5 #include "split-index.h"
8 struct split_index
*init_split_index(struct index_state
*istate
)
10 if (!istate
->split_index
) {
11 if (istate
->sparse_index
)
12 die(_("cannot use split index with a sparse index"));
14 CALLOC_ARRAY(istate
->split_index
, 1);
15 istate
->split_index
->refcount
= 1;
17 return istate
->split_index
;
20 int read_link_extension(struct index_state
*istate
,
21 const void *data_
, unsigned long sz
)
23 const unsigned char *data
= data_
;
24 struct split_index
*si
;
27 if (sz
< the_hash_algo
->rawsz
)
28 return error("corrupt link extension (too short)");
29 si
= init_split_index(istate
);
30 oidread(&si
->base_oid
, data
);
31 data
+= the_hash_algo
->rawsz
;
32 sz
-= the_hash_algo
->rawsz
;
35 si
->delete_bitmap
= ewah_new();
36 ret
= ewah_read_mmap(si
->delete_bitmap
, data
, sz
);
38 return error("corrupt delete bitmap in link extension");
41 si
->replace_bitmap
= ewah_new();
42 ret
= ewah_read_mmap(si
->replace_bitmap
, data
, sz
);
44 return error("corrupt replace bitmap in link extension");
46 return error("garbage at the end of link extension");
50 int write_link_extension(struct strbuf
*sb
,
51 struct index_state
*istate
)
53 struct split_index
*si
= istate
->split_index
;
54 strbuf_add(sb
, si
->base_oid
.hash
, the_hash_algo
->rawsz
);
55 if (!si
->delete_bitmap
&& !si
->replace_bitmap
)
57 ewah_serialize_strbuf(si
->delete_bitmap
, sb
);
58 ewah_serialize_strbuf(si
->replace_bitmap
, sb
);
62 static void mark_base_index_entries(struct index_state
*base
)
66 * To keep track of the shared entries between
67 * istate->base->cache[] and istate->cache[], base entry
68 * position is stored in each base entry. All positions start
69 * from 1 instead of 0, which is reserved to say "this is a new
72 for (i
= 0; i
< base
->cache_nr
; i
++)
73 base
->cache
[i
]->index
= i
+ 1;
76 void move_cache_to_base_index(struct index_state
*istate
)
78 struct split_index
*si
= istate
->split_index
;
82 * If there was a previous base index, then transfer ownership of allocated
83 * entries to the parent index.
86 si
->base
->ce_mem_pool
) {
88 if (!istate
->ce_mem_pool
) {
89 istate
->ce_mem_pool
= xmalloc(sizeof(struct mem_pool
));
90 mem_pool_init(istate
->ce_mem_pool
, 0);
93 mem_pool_combine(istate
->ce_mem_pool
, istate
->split_index
->base
->ce_mem_pool
);
96 ALLOC_ARRAY(si
->base
, 1);
97 index_state_init(si
->base
, istate
->repo
);
98 si
->base
->version
= istate
->version
;
99 /* zero timestamp disables racy test in ce_write_index() */
100 si
->base
->timestamp
= istate
->timestamp
;
101 ALLOC_GROW(si
->base
->cache
, istate
->cache_nr
, si
->base
->cache_alloc
);
102 si
->base
->cache_nr
= istate
->cache_nr
;
105 * The mem_pool needs to move with the allocated entries.
