Sync with Git 2.45-rc0
[git.git] / split-index.c
blob8c38687c04b81ceaa57746a96daf670874e31643
1 #include "git-compat-util.h"
2 #include "gettext.h"
3 #include "hash.h"
4 #include "mem-pool.h"
5 #include "read-cache-ll.h"
6 #include "split-index.h"
7 #include "strbuf.h"
8 #include "ewah/ewok.h"
10 struct split_index *init_split_index(struct index_state *istate)
12 if (!istate->split_index) {
13 if (istate->sparse_index)
14 die(_("cannot use split index with a sparse index"));
16 CALLOC_ARRAY(istate->split_index, 1);
17 istate->split_index->refcount = 1;
19 return istate->split_index;
22 int read_link_extension(struct index_state *istate,
23 const void *data_, unsigned long sz)
25 const unsigned char *data = data_;
26 struct split_index *si;
27 int ret;
29 if (sz < the_hash_algo->rawsz)
30 return error("corrupt link extension (too short)");
31 si = init_split_index(istate);
32 oidread(&si->base_oid, data);
33 data += the_hash_algo->rawsz;
34 sz -= the_hash_algo->rawsz;
35 if (!sz)
36 return 0;
37 si->delete_bitmap = ewah_new();
38 ret = ewah_read_mmap(si->delete_bitmap, data, sz);
39 if (ret < 0)
40 return error("corrupt delete bitmap in link extension");
41 data += ret;
42 sz -= ret;
43 si->replace_bitmap = ewah_new();
44 ret = ewah_read_mmap(si->replace_bitmap, data, sz);
45 if (ret < 0)
46 return error("corrupt replace bitmap in link extension");
47 if (ret != sz)
48 return error("garbage at the end of link extension");
49 return 0;
52 int write_link_extension(struct strbuf *sb,
53 struct index_state *istate)
55 struct split_index *si = istate->split_index;
56 strbuf_add(sb, si->base_oid.hash, the_hash_algo->rawsz);
57 if (!si->delete_bitmap && !si->replace_bitmap)
58 return 0;
59 ewah_serialize_strbuf(si->delete_bitmap, sb);
60 ewah_serialize_strbuf(si->replace_bitmap, sb);
61 return 0;
64 static void mark_base_index_entries(struct index_state *base)
66 int i;
68 * To keep track of the shared entries between
69 * istate->base->cache[] and istate->cache[], base entry
70 * position is stored in each base entry. All positions start
71 * from 1 instead of 0, which is reserved to say "this is a new
72 * entry".
74 for (i = 0; i < base->cache_nr; i++)
75 base->cache[i]->index = i + 1;
78 void move_cache_to_base_index(struct index_state *istate)
80 struct split_index *si = istate->split_index;
81 int i;
84 * If there was a previous base index, then transfer ownership of allocated
85 * entries to the parent index.
87 if (si->base &&
88 si->base->ce_mem_pool) {
90 if (!istate->ce_mem_pool) {
91 istate->ce_mem_pool = xmalloc(sizeof(struct mem_pool));
92 mem_pool_init(istate->ce_mem_pool, 0);
95 mem_pool_combine(istate->ce_mem_pool, istate->split_index->base->ce_mem_pool);
98 ALLOC_ARRAY(si->base, 1);
99 index_state_init(si->base, istate->repo);
100 si->base->version = istate->version;
101 /* zero timestamp disables racy test in ce_write_index() */
102 si->base->timestamp = istate->timestamp;
103 ALLOC_GROW(si->base->cache, istate->cache_nr, si->base->cache_alloc);
104 si->base->cache_nr = istate->cache_nr;
107 * The mem_pool needs to move with the allocated entries.
