4 #include "split-index.h"
7 struct split_index
*init_split_index(struct index_state
*istate
)
9 if (!istate
->split_index
) {
10 if (istate
->sparse_index
)
11 die(_("cannot use split index with a sparse index"));
13 CALLOC_ARRAY(istate
->split_index
, 1);
14 istate
->split_index
->refcount
= 1;
16 return istate
->split_index
;
19 int read_link_extension(struct index_state
*istate
,
20 const void *data_
, unsigned long sz
)
22 const unsigned char *data
= data_
;
23 struct split_index
*si
;
26 if (sz
< the_hash_algo
->rawsz
)
27 return error("corrupt link extension (too short)");
28 si
= init_split_index(istate
);
29 oidread(&si
->base_oid
, data
);
30 data
+= the_hash_algo
->rawsz
;
31 sz
-= the_hash_algo
->rawsz
;
34 si
->delete_bitmap
= ewah_new();
35 ret
= ewah_read_mmap(si
->delete_bitmap
, data
, sz
);
37 return error("corrupt delete bitmap in link extension");
40 si
->replace_bitmap
= ewah_new();
41 ret
= ewah_read_mmap(si
->replace_bitmap
, data
, sz
);
43 return error("corrupt replace bitmap in link extension");
45 return error("garbage at the end of link extension");
49 int write_link_extension(struct strbuf
*sb
,
50 struct index_state
*istate
)
52 struct split_index
*si
= istate
->split_index
;
53 strbuf_add(sb
, si
->base_oid
.hash
, the_hash_algo
->rawsz
);
54 if (!si
->delete_bitmap
&& !si
->replace_bitmap
)
56 ewah_serialize_strbuf(si
->delete_bitmap
, sb
);
57 ewah_serialize_strbuf(si
->replace_bitmap
, sb
);
61 static void mark_base_index_entries(struct index_state
*base
)
65 * To keep track of the shared entries between
66 * istate->base->cache[] and istate->cache[], base entry
67 * position is stored in each base entry. All positions start
68 * from 1 instead of 0, which is reserved to say "this is a new
71 for (i
= 0; i
< base
->cache_nr
; i
++)
72 base
->cache
[i
]->index
= i
+ 1;
75 void move_cache_to_base_index(struct index_state
*istate
)
77 struct split_index
*si
= istate
->split_index
;
81 * If there was a previous base index, then transfer ownership of allocated
82 * entries to the parent index.
85 si
->base
->ce_mem_pool
) {
87 if (!istate
->ce_mem_pool
) {
88 istate
->ce_mem_pool
= xmalloc(sizeof(struct mem_pool
));
89 mem_pool_init(istate
->ce_mem_pool
, 0);
92 mem_pool_combine(istate
->ce_mem_pool
, istate
->split_index
->base
->ce_mem_pool
);
95 ALLOC_ARRAY(si
->base
, 1);
96 index_state_init(si
->base
, istate
->repo
);
97 si
->base
->version
= istate
->version
;
98 /* zero timestamp disables racy test in ce_write_index() */
99 si
->base
->timestamp
= istate
->timestamp
;
100 ALLOC_GROW(si
->base
->cache
, istate
->cache_nr
, si
->base
->cache_alloc
);
101 si
->base
->cache_nr
= istate
->cache_nr
;
104 * The mem_pool needs to move with the allocated entries.
106 si
->base
->ce_mem_pool
= istate
->ce_mem_pool
;
107 istate
->ce_mem_pool
= NULL
;
109 COPY_ARRAY(si
->base
->cache
, istate
->cache
, istate
->cache_nr
);
110 mark_base_index_entries(si
->base
);
111 for (i
= 0; i
< si
->base
->cache_nr
; i
++)
112 si
->base
->cache
[i
]->ce_flags
&= ~CE_UPDATE_IN_BASE
;
115 static void mark_entry_for_delete(size_t pos
, void *data
)
117 struct index_state
*istate
= data
;
118 if (pos
>= istate
->cache_nr
)
119 die("position for delete %d exceeds base index size %d",
120 (int)pos
, istate
->cache_nr
);
121 istate
->cache
[pos
]->ce_flags
|= CE_REMOVE
;
122 istate
->split_index
->nr_deletions
++;
125 static void replace_entry(size_t pos
, void *data
)
127 struct index_state
*istate
= data
;
128 struct split_index
*si
= istate
->split_index
;
129 struct cache_entry
*dst
, *src
;
131 if (pos
>= istate
->cache_nr
)
132 die("position for replacement %d exceeds base index size %d",
133 (int)pos
, istate
->cache_nr
);
134 if (si
->nr_replacements
>= si
->saved_cache_nr
)
135 die("too many replacements (%d vs %d)",
136 si
->nr_replacements
, si
->saved_cache_nr
);
137 dst
= istate
->cache
[pos
];
138 if (dst
->ce_flags
& CE_REMOVE
)
139 die("entry %d is marked as both replaced and deleted",
141 src
= si
->saved_cache
[si
->nr_replacements
];
143 die("corrupt link extension, entry %d should have "
144 "zero length name", (int)pos
);
145 src
->index
= pos
+ 1;
146 src
->ce_flags
|= CE_UPDATE_IN_BASE
;
147 src
->ce_namelen
= dst
->ce_namelen
;
148 copy_cache_entry(dst
, src
);
149 discard_cache_entry(src
);
150 si
->nr_replacements
++;
153 void merge_base_index(struct index_state
*istate
)
155 struct split_index
*si
= istate
->split_index
;
158 mark_base_index_entries(si
->base
);
160 si
->saved_cache
= istate
->cache
;
161 si
->saved_cache_nr
= istate
->cache_nr
;
162 istate
->cache_nr
= si
->base
->cache_nr
;
163 istate
->cache
= NULL
;
164 istate
->cache_alloc
= 0;
165 ALLOC_GROW(istate
->cache
, istate
->cache_nr
, istate
->cache_alloc