107 si
->base
->ce_mem_pool
= istate
->ce_mem_pool
;
108 istate
->ce_mem_pool
= NULL
;
110 COPY_ARRAY(si
->base
->cache
, istate
->cache
, istate
->cache_nr
);
111 mark_base_index_entries(si
->base
);
112 for (i
= 0; i
< si
->base
->cache_nr
; i
++)
113 si
->base
->cache
[i
]->ce_flags
&= ~CE_UPDATE_IN_BASE
;
116 static void mark_entry_for_delete(size_t pos
, void *data
)
118 struct index_state
*istate
= data
;
119 if (pos
>= istate
->cache_nr
)
120 die("position for delete %d exceeds base index size %d",
121 (int)pos
, istate
->cache_nr
);
122 istate
->cache
[pos
]->ce_flags
|= CE_REMOVE
;
123 istate
->split_index
->nr_deletions
++;
126 static void replace_entry(size_t pos
, void *data
)
128 struct index_state
*istate
= data
;
129 struct split_index
*si
= istate
->split_index
;
130 struct cache_entry
*dst
, *src
;
132 if (pos
>= istate
->cache_nr
)
133 die("position for replacement %d exceeds base index size %d",
134 (int)pos
, istate
->cache_nr
);
135 if (si
->nr_replacements
>= si
->saved_cache_nr
)
136 die("too many replacements (%d vs %d)",
137 si
->nr_replacements
, si
->saved_cache_nr
);
138 dst
= istate
->cache
[pos
];
139 if (dst
->ce_flags
& CE_REMOVE
)
140 die("entry %d is marked as both replaced and deleted",
142 src
= si
->saved_cache
[si
->nr_replacements
];
144 die("corrupt link extension, entry %d should have "
145 "zero length name", (int)pos
);
146 src
->index
= pos
+ 1;
147 src
->ce_flags
|= CE_UPDATE_IN_BASE
;
148 src
->ce_namelen
= dst
->ce_namelen
;
149 copy_cache_entry(dst
, src
);
150 discard_cache_entry(src
);
151 si
->nr_replacements
++;
154 void merge_base_index(struct index_state
*istate
)
156 struct split_index
*si
= istate
->split_index
;
159 mark_base_index_entries(si
->base
);
161 si
->saved_cache
= istate
->cache
;
162 si
->saved_cache_nr
= istate
->cache_nr
;
163 istate
->cache_nr
= si
->base
->cache_nr
;
164 istate
->cache
= NULL
;
165 istate
->cache_alloc
= 0;
166 ALLOC_GROW(istate
->cache
, istate
->cache_nr
, istate
->cache_alloc
);
167 COPY_ARRAY(istate
->cache
, si
->base
->cache
, istate
->cache_nr
);
169 si
->nr_deletions
= 0;
170 si
->nr_replacements
= 0;
171 ewah_each_bit(si
->replace_bitmap
, replace_entry
, istate
);
172 ewah_each_bit(si
->delete_bitmap
, mark_entry_for_delete
, istate
);
173 if (si
->nr_deletions
)
174 remove_marked_cache_entries(istate
, 0);
176 for (i
= si
->nr_replacements
; i
< si
->saved_cache_nr
; i
++) {
177 if (!ce_namelen(si
->saved_cache
[i
]))
178 die("corrupt link extension, entry %d should "
179 "have non-zero length name", i
);
180 add_index_entry(istate
, si
->saved_cache
[i
],
181 ADD_CACHE_OK_TO_ADD
|
182 ADD_CACHE_KEEP_CACHE_TREE
|
184 * we may have to replay what
185 * merge-recursive.c:update_stages()
186 * does, which has this flag on
188 ADD_CACHE_SKIP_DFCHECK
);
189 si
->saved_cache
[i
] = NULL
;
192 ewah_free(si
->delete_bitmap
);
193 ewah_free(si
->replace_bitmap
);
194 FREE_AND_NULL(si
->saved_cache
);
195 si
->delete_bitmap
= NULL
;
196 si
->replace_bitmap
= NULL
;
197 si
->saved_cache_nr
= 0;
201 * Compare most of the fields in two cache entries, i.e. all except the
202 * hashmap_entry and the name.
