Merge branch 'sg/t4051-fix'
[git.git] / split-index.c
blob84f067e10d2cec1d77ec1a51f0f38cc5a29dd89f
1 #include "cache.h"
2 #include "split-index.h"
3 #include "ewah/ewok.h"
5 struct split_index *init_split_index(struct index_state *istate)
7 if (!istate->split_index) {
8 istate->split_index = xcalloc(1, sizeof(*istate->split_index));
9 istate->split_index->refcount = 1;
11 return istate->split_index;
14 int read_link_extension(struct index_state *istate,
15 const void *data_, unsigned long sz)
17 const unsigned char *data = data_;
18 struct split_index *si;
19 int ret;
21 if (sz < the_hash_algo->rawsz)
22 return error("corrupt link extension (too short)");
23 si = init_split_index(istate);
24 hashcpy(si->base_oid.hash, data);
25 data += the_hash_algo->rawsz;
26 sz -= the_hash_algo->rawsz;
27 if (!sz)
28 return 0;
29 si->delete_bitmap = ewah_new();
30 ret = ewah_read_mmap(si->delete_bitmap, data, sz);
31 if (ret < 0)
32 return error("corrupt delete bitmap in link extension");
33 data += ret;
34 sz -= ret;
35 si->replace_bitmap = ewah_new();
36 ret = ewah_read_mmap(si->replace_bitmap, data, sz);
37 if (ret < 0)
38 return error("corrupt replace bitmap in link extension");
39 if (ret != sz)
40 return error("garbage at the end of link extension");
41 return 0;
44 int write_link_extension(struct strbuf *sb,
45 struct index_state *istate)
47 struct split_index *si = istate->split_index;
48 strbuf_add(sb, si->base_oid.hash, the_hash_algo->rawsz);
49 if (!si->delete_bitmap && !si->replace_bitmap)
50 return 0;
51 ewah_serialize_strbuf(si->delete_bitmap, sb);
52 ewah_serialize_strbuf(si->replace_bitmap, sb);
53 return 0;
56 static void mark_base_index_entries(struct index_state *base)
58 int i;
60 * To keep track of the shared entries between
61 * istate->base->cache[] and istate->cache[], base entry
62 * position is stored in each base entry. All positions start
63 * from 1 instead of 0, which is reserved to say "this is a new
64 * entry".
66 for (i = 0; i < base->cache_nr; i++)
67 base->cache[i]->index = i + 1;
70 void move_cache_to_base_index(struct index_state *istate)
72 struct split_index *si = istate->split_index;
73 int i;
76 * If there was a previous base index, then transfer ownership of allocated
77 * entries to the parent index.
79 if (si->base &&
80 si->base->ce_mem_pool) {
82 if (!istate->ce_mem_pool)
83 mem_pool_init(&istate->ce_mem_pool, 0);
85 mem_pool_combine(istate->ce_mem_pool, istate->split_index->base->ce_mem_pool);
88 si->base = xcalloc(1, sizeof(*si->base));
89 si->base->version = istate->version;
90 /* zero timestamp disables racy test in ce_write_index() */
91 si->base->timestamp = istate->timestamp;
92 ALLOC_GROW(si->base->cache, istate->cache_nr, si->base->cache_alloc);
93 si->base->cache_nr = istate->cache_nr;
96 * The mem_pool needs to move with the allocated entries.
