Merge branch 'nd/pthreads'
[git/git.git] / pack-objects.h
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 */
27 enum 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 */
85 struct object_entry {
86 struct pack_idx_entry idx;
87 void *delta_data; /* cached delta (uncompressed) */
88 off_t in_pack_offset;
89 uint32_t hash; /* name hint hash */
90 unsigned size_:OE_SIZE_BITS;
91 unsigned size_valid:1;
92 uint32_t delta_idx; /* delta base object */
93 uint32_t delta_child_idx; /* deltified objects who bases me */
94 uint32_t delta_sibling_idx; /* other deltified objects who
95 * uses the same base as me
96 */
97 unsigned delta_size_:OE_DELTA_SIZE_BITS; /* delta data size (uncompressed) */
98 unsigned delta_size_valid:1;
99 unsigned char in_pack_header_size;
100 unsigned in_pack_idx:OE_IN_PACK_BITS; /* already in pack */
101 unsigned z_delta_size:OE_Z_DELTA_BITS;
102 unsigned type_valid:1;
103 unsigned no_try_delta:1;
104 unsigned type_:TYPE_BITS;
105 unsigned in_pack_type:TYPE_BITS; /* could be delta */
106
107 unsigned preferred_base:1; /*
108 * we do not pack this, but is available
109 * to be used as the base object to delta
110 * objects against.
111 */
112 unsigned tagged:1; /* near the very tip of refs */
113 unsigned filled:1; /* assigned write-order */
114 unsigned dfs_state:OE_DFS_STATE_BITS;
115 unsigned depth:OE_DEPTH_BITS;
116 unsigned ext_base:1; /* delta_idx points outside packlist */
117
118 /*
119 * pahole results on 64-bit linux (gcc and clang)
120 *
121 * size: 80, bit_padding: 9 bits
122 *
123 * and on 32-bit (gcc)
124 *
125 * size: 76, bit_padding: 9 bits
126 */
127 };
128
129 struct packing_data {
130 struct object_entry *objects;
131 uint32_t nr_objects, nr_alloc;
132
133 int32_t *index;
134 uint32_t index_size;
135
136 unsigned int *in_pack_pos;
137 unsigned long *delta_size;
138
139 /*
140 * Only one of these can be non-NULL and they have different
141 * sizes. if in_pack_by_idx is allocated, oe_in_pack() returns
142 * the pack of an object using in_pack_idx field. If not,
143 * in_pack[] array is used the same way as in_pack_pos[]
144 */
145 struct packed_git **in_pack_by_idx;
146 struct packed_git **in_pack;
147
148 pthread_mutex_t lock;
149
150 /*
151 * This list contains entries for bases which we know the other side
152 * has (e.g., via reachability bitmaps), but which aren't in our
153 * "objects" list.
154 */
155 struct object_entry *ext_bases;
156 uint32_t nr_ext, alloc_ext;
157
158 uintmax_t oe_size_limit;
159 uintmax_t oe_delta_size_limit;
160
161 /* delta islands */
162 unsigned int *tree_depth;
163 unsigned char *layer;
164 };
165
166 void prepare_packing_data(struct packing_data *pdata);
167
168 static inline void packing_data_lock(struct packing_data *pdata)
169 {
170 pthread_mutex_lock(&pdata->lock);
171 }
172 static inline void packing_data_unlock(struct packing_data *pdata)
173 {
174 pthread_mutex_unlock(&pdata->lock);
175 }
176
177 struct object_entry *packlist_alloc(struct packing_data *pdata,
178 const unsigned char *sha1,
179 uint32_t index_pos);
180
181 struct object_entry *packlist_find(struct packing_data *pdata,
182 const unsigned char *sha1,
183 uint32_t *index_pos);
184
185 static inline uint32_t pack_name_hash(const char *name)
186 {
187 uint32_t c, hash = 0;
188
189 if (!name)
190 return 0;
191
192 /*
193 * This effectively just creates a sortable number from the
194 * last sixteen non-whitespace characters. Last characters
195 * count "most", so things that end in ".c" sort together.