204 static int compare_ce_content(struct cache_entry
*a
, struct cache_entry
*b
)
206 const unsigned int ondisk_flags
= CE_STAGEMASK
| CE_VALID
|
208 unsigned int ce_flags
= a
->ce_flags
;
209 unsigned int base_flags
= b
->ce_flags
;
212 /* only on-disk flags matter */
213 a
->ce_flags
&= ondisk_flags
;
214 b
->ce_flags
&= ondisk_flags
;
215 ret
= memcmp(&a
->ce_stat_data
, &b
->ce_stat_data
,
216 offsetof(struct cache_entry
, name
) -
217 offsetof(struct cache_entry
, oid
)) ||
218 !oideq(&a
->oid
, &b
->oid
);
219 a
->ce_flags
= ce_flags
;
220 b
->ce_flags
= base_flags
;
225 void prepare_to_write_split_index(struct index_state
*istate
)
227 struct split_index
*si
= init_split_index(istate
);
228 struct cache_entry
**entries
= NULL
, *ce
;
229 int i
, nr_entries
= 0, nr_alloc
= 0;
231 si
->delete_bitmap
= ewah_new();
232 si
->replace_bitmap
= ewah_new();
235 /* Go through istate->cache[] and mark CE_MATCHED to
236 * entry with positive index. We'll go through
237 * base->cache[] later to delete all entries in base
238 * that are not marked with either CE_MATCHED or
239 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
240 * duplicate, deduplicate it.
242 for (i
= 0; i
< istate
->cache_nr
; i
++) {
243 struct cache_entry
*base
;
244 ce
= istate
->cache
[i
];
247 * During simple update index operations this
248 * is a cache entry that is not present in
249 * the shared index. It will be added to the
252 * However, it might also represent a file
253 * that already has a cache entry in the
254 * shared index, but a new index has just
255 * been constructed by unpack_trees(), and
256 * this entry now refers to different content
257 * than what was recorded in the original
258 * index, e.g. during 'read-tree -m HEAD^' or
259 * 'checkout HEAD^'. In this case the
260 * original entry in the shared index will be
261 * marked as deleted, and this entry will be
262 * added to the split index.
266 if (ce
->index
> si
->base
->cache_nr
) {
267 BUG("ce refers to a shared ce at %d, which is beyond the shared index size %d",
268 ce
->index
, si
->base
->cache_nr
);
270 ce
->ce_flags
|= CE_MATCHED
; /* or "shared" */
271 base
= si
->base
->cache
[ce
->index
- 1];
273 /* The entry is present in the shared index. */
274 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
276 * Already marked for inclusion in
277 * the split index, either because
278 * the corresponding file was
279 * modified and the cached stat data
280 * was refreshed, or because there
281 * is already a replacement entry in
283 * Nothing more to do here.
285 } else if (!ce_uptodate(ce
) &&
286 is_racy_timestamp(istate
, ce
)) {
288 * A racily clean cache entry stored
289 * only in the shared index: it must
290 * be added to the split index, so
291 * the subsequent do_write_index()
292 * can smudge its stat data.
294 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
297 * The entry is only present in the
298 * shared index and it was not
300 * Just leave it there.
305 if (ce
->ce_namelen
!= base
->ce_namelen
||
306 strcmp(ce
->name
, base
->name
)) {
311 * This is the copy of a cache entry that is present
312 * in the shared index, created by unpack_trees()
313 * while it constructed a new index.
315 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
317 * Already marked for inclusion in the split
318 * index, either because the corresponding
319 * file was modified and the cached stat data
320 * was refreshed, or because the original
321 * entry already had a replacement entry in
325 } else if (!ce_uptodate(ce
) &&
326 is_racy_timestamp(istate
, ce
)) {
328 * A copy of a racily clean cache entry from
329 * the shared index. It must be added to
330 * the split index, so the subsequent
331 * do_write_index() can smudge its stat data.
333 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
336 * Thoroughly compare the cached data to see
337 * whether it should be marked for inclusion
338 * in the split index.
340 * This comparison might be unnecessary, as
341 * code paths modifying the cached data do
342 * set CE_UPDATE_IN_BASE as well.