98 si->base->ce_mem_pool = istate->ce_mem_pool;
99 istate->ce_mem_pool = NULL;
101 COPY_ARRAY(si->base->cache, istate->cache, istate->cache_nr);
102 mark_base_index_entries(si->base);
103 for (i = 0; i < si->base->cache_nr; i++)
104 si->base->cache[i]->ce_flags &= ~CE_UPDATE_IN_BASE;
107 static void mark_entry_for_delete(size_t pos, void *data)
109 struct index_state *istate = data;
110 if (pos >= istate->cache_nr)
111 die("position for delete %d exceeds base index size %d",
112 (int)pos, istate->cache_nr);
113 istate->cache[pos]->ce_flags |= CE_REMOVE;
114 istate->split_index->nr_deletions = 1;
117 static void replace_entry(size_t pos, void *data)
119 struct index_state *istate = data;
120 struct split_index *si = istate->split_index;
121 struct cache_entry *dst, *src;
123 if (pos >= istate->cache_nr)
124 die("position for replacement %d exceeds base index size %d",
125 (int)pos, istate->cache_nr);
126 if (si->nr_replacements >= si->saved_cache_nr)
127 die("too many replacements (%d vs %d)",
128 si->nr_replacements, si->saved_cache_nr);
129 dst = istate->cache[pos];
130 if (dst->ce_flags & CE_REMOVE)
131 die("entry %d is marked as both replaced and deleted",
132 (int)pos);
133 src = si->saved_cache[si->nr_replacements];
134 if (ce_namelen(src))
135 die("corrupt link extension, entry %d should have "
136 "zero length name", (int)pos);
137 src->index = pos + 1;
138 src->ce_flags |= CE_UPDATE_IN_BASE;
139 src->ce_namelen = dst->ce_namelen;
140 copy_cache_entry(dst, src);
141 discard_cache_entry(src);
142 si->nr_replacements++;
145 void merge_base_index(struct index_state *istate)
147 struct split_index *si = istate->split_index;
148 unsigned int i;
150 mark_base_index_entries(si->base);
152 si->saved_cache = istate->cache;
153 si->saved_cache_nr = istate->cache_nr;
154 istate->cache_nr = si->base->cache_nr;
155 istate->cache = NULL;
156 istate->cache_alloc = 0;
157 ALLOC_GROW(istate->cache, istate->cache_nr, istate->cache_alloc);
158 COPY_ARRAY(istate->cache, si->base->cache, istate->cache_nr);
160 si->nr_deletions = 0;
161 si->nr_replacements = 0;
162 ewah_each_bit(si->replace_bitmap, replace_entry, istate);
163 ewah_each_bit(si->delete_bitmap, mark_entry_for_delete, istate);
164 if (si->nr_deletions)
165 remove_marked_cache_entries(istate);
167 for (i = si->nr_replacements; i < si->saved_cache_nr; i++) {
168 if (!ce_namelen(si->saved_cache[i]))
169 die("corrupt link extension, entry %d should "
170 "have non-zero length name", i);
171 add_index_entry(istate, si->saved_cache[i],
172 ADD_CACHE_OK_TO_ADD |
173 ADD_CACHE_KEEP_CACHE_TREE |
175 * we may have to replay what
176 * merge-recursive.c:update_stages()
177 * does, which has this flag on
179 ADD_CACHE_SKIP_DFCHECK);
180 si->saved_cache[i] = NULL;
183 ewah_free(si->delete_bitmap);
184 ewah_free(si->replace_bitmap);
185 FREE_AND_NULL(si->saved_cache);
186 si->delete_bitmap = NULL;
187 si->replace_bitmap = NULL;
188 si->saved_cache_nr = 0;
191 void prepare_to_write_split_index(struct index_state *istate)
193 struct split_index *si = init_split_index(istate);
194 struct cache_entry **entries = NULL, *ce;
195 int i, nr_entries = 0, nr_alloc = 0;
197 si->delete_bitmap = ewah_new();
198 si->replace_bitmap = ewah_new();
200 if (si->base) {
201 /* Go through istate->cache[] and mark CE_MATCHED to
202 * entry with positive index. We'll go through
203 * base->cache[] later to delete all entries in base
204 * that are not marked with either CE_MATCHED or
205 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
206 * duplicate, deduplicate it.
