Merge branch 'rs/pcre2-utf'
[git/debian.git] / split-index.c
blob8e52e891c3bc34d35eab1124302d6f0294d9cedb
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 CALLOC_ARRAY(istate->split_index, 1);
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 oidread(&si->base_oid, 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 istate->ce_mem_pool = xmalloc(sizeof(struct mem_pool));
84 mem_pool_init(istate->ce_mem_pool, 0);
87 mem_pool_combine(istate->ce_mem_pool, istate->split_index->base->ce_mem_pool);
90 CALLOC_ARRAY(si->base, 1);
91 si->base->version = istate->version;
92 /* zero timestamp disables racy test in ce_write_index() */
93 si->base->timestamp = istate->timestamp;
94 ALLOC_GROW(si->base->cache, istate->cache_nr, si->base->cache_alloc);
95 si->base->cache_nr = istate->cache_nr;
98 * The mem_pool needs to move with the allocated entries.
100 si->base->ce_mem_pool = istate->ce_mem_pool;
101 istate->ce_mem_pool = NULL;
103 COPY_ARRAY(si->base->cache, istate->cache, istate->cache_nr);
104 mark_base_index_entries(si->base);
105 for (i = 0; i < si->base->cache_nr; i++)
106 si->base->cache[i]->ce_flags &= ~CE_UPDATE_IN_BASE;
109 static void mark_entry_for_delete(size_t pos, void *data)
111 struct index_state *istate = data;
112 if (pos >= istate->cache_nr)
113 die("position for delete %d exceeds base index size %d",
114 (int)pos, istate->cache_nr);
115 istate->cache[pos]->ce_flags |= CE_REMOVE;
116 istate->split_index->nr_deletions++;
119 static void replace_entry(size_t pos, void *data)
121 struct index_state *istate = data;
122 struct split_index *si = istate->split_index;
123 struct cache_entry *dst, *src;
125 if (pos >= istate->cache_nr)
126 die("position for replacement %d exceeds base index size %d",
127 (int)pos, istate->cache_nr);
128 if (si->nr_replacements >= si->saved_cache_nr)
129 die("too many replacements (%d vs %d)",
130 si->nr_replacements, si->saved_cache_nr);
131 dst = istate->cache[pos];
132 if (dst->ce_flags & CE_REMOVE)
133 die("entry %d is marked as both replaced and deleted",
134 (int)pos);
135 src = si->saved_cache[si->nr_replacements];
136 if (ce_namelen(src))
137 die("corrupt link extension, entry %d should have "
138 "zero length name", (int)pos);
139 src->index = pos + 1;
140 src->ce_flags |= CE_UPDATE_IN_BASE;
141 src->ce_namelen = dst->ce_namelen;
142 copy_cache_entry(dst, src);
143 discard_cache_entry(src);
144 si->nr_replacements++;
147 void merge_base_index(struct index_state *istate)
149 struct split_index *si = istate->split_index;
150 unsigned int i;
152 mark_base_index_entries(si->base);
154 si->saved_cache = istate->cache;
155 si->saved_cache_nr = istate->cache_nr;
156 istate->cache_nr = si->base->cache_nr;
157 istate->cache = NULL;
158 istate->cache_alloc = 0;
159 ALLOC_GROW(istate->cache, istate->cache_nr, istate->cache_alloc);
160 COPY_ARRAY(istate->cache, si->base->cache, istate->cache_nr);
162 si->nr_deletions = 0;
163 si->nr_replacements = 0;
164 ewah_each_bit(si->replace_bitmap, replace_entry, istate);
165 ewah_each_bit(si->delete_bitmap, mark_entry_for_delete, istate);
166 if (si->nr_deletions)
167 remove_marked_cache_entries(istate, 0);
169 for (i = si->nr_replacements; i < si->saved_cache_nr; i++) {
170 if (!ce_namelen(si->saved_cache[i]))
171 die("corrupt link extension, entry %d should "
172 "have non-zero length name", i);
173 add_index_entry(istate, si->saved_cache[i],
174 ADD_CACHE_OK_TO_ADD |
175 ADD_CACHE_KEEP_CACHE_TREE |
177 * we may have to replay what
178 * merge-recursive.c:update_stages()
179 * does, which has this flag on
181 ADD_CACHE_SKIP_DFCHECK);
182 si->saved_cache[i] = NULL;
185 ewah_free(si->delete_bitmap);
186 ewah_free(si->replace_bitmap);
187 FREE_AND_NULL(si->saved_cache);
188 si->delete_bitmap = NULL;
189 si->replace_bitmap = NULL;
190 si->saved_cache_nr = 0;
194 * Compare most of the fields in two cache entries, i.e. all except the
195 * hashmap_entry and the name.