196 */
197 while ((c = *name++) != 0) {
198 if (isspace(c))
199 continue;
200 hash = (hash >> 2) + (c << 24);
201 }
202 return hash;
203 }
204
205 static inline enum object_type oe_type(const struct object_entry *e)
206 {
207 return e->type_valid ? e->type_ : OBJ_BAD;
208 }
209
210 static inline void oe_set_type(struct object_entry *e,
211 enum object_type type)
212 {
213 if (type >= OBJ_ANY)
214 BUG("OBJ_ANY cannot be set in pack-objects code");
215
216 e->type_valid = type >= OBJ_NONE;
217 e->type_ = (unsigned)type;
218 }
219
220 static inline unsigned int oe_in_pack_pos(const struct packing_data *pack,
221 const struct object_entry *e)
222 {
223 return pack->in_pack_pos[e - pack->objects];
224 }
225
226 static inline void oe_set_in_pack_pos(const struct packing_data *pack,
227 const struct object_entry *e,
228 unsigned int pos)
229 {
230 pack->in_pack_pos[e - pack->objects] = pos;
231 }
232
233 static inline struct packed_git *oe_in_pack(const struct packing_data *pack,
234 const struct object_entry *e)
235 {
236 if (pack->in_pack_by_idx)
237 return pack->in_pack_by_idx[e->in_pack_idx];
238 else
239 return pack->in_pack[e - pack->objects];
240 }
241
242 void oe_map_new_pack(struct packing_data *pack,
243 struct packed_git *p);
244 static inline void oe_set_in_pack(struct packing_data *pack,
245 struct object_entry *e,
246 struct packed_git *p)
247 {
248 if (!p->index)
249 oe_map_new_pack(pack, p);
250 if (pack->in_pack_by_idx)
251 e->in_pack_idx = p->index;
252 else
253 pack->in_pack[e - pack->objects] = p;
254 }
255
256 static inline struct object_entry *oe_delta(
257 const struct packing_data *pack,
258 const struct object_entry *e)
259 {
260 if (!e->delta_idx)
261 return NULL;
262 if (e->ext_base)
263 return &pack->ext_bases[e->delta_idx - 1];
264 else
265 return &pack->objects[e->delta_idx - 1];
266 }
267
268 static inline void oe_set_delta(struct packing_data *pack,
269 struct object_entry *e,
270 struct object_entry *delta)
271 {
272 if (delta)
273 e->delta_idx = (delta - pack->objects) + 1;
274 else
275 e->delta_idx = 0;
276 }
277
278 void oe_set_delta_ext(struct packing_data *pack,
279 struct object_entry *e,
280 const unsigned char *sha1);
281
282 static inline struct object_entry *oe_delta_child(
283 const struct packing_data *pack,
284 const struct object_entry *e)
285 {
286 if (e->delta_child_idx)
287 return &pack->objects[e->delta_child_idx - 1];
288 return NULL;
289 }
290
291 static inline void oe_set_delta_child(struct packing_data *pack,
292 struct object_entry *e,
293 struct object_entry *delta)
294 {
295 if (delta)
296 e->delta_child_idx = (delta - pack->objects) + 1;
297 else
298 e->delta_child_idx = 0;
299 }
300
301 static inline struct object_entry *oe_delta_sibling(
302 const struct packing_data *pack,
303 const struct object_entry *e)
304 {
305 if (e->delta_sibling_idx)
306 return &pack->objects[e->delta_sibling_idx - 1];
307 return NULL;
308 }
309
310 static inline void oe_set_delta_sibling(struct packing_data *pack,
311 struct object_entry *e,
312 struct object_entry *delta)
313 {
314 if (delta)
315 e->delta_sibling_idx = (delta - pack->objects) + 1;
316 else
317 e->delta_sibling_idx = 0;
318 }
319