208 for (i = 0; i < istate->cache_nr; i++) {
209 struct cache_entry *base;
210 /* namelen is checked separately */
211 const unsigned int ondisk_flags =
212 CE_STAGEMASK | CE_VALID | CE_EXTENDED_FLAGS;
213 unsigned int ce_flags, base_flags, ret;
214 ce = istate->cache[i];
215 if (!ce->index)
216 continue;
217 if (ce->index > si->base->cache_nr) {
218 ce->index = 0;
219 continue;
221 ce->ce_flags |= CE_MATCHED; /* or "shared" */
222 base = si->base->cache[ce->index - 1];
223 if (ce == base)
224 continue;
225 if (ce->ce_namelen != base->ce_namelen ||
226 strcmp(ce->name, base->name)) {
227 ce->index = 0;
228 continue;
230 ce_flags = ce->ce_flags;
231 base_flags = base->ce_flags;
232 /* only on-disk flags matter */
233 ce->ce_flags &= ondisk_flags;
234 base->ce_flags &= ondisk_flags;
235 ret = memcmp(&ce->ce_stat_data, &base->ce_stat_data,
236 offsetof(struct cache_entry, name) -
237 offsetof(struct cache_entry, ce_stat_data));
238 ce->ce_flags = ce_flags;
239 base->ce_flags = base_flags;
240 if (ret)
241 ce->ce_flags |= CE_UPDATE_IN_BASE;
242 discard_cache_entry(base);
243 si->base->cache[ce->index - 1] = ce;
245 for (i = 0; i < si->base->cache_nr; i++) {
246 ce = si->base->cache[i];
247 if ((ce->ce_flags & CE_REMOVE) ||
248 !(ce->ce_flags & CE_MATCHED))
249 ewah_set(si->delete_bitmap, i);
250 else if (ce->ce_flags & CE_UPDATE_IN_BASE) {
251 ewah_set(si->replace_bitmap, i);
252 ce->ce_flags |= CE_STRIP_NAME;
253 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
254 entries[nr_entries++] = ce;
256 if (is_null_oid(&ce->oid))
257 istate->drop_cache_tree = 1;
261 for (i = 0; i < istate->cache_nr; i++) {
262 ce = istate->cache[i];
263 if ((!si->base || !ce->index) && !(ce->ce_flags & CE_REMOVE)) {
264 assert(!(ce->ce_flags & CE_STRIP_NAME));
265 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
266 entries[nr_entries++] = ce;
268 ce->ce_flags &= ~CE_MATCHED;
272 * take cache[] out temporarily, put entries[] in its place
273 * for writing
275 si->saved_cache = istate->cache;
276 si->saved_cache_nr = istate->cache_nr;
277 istate->cache = entries;
278 istate->cache_nr = nr_entries;
281 void finish_writing_split_index(struct index_state *istate)
283 struct split_index *si = init_split_index(istate);
285 ewah_free(si->delete_bitmap);
286 ewah_free(si->replace_bitmap);
287 si->delete_bitmap = NULL;
288 si->replace_bitmap = NULL;
289 free(istate->cache);
290 istate->cache = si->saved_cache;
291 istate->cache_nr = si->saved_cache_nr;
294 void discard_split_index(struct index_state *istate)
296 struct split_index *si = istate->split_index;
297 if (!si)
298 return;
299 istate->split_index = NULL;
300 si->refcount--;
301 if (si->refcount)
302 return;
303 if (si->base) {
304 discard_index(si->base);
305 free(si->base);
307 free(si);
310 void save_or_free_index_entry(struct index_state *istate, struct cache_entry *ce)
312 if (ce->index &&
313 istate->split_index &&
314 istate->split_index->base &&
315 ce->index <= istate->split_index->base->cache_nr &&
316 ce == istate->split_index->base->cache[ce->index - 1])
317 ce->ce_flags |= CE_REMOVE;
318 else
319 discard_cache_entry(ce);
322 void replace_index_entry_in_base(struct index_state *istate,
323 struct cache_entry *old_entry,
324 struct cache_entry *new_entry)
326 if (old_entry->index &&
327 istate->split_index &&
328 istate->split_index->base &&
329 old_entry->index <= istate->split_index->base->cache_nr) {
330 new_entry->index = old_entry->index;
331 if (old_entry != istate->split_index->base->cache[new_entry->index - 1])
332 discard_cache_entry(istate->split_index->base->cache[new_entry->index - 1]);
333 istate->split_index->base->cache[new_entry->index - 1] = new_entry;
337 void add_split_index(struct index_state *istate)
339 if (!istate->split_index) {
340 init_split_index(istate);
341 istate->cache_changed |= SPLIT_INDEX_ORDERED;
345 void remove_split_index(struct index_state *istate)
347 if (istate->split_index) {
349 * When removing the split index, we need to move
350 * ownership of the mem_pool associated with the
351 * base index to the main index. There may be cache entries
352 * allocated from the base's memory pool that are shared with
353 * the_index.cache[].
355 mem_pool_combine(istate->ce_mem_pool, istate->split_index->base->ce_mem_pool);
358 * The split index no longer owns the mem_pool backing
359 * its cache array. As we are discarding this index,
360 * mark the index as having no cache entries, so it
361 * will not attempt to clean up the cache entries or
362 * validate them.
364 if (istate->split_index->base)
365 istate->split_index->base->cache_nr = 0;
368 * We can discard the split index because its
369 * memory pool has been incorporated into the
370 * memory pool associated with the the_index.
372 discard_split_index(istate);
374 istate->cache_changed |= SOMETHING_CHANGED;