197 static int compare_ce_content(struct cache_entry *a, struct cache_entry *b)
199 const unsigned int ondisk_flags = CE_STAGEMASK | CE_VALID |
200 CE_EXTENDED_FLAGS;
201 unsigned int ce_flags = a->ce_flags;
202 unsigned int base_flags = b->ce_flags;
203 int ret;
205 /* only on-disk flags matter */
206 a->ce_flags &= ondisk_flags;
207 b->ce_flags &= ondisk_flags;
208 ret = memcmp(&a->ce_stat_data, &b->ce_stat_data,
209 offsetof(struct cache_entry, name) -
210 offsetof(struct cache_entry, oid)) ||
211 !oideq(&a->oid, &b->oid);
212 a->ce_flags = ce_flags;
213 b->ce_flags = base_flags;
215 return ret;
218 void prepare_to_write_split_index(struct index_state *istate)
220 struct split_index *si = init_split_index(istate);
221 struct cache_entry **entries = NULL, *ce;
222 int i, nr_entries = 0, nr_alloc = 0;
224 si->delete_bitmap = ewah_new();
225 si->replace_bitmap = ewah_new();
227 if (si->base) {
228 /* Go through istate->cache[] and mark CE_MATCHED to
229 * entry with positive index. We'll go through
230 * base->cache[] later to delete all entries in base
231 * that are not marked with either CE_MATCHED or
232 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
233 * duplicate, deduplicate it.
235 for (i = 0; i < istate->cache_nr; i++) {
236 struct cache_entry *base;
237 ce = istate->cache[i];
238 if (!ce->index) {
240 * During simple update index operations this
241 * is a cache entry that is not present in
242 * the shared index. It will be added to the
243 * split index.
245 * However, it might also represent a file
246 * that already has a cache entry in the
247 * shared index, but a new index has just
248 * been constructed by unpack_trees(), and
249 * this entry now refers to different content
250 * than what was recorded in the original
251 * index, e.g. during 'read-tree -m HEAD^' or
252 * 'checkout HEAD^'. In this case the
253 * original entry in the shared index will be
254 * marked as deleted, and this entry will be
255 * added to the split index.
257 continue;
259 if (ce->index > si->base->cache_nr) {
260 BUG("ce refers to a shared ce at %d, which is beyond the shared index size %d",
261 ce->index, si->base->cache_nr);
263 ce->ce_flags |= CE_MATCHED; /* or "shared" */
264 base = si->base->cache[ce->index - 1];
265 if (ce == base) {
266 /* The entry is present in the shared index. */
267 if (ce->ce_flags & CE_UPDATE_IN_BASE) {
269 * Already marked for inclusion in
270 * the split index, either because
271 * the corresponding file was
272 * modified and the cached stat data
273 * was refreshed, or because there
274 * is already a replacement entry in
275 * the split index.
276 * Nothing more to do here.
278 } else if (!ce_uptodate(ce) &&
279 is_racy_timestamp(istate, ce)) {
281 * A racily clean cache entry stored
282 * only in the shared index: it must
283 * be added to the split index, so
284 * the subsequent do_write_index()
285 * can smudge its stat data.
287 ce->ce_flags |= CE_UPDATE_IN_BASE;
288 } else {
290 * The entry is only present in the
291 * shared index and it was not
292 * refreshed.
293 * Just leave it there.
296 continue;
298 if (ce->ce_namelen != base->ce_namelen ||
299 strcmp(ce->name, base->name)) {
300 ce->index = 0;
301 continue;
304 * This is the copy of a cache entry that is present
305 * in the shared index, created by unpack_trees()
306 * while it constructed a new index.
308 if (ce->ce_flags & CE_UPDATE_IN_BASE) {
310 * Already marked for inclusion in the split
311 * index, either because the corresponding
312 * file was modified and the cached stat data
313 * was refreshed, or because the original
314 * entry already had a replacement entry in
315 * the split index.
316 * Nothing to do.
318 } else if (!ce_uptodate(ce) &&
319 is_racy_timestamp(istate, ce)) {
321 * A copy of a racily clean cache entry from
322 * the shared index. It must be added to
323 * the split index, so the subsequent
324 * do_write_index() can smudge its stat data.
326 ce->ce_flags |= CE_UPDATE_IN_BASE;
327 } else {
329 * Thoroughly compare the cached data to see
330 * whether it should be marked for inclusion
331 * in the split index.