344 if (compare_ce_content(ce
, base
))
345 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
347 discard_cache_entry(base
);
348 si
->base
->cache
[ce
->index
- 1] = ce
;
350 for (i
= 0; i
< si
->base
->cache_nr
; i
++) {
351 ce
= si
->base
->cache
[i
];
352 if ((ce
->ce_flags
& CE_REMOVE
) ||
353 !(ce
->ce_flags
& CE_MATCHED
))
354 ewah_set(si
->delete_bitmap
, i
);
355 else if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
356 ewah_set(si
->replace_bitmap
, i
);
357 ce
->ce_flags
|= CE_STRIP_NAME
;
358 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
359 entries
[nr_entries
++] = ce
;
361 if (is_null_oid(&ce
->oid
))
362 istate
->drop_cache_tree
= 1;
366 for (i
= 0; i
< istate
->cache_nr
; i
++) {
367 ce
= istate
->cache
[i
];
368 if ((!si
->base
|| !ce
->index
) && !(ce
->ce_flags
& CE_REMOVE
)) {
369 assert(!(ce
->ce_flags
& CE_STRIP_NAME
));
370 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
371 entries
[nr_entries
++] = ce
;
373 ce
->ce_flags
&= ~CE_MATCHED
;
377 * take cache[] out temporarily, put entries[] in its place
380 si
->saved_cache
= istate
->cache
;
381 si
->saved_cache_nr
= istate
->cache_nr
;
382 istate
->cache
= entries
;
383 istate
->cache_nr
= nr_entries
;
386 void finish_writing_split_index(struct index_state
*istate
)
388 struct split_index
*si
= init_split_index(istate
);
390 ewah_free(si
->delete_bitmap
);
391 ewah_free(si
->replace_bitmap
);
392 si
->delete_bitmap
= NULL
;
393 si
->replace_bitmap
= NULL
;
395 istate
->cache
= si
->saved_cache
;
396 istate
->cache_nr
= si
->saved_cache_nr
;
399 void discard_split_index(struct index_state
*istate
)
401 struct split_index
*si
= istate
->split_index
;
404 istate
->split_index
= NULL
;
409 discard_index(si
->base
);
415 void save_or_free_index_entry(struct index_state
*istate
, struct cache_entry
*ce
)
418 istate
->split_index
&&
419 istate
->split_index
->base
&&
420 ce
->index
<= istate
->split_index
->base
->cache_nr
&&
421 ce
== istate
->split_index
->base
->cache
[ce
->index
- 1])
422 ce
->ce_flags
|= CE_REMOVE
;
424 discard_cache_entry(ce
);
427 void replace_index_entry_in_base(struct index_state
*istate
,
428 struct cache_entry
*old_entry
,
429 struct cache_entry
*new_entry
)
431 if (old_entry
->index
&&
432 istate
->split_index
&&
433 istate
->split_index
->base
&&
434 old_entry
->index
<= istate
->split_index
->base
->cache_nr
) {
435 new_entry
->index
= old_entry
->index
;
436 if (old_entry
!= istate
->split_index
->base
->cache
[new_entry
->index
- 1])
437 discard_cache_entry(istate
->split_index
->base
->cache
[new_entry
->index
- 1]);
438 istate
->split_index
->base
->cache
[new_entry
->index
- 1] = new_entry
;
442 void add_split_index(struct index_state
*istate
)
444 if (!istate
->split_index
) {
445 init_split_index(istate
);
446 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
450 void remove_split_index(struct index_state
*istate
)
452 if (istate
->split_index
) {
453 if (istate
->split_index
->base
) {
455 * When removing the split index, we need to move
456 * ownership of the mem_pool associated with the
457 * base index to the main index. There may be cache entries
458 * allocated from the base's memory pool that are shared with
461 mem_pool_combine(istate
->ce_mem_pool
,
462 istate
->split_index
->base
->ce_mem_pool
);
465 * The split index no longer owns the mem_pool backing
466 * its cache array. As we are discarding this index,
467 * mark the index as having no cache entries, so it
468 * will not attempt to clean up the cache entries or
471 istate
->split_index
->base
->cache_nr
= 0;
475 * We can discard the split index because its
476 * memory pool has been incorporated into the
477 * memory pool associated with the the_index.
479 discard_split_index(istate
);
481 istate
->cache_changed
|= SOMETHING_CHANGED
;