109 si->base->ce_mem_pool = istate->ce_mem_pool;
110 istate->ce_mem_pool = NULL;
112 COPY_ARRAY(si->base->cache, istate->cache, istate->cache_nr);
113 mark_base_index_entries(si->base);
114 for (i = 0; i < si->base->cache_nr; i++)
115 si->base->cache[i]->ce_flags &= ~CE_UPDATE_IN_BASE;
118 static void mark_entry_for_delete(size_t pos, void *data)
120 struct index_state *istate = data;
121 if (pos >= istate->cache_nr)
122 die("position for delete %d exceeds base index size %d",
123 (int)pos, istate->cache_nr);
124 istate->cache[pos]->ce_flags |= CE_REMOVE;
125 istate->split_index->nr_deletions++;
128 static void replace_entry(size_t pos, void *data)
130 struct index_state *istate = data;
131 struct split_index *si = istate->split_index;
132 struct cache_entry *dst, *src;
134 if (pos >= istate->cache_nr)
135 die("position for replacement %d exceeds base index size %d",
136 (int)pos, istate->cache_nr);
137 if (si->nr_replacements >= si->saved_cache_nr)
138 die("too many replacements (%d vs %d)",
139 si->nr_replacements, si->saved_cache_nr);
140 dst = istate->cache[pos];
141 if (dst->ce_flags & CE_REMOVE)
142 die("entry %d is marked as both replaced and deleted",
143 (int)pos);
144 src = si->saved_cache[si->nr_replacements];
145 if (ce_namelen(src))
146 die("corrupt link extension, entry %d should have "
147 "zero length name", (int)pos);
148 src->index = pos + 1;
149 src->ce_flags |= CE_UPDATE_IN_BASE;
150 src->ce_namelen = dst->ce_namelen;
151 copy_cache_entry(dst, src);
152 discard_cache_entry(src);
153 si->nr_replacements++;
156 void merge_base_index(struct index_state *istate)
158 struct split_index *si = istate->split_index;
159 unsigned int i;
161 mark_base_index_entries(si->base);
163 si->saved_cache = istate->cache;
164 si->saved_cache_nr = istate->cache_nr;
165 istate->cache_nr = si->base->cache_nr;
166 istate->cache = NULL;
167 istate->cache_alloc = 0;
168 ALLOC_GROW(istate->cache, istate->cache_nr, istate->cache_alloc);
169 COPY_ARRAY(istate->cache, si->base->cache, istate->cache_nr);
171 si->nr_deletions = 0;
172 si->nr_replacements = 0;
173 ewah_each_bit(si->replace_bitmap, replace_entry, istate);
174 ewah_each_bit(si->delete_bitmap, mark_entry_for_delete, istate);
175 if (si->nr_deletions)
176 remove_marked_cache_entries(istate, 0);
178 for (i = si->nr_replacements; i < si->saved_cache_nr; i++) {
179 if (!ce_namelen(si->saved_cache[i]))
180 die("corrupt link extension, entry %d should "
181 "have non-zero length name", i);
182 add_index_entry(istate, si->saved_cache[i],
183 ADD_CACHE_OK_TO_ADD |
184 ADD_CACHE_KEEP_CACHE_TREE |
186 * we may have to replay what
187 * merge-recursive.c:update_stages()
188 * does, which has this flag on
190 ADD_CACHE_SKIP_DFCHECK);
191 si->saved_cache[i] = NULL;
194 ewah_free(si->delete_bitmap);
195 ewah_free(si->replace_bitmap);
196 FREE_AND_NULL(si->saved_cache);
197 si->delete_bitmap = NULL;
198 si->replace_bitmap = NULL;
199 si->saved_cache_nr = 0;
203 * Compare most of the fields in two cache entries, i.e. all except the
204 * hashmap_entry and the name.
206 static int compare_ce_content(struct cache_entry *a, struct cache_entry *b)
208 const unsigned int ondisk_flags = CE_STAGEMASK | CE_VALID |
209 CE_EXTENDED_FLAGS;
210 unsigned int ce_flags = a->ce_flags;
211 unsigned int base_flags = b->ce_flags;
212 int ret;
214 /* only on-disk flags matter */
215 a->ce_flags &= ondisk_flags;
216 b->ce_flags &= ondisk_flags;
217 ret = memcmp(&a->ce_stat_data, &b->ce_stat_data,
218 offsetof(struct cache_entry, name) -
219 offsetof(struct cache_entry, oid)) ||
220 !oideq(&a->oid, &b->oid);
221 a->ce_flags = ce_flags;
222 b->ce_flags = base_flags;
224 return ret;
227 void prepare_to_write_split_index(struct index_state *istate)
229 struct split_index *si = init_split_index(istate);
230 struct cache_entry **entries = NULL, *ce;
231 int i, nr_entries = 0, nr_alloc = 0;
233 si->delete_bitmap = ewah_new();
234 si->replace_bitmap = ewah_new();
236 if (si->base) {
237 /* Go through istate->cache[] and mark CE_MATCHED to
238 * entry with positive index. We'll go through
239 * base->cache[] later to delete all entries in base
240 * that are not marked with either CE_MATCHED or
241 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
242 * duplicate, deduplicate it.