333 * This comparison might be unnecessary, as
334 * code paths modifying the cached data do
335 * set CE_UPDATE_IN_BASE as well.
337 if (compare_ce_content(ce, base))
338 ce->ce_flags |= CE_UPDATE_IN_BASE;
340 discard_cache_entry(base);
341 si->base->cache[ce->index - 1] = ce;
343 for (i = 0; i < si->base->cache_nr; i++) {
344 ce = si->base->cache[i];
345 if ((ce->ce_flags & CE_REMOVE) ||
346 !(ce->ce_flags & CE_MATCHED))
347 ewah_set(si->delete_bitmap, i);
348 else if (ce->ce_flags & CE_UPDATE_IN_BASE) {
349 ewah_set(si->replace_bitmap, i);
350 ce->ce_flags |= CE_STRIP_NAME;
351 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
352 entries[nr_entries++] = ce;
354 if (is_null_oid(&ce->oid))
355 istate->drop_cache_tree = 1;
359 for (i = 0; i < istate->cache_nr; i++) {
360 ce = istate->cache[i];
361 if ((!si->base || !ce->index) && !(ce->ce_flags & CE_REMOVE)) {
362 assert(!(ce->ce_flags & CE_STRIP_NAME));
363 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
364 entries[nr_entries++] = ce;
366 ce->ce_flags &= ~CE_MATCHED;
370 * take cache[] out temporarily, put entries[] in its place
371 * for writing
373 si->saved_cache = istate->cache;
374 si->saved_cache_nr = istate->cache_nr;
375 istate->cache = entries;
376 istate->cache_nr = nr_entries;
379 void finish_writing_split_index(struct index_state *istate)
381 struct split_index *si = init_split_index(istate);
383 ewah_free(si->delete_bitmap);
384 ewah_free(si->replace_bitmap);
385 si->delete_bitmap = NULL;
386 si->replace_bitmap = NULL;
387 free(istate->cache);
388 istate->cache = si->saved_cache;
389 istate->cache_nr = si->saved_cache_nr;
392 void discard_split_index(struct index_state *istate)
394 struct split_index *si = istate->split_index;
395 if (!si)
396 return;
397 istate->split_index = NULL;
398 si->refcount--;
399 if (si->refcount)
400 return;
401 if (si->base) {
402 discard_index(si->base);
403 free(si->base);
405 free(si);
408 void save_or_free_index_entry(struct index_state *istate, struct cache_entry *ce)
410 if (ce->index &&
411 istate->split_index &&
412 istate->split_index->base &&
413 ce->index <= istate->split_index->base->cache_nr &&
414 ce == istate->split_index->base->cache[ce->index - 1])
415 ce->ce_flags |= CE_REMOVE;
416 else
417 discard_cache_entry(ce);
420 void replace_index_entry_in_base(struct index_state *istate,
421 struct cache_entry *old_entry,
422 struct cache_entry *new_entry)
424 if (old_entry->index &&
425 istate->split_index &&
426 istate->split_index->base &&
427 old_entry->index <= istate->split_index->base->cache_nr) {
428 new_entry->index = old_entry->index;
429 if (old_entry != istate->split_index->base->cache[new_entry->index - 1])
430 discard_cache_entry(istate->split_index->base->cache[new_entry->index - 1]);
431 istate->split_index->base->cache[new_entry->index - 1] = new_entry;
435 void add_split_index(struct index_state *istate)
437 if (!istate->split_index) {
438 init_split_index(istate);
439 istate->cache_changed |= SPLIT_INDEX_ORDERED;
443 void remove_split_index(struct index_state *istate)
445 if (istate->split_index) {
446 if (istate->split_index->base) {
448 * When removing the split index, we need to move
449 * ownership of the mem_pool associated with the
450 * base index to the main index. There may be cache entries
451 * allocated from the base's memory pool that are shared with
452 * the_index.cache[].
454 mem_pool_combine(istate->ce_mem_pool,
455 istate->split_index->base->ce_mem_pool);
458 * The split index no longer owns the mem_pool backing
459 * its cache array. As we are discarding this index,
460 * mark the index as having no cache entries, so it
461 * will not attempt to clean up the cache entries or
462 * validate them.
464 istate->split_index->base->cache_nr = 0;
468 * We can discard the split index because its
469 * memory pool has been incorporated into the
470 * memory pool associated with the the_index.
472 discard_split_index(istate);
474 istate->cache_changed |= SOMETHING_CHANGED;