inline void oe_set_cruft_mtime(struct packing_data *pack,
300 struct object_entry *e,
301 uint32_t mtime)
303 if (!pack->cruft_mtime)
304 CALLOC_ARRAY(pack->cruft_mtime, pack->nr_alloc);
305 pack->cruft_mtime[e - pack->objects] = mtime;
308 #endif