244 for (i = 0; i < istate->cache_nr; i++) {
245 struct cache_entry *base;
246 ce = istate->cache[i];
247 if (!ce->index) {
249 * During simple update index operations this
250 * is a cache entry that is not present in
251 * the shared index. It will be added to the
252 * split index.
254 * However, it might also represent a file
255 * that already has a cache entry in the
256 * shared index, but a new index has just
257 * been constructed by unpack_trees(), and
258 * this entry now refers to different content
259 * than what was recorded in the original
260 * index, e.g. during 'read-tree -m HEAD^' or
261 * 'checkout HEAD^'. In this case the
262 * original entry in the shared index will be
263 * marked as deleted, and this entry will be
264 * added to the split index.
266 continue;
268 if (ce->index > si->base->cache_nr) {
269 BUG("ce refers to a shared ce at %d, which is beyond the shared index size %d",
270 ce->index, si->base->cache_nr);
272 ce->ce_flags |= CE_MATCHED; /* or "shared" */
273 base = si->base->cache[ce->index - 1];
274 if (ce == base) {
275 /* The entry is present in the shared index. */
276 if (ce->ce_flags & CE_UPDATE_IN_BASE) {
278 * Already marked for inclusion in
279 * the split index, either because
280 * the corresponding file was
281 * modified and the cached stat data
282 * was refreshed, or because there
283 * is already a replacement entry in
284 * the split index.
285 * Nothing more to do here.
287 } else if (!ce_uptodate(ce) &&
288 is_racy_timestamp(istate, ce)) {
290 * A racily clean cache entry stored
291 * only in the shared index: it must
292 * be added to the split index, so
293 * the subsequent do_write_index()
294 * can smudge its stat data.
296 ce->ce_flags |= CE_UPDATE_IN_BASE;
297 } else {
299 * The entry is only present in the
300 * shared index and it was not
301 * refreshed.
302 * Just leave it there.
305 continue;
307 if (ce->ce_namelen != base->ce_namelen ||
308 strcmp(ce->name, base->name)) {
309 ce->index = 0;
310 continue;
313 * This is the copy of a cache entry that is present
314 * in the shared index, created by unpack_trees()
315 * while it constructed a new index.
317 if (ce->ce_flags & CE_UPDATE_IN_BASE) {
319 * Already marked for inclusion in the split
320 * index, either because the corresponding
321 * file was modified and the cached stat data
322 * was refreshed, or because the original
323 * entry already had a replacement entry in
324 * the split index.
325 * Nothing to do.
327 } else if (!ce_uptodate(ce) &&
328 is_racy_timestamp(istate, ce)) {
330 * A copy of a racily clean cache entry from
331 * the shared index. It must be added to
332 * the split index, so the subsequent
333 * do_write_index() can smudge its stat data.
335 ce->ce_flags |= CE_UPDATE_IN_BASE;
336 } else {
338 * Thoroughly compare the cached data to see
339 * whether it should be marked for inclusion
340 * in the split index.
342 * This comparison might be unnecessary, as
343 * code paths modifying the cached data do
344 * set CE_UPDATE_IN_BASE as well.