320 unsigned long oe_get_size_slow(struct packing_data *pack,
321 const struct object_entry *e);
322 static inline unsigned long oe_size(struct packing_data *pack,
323 const struct object_entry *e)
324 {
325 if (e->size_valid)
326 return e->size_;
327
328 return oe_get_size_slow(pack, e);
329 }
330
331 static inline int oe_size_less_than(struct packing_data *pack,
332 const struct object_entry *lhs,
333 unsigned long rhs)
334 {
335 if (lhs->size_valid)
336 return lhs->size_ < rhs;
337 if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
338 return 0;
339 return oe_get_size_slow(pack, lhs) < rhs;
340 }
341
342 static inline int oe_size_greater_than(struct packing_data *pack,
343 const struct object_entry *lhs,
344 unsigned long rhs)
345 {
346 if (lhs->size_valid)
347 return lhs->size_ > rhs;
348 if (rhs < pack->oe_size_limit) /* rhs < 2^x <= lhs ? */
349 return 1;
350 return oe_get_size_slow(pack, lhs) > rhs;
351 }
352
353 static inline void oe_set_size(struct packing_data *pack,
354 struct object_entry *e,
355 unsigned long size)
356 {
357 if (size < pack->oe_size_limit) {
358 e->size_ = size;
359 e->size_valid = 1;
360 } else {
361 e->size_valid = 0;
362 if (oe_get_size_slow(pack, e) != size)
363 BUG("'size' is supposed to be the object size!");
364 }
365 }
366
367 static inline unsigned long oe_delta_size(struct packing_data *pack,
368 const struct object_entry *e)
369 {
370 if (e->delta_size_valid)
371 return e->delta_size_;
372
373 /*
374 * pack->delta_size[] can't be NULL because oe_set_delta_size()
375 * must have been called when a new delta is saved with
376 * oe_set_delta().
377 * If oe_delta() returns NULL (i.e. default state, which means
378 * delta_size_valid is also false), then the caller must never
379 * call oe_delta_size().
380 */
381 return pack->delta_size[e - pack->objects];
382 }
383
384 static inline void oe_set_delta_size(struct packing_data *pack,
385 struct object_entry *e,
386 unsigned long size)
387 {
388 if (size < pack->oe_delta_size_limit) {
389 e->delta_size_ = size;
390 e->delta_size_valid = 1;
391 } else {
392 packing_data_lock(pack);
393 if (!pack->delta_size)
394 ALLOC_ARRAY(pack->delta_size, pack->nr_alloc);
395 packing_data_unlock(pack);
396
397 pack->delta_size[e - pack->objects] = size;
398 e->delta_size_valid = 0;
399 }
400 }
401
402 static inline unsigned int oe_tree_depth(struct packing_data *pack,
403 struct object_entry *e)
404 {
405 if (!pack->tree_depth)
406 return 0;
407 return pack->tree_depth[e - pack->objects];
408 }
409
410 static inline void oe_set_tree_depth(struct packing_data *pack,
411 struct object_entry *e,
412 unsigned int tree_depth)
413 {
414 if (!pack->tree_depth)
415 ALLOC_ARRAY(pack->tree_depth, pack->nr_objects);
416 pack->tree_depth[e - pack->objects] = tree_depth;
417 }
418
419 static inline unsigned char oe_layer(struct packing_data *pack,
420 struct object_entry *e)
421 {
422 if (!pack->layer)
423 return 0;
424 return pack->layer[e - pack->objects];
425 }
426
427 static inline void oe_set_layer(struct packing_data *pack,
428 struct object_entry *e,
429 unsigned char layer)
430 {
431 if (!pack->layer)
432 ALLOC_ARRAY(pack->layer, pack->nr_objects);
433 pack->layer[e - pack->objects] = layer;
434 }
435
436 #endif