346 if (compare_ce_content(ce, base))
347 ce->ce_flags |= CE_UPDATE_IN_BASE;
349 discard_cache_entry(base);
350 si->base->cache[ce->index - 1] = ce;
352 for (i = 0; i < si->base->cache_nr; i++) {
353 ce = si->base->cache[i];
354 if ((ce->ce_flags & CE_REMOVE) ||
355 !(ce->ce_flags & CE_MATCHED))
356 ewah_set(si->delete_bitmap, i);
357 else if (ce->ce_flags & CE_UPDATE_IN_BASE) {
358 ewah_set(si->replace_bitmap, i);
359 ce->ce_flags |= CE_STRIP_NAME;
360 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
361 entries[nr_entries++] = ce;
363 if (is_null_oid(&ce->oid))
364 istate->drop_cache_tree = 1;
368 for (i = 0; i < istate->cache_nr; i++) {
369 ce = istate->cache[i];
370 if ((!si->base || !ce->index) && !(ce->ce_flags & CE_REMOVE)) {
371 assert(!(ce->ce_flags & CE_STRIP_NAME));
372 ALLOC_GROW(entries, nr_entries+1, nr_alloc);
373 entries[nr_entries++] = ce;
375 ce->ce_flags &= ~CE_MATCHED;
379 * take cache[] out temporarily, put entries[] in its place
380 * for writing
382 si->saved_cache = istate->cache;
383 si->saved_cache_nr = istate->cache_nr;
384 istate->cache = entries;
385 istate->cache_nr = nr_entries;
388 void finish_writing_split_index(struct index_state *istate)
390 struct split_index *si = init_split_index(istate);
392 ewah_free(si->delete_bitmap);
393 ewah_free(si->replace_bitmap);
394 si->delete_bitmap = NULL;
395 si->replace_bitmap = NULL;
396 free(istate->cache);
397 istate->cache = si->saved_cache;
398 istate->cache_nr = si->saved_cache_nr;
401 void discard_split_index(struct index_state *istate)
403 struct split_index *si = istate->split_index;
404 if (!si)
405 return;
406 istate->split_index = NULL;
407 si->refcount--;
408 if (si->refcount)
409 return;
410 if (si->base) {
411 discard_index(si->base);
412 free(si->base);
414 free(si);
417 void save_or_free_index_entry(struct index_state *istate, struct cache_entry *ce)
419 if (ce->index &&
420 istate->split_index &&
421 istate->split_index->base &&
422 ce->index <= istate->split_index->base->cache_nr &&
423 ce == istate->split_index->base->cache[ce->index - 1])
424 ce->ce_flags |= CE_REMOVE;
425 else
426 discard_cache_entry(ce);
429 void replace_index_entry_in_base(struct index_state *istate,
430 struct cache_entry *old_entry,
431 struct cache_entry *new_entry)
433 if (old_entry->index &&
434 istate->split_index &&
435 istate->split_index->base &&
436 old_entry->index <= istate->split_index->base->cache_nr) {
437 new_entry->index = old_entry->index;
438 if (old_entry != istate->split_index->base->cache[new_entry->index - 1])
439 discard_cache_entry(istate->split_index->base->cache[new_entry->index - 1]);
440 istate->split_index->base->cache[new_entry->index - 1] = new_entry;
444 void add_split_index(struct index_state *istate)
446 if (!istate->split_index) {
447 init_split_index(istate);
448 istate->cache_changed |= SPLIT_INDEX_ORDERED;
452 void remove_split_index(struct index_state *istate)
454 if (istate->split_index) {
455 if (istate->split_index->base) {
457 * When removing the split index, we need to move
458 * ownership of the mem_pool associated with the
459 * base index to the main index. There may be cache entries
460 * allocated from the base's memory pool that are shared with
461 * the_index.cache[].
463 mem_pool_combine(istate->ce_mem_pool,
464 istate->split_index->base->ce_mem_pool);
467 * The split index no longer owns the mem_pool backing
468 * its cache array. As we are discarding this index,
469 * mark the index as having no cache entries, so it
470 * will not attempt to clean up the cache entries or
471 * validate them.
473 istate->split_index->base->cache_nr = 0;
477 * We can discard the split index because its
478 * memory pool has been incorporated into the
479 * memory pool associated with the the_index.
481 discard_split_index(istate);
483 istate->cache_changed |= SOMETHING_CHANGED;