);
166 COPY_ARRAY(istate
->cache
, si
->base
->cache
, istate
->cache_nr
);
168 si
->nr_deletions
= 0;
169 si
->nr_replacements
= 0;
170 ewah_each_bit(si
->replace_bitmap
, replace_entry
, istate
);
171 ewah_each_bit(si
->delete_bitmap
, mark_entry_for_delete
, istate
);
172 if (si
->nr_deletions
)
173 remove_marked_cache_entries(istate
, 0);
175 for (i
= si
->nr_replacements
; i
< si
->saved_cache_nr
; i
++) {
176 if (!ce_namelen(si
->saved_cache
[i
]))
177 die("corrupt link extension, entry %d should "
178 "have non-zero length name", i
);
179 add_index_entry(istate
, si
->saved_cache
[i
],
180 ADD_CACHE_OK_TO_ADD
|
181 ADD_CACHE_KEEP_CACHE_TREE
|
183 * we may have to replay what
184 * merge-recursive.c:update_stages()
185 * does, which has this flag on
187 ADD_CACHE_SKIP_DFCHECK
);
188 si
->saved_cache
[i
] = NULL
;
191 ewah_free(si
->delete_bitmap
);
192 ewah_free(si
->replace_bitmap
);
193 FREE_AND_NULL(si
->saved_cache
);
194 si
->delete_bitmap
= NULL
;
195 si
->replace_bitmap
= NULL
;
196 si
->saved_cache_nr
= 0;
200 * Compare most of the fields in two cache entries, i.e. all except the
201 * hashmap_entry and the name.
203 static int compare_ce_content(struct cache_entry
*a
, struct cache_entry
*b
)
205 const unsigned int ondisk_flags
= CE_STAGEMASK
| CE_VALID
|
207 unsigned int ce_flags
= a
->ce_flags
;
208 unsigned int base_flags
= b
->ce_flags
;
211 /* only on-disk flags matter */
212 a
->ce_flags
&= ondisk_flags
;
213 b
->ce_flags
&= ondisk_flags
;
214 ret
= memcmp(&a
->ce_stat_data
, &b
->ce_stat_data
,
215 offsetof(struct cache_entry
, name
) -
216 offsetof(struct cache_entry
, oid
)) ||
217 !oideq(&a
->oid
, &b
->oid
);
218 a
->ce_flags
= ce_flags
;
219 b
->ce_flags
= base_flags
;
224 void prepare_to_write_split_index(struct index_state
*istate
)
226 struct split_index
*si
= init_split_index(istate
);
227 struct cache_entry
**entries
= NULL
, *ce
;
228 int i
, nr_entries
= 0, nr_alloc
= 0;
230 si
->delete_bitmap
= ewah_new();
231 si
->replace_bitmap
= ewah_new();
234 /* Go through istate->cache[] and mark CE_MATCHED to
235 * entry with positive index. We'll go through
236 * base->cache[] later to delete all entries in base
237 * that are not marked with either CE_MATCHED or
238 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
239 * duplicate, deduplicate it.
241 for (i
= 0; i
< istate
->cache_nr
; i
++) {
242 struct cache_entry
*base
;
243 ce
= istate
->cache
[i
];
246 * During simple update index operations this
247 * is a cache entry that is not present in
248 * the shared index. It will be added to the
251 * However, it might also represent a file
252 * that already has a cache entry in the
253 * shared index, but a new index has just
254 * been constructed by unpack_trees(), and
255 * this entry now refers to different content
256 * than what was recorded in the original
257 * index, e.g. during 'read-tree -m HEAD^' or
258 * 'checkout HEAD^'. In this case the
259 * original entry in the shared index will be
260 * marked as deleted, and this entry will be
261 * added to the split index.
265 if (ce
->index
> si
->base
->cache_nr
) {
266 BUG("ce refers to a shared ce at %d, which is beyond the shared index size %d",
267 ce
->index
, si
->base
->cache_nr
);
269 ce
->ce_flags
|= CE_MATCHED
; /* or "shared" */
270 base
= si
->base
->cache
[ce
->index
- 1];
272 /* The entry is present in the shared index. */
273 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
275 * Already marked for inclusion in
276 * the split index, either because
277 * the corresponding file was
278 * modified and the cached stat data
279 * was refreshed, or because there
280 * is already a replacement entry in
282 * Nothing more to do here.
284 } else if (!ce_uptodate(ce
) &&
285 is_racy_timestamp(istate
, ce
)) {
287 * A racily clean cache entry stored
288 * only in the shared index: it must
289 * be added to the split index, so
290 * the subsequent do_write_index()
291 * can smudge its stat data.
293 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
296 * The entry is only present in the
297 * shared index and it was not
299 * Just leave it there.
304 if (ce
->ce_namelen
!= base
->ce_namelen
||
305 strcmp(ce
->name
, base
->name
)) {
310 * This is the copy of a cache entry that is present
311 * in the shared index, created by unpack_trees()
312 * while it constructed a new index.
314 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
316 * Already marked for inclusion in the split
317 * index, either because the corresponding
318 * file was modified and the cached stat data
319 * was refreshed, or because the original
320 * entry already had a replacement entry in
324 } else if (!ce_uptodate(ce
) &&
325 is_racy_timestamp(istate
, ce
)) {
327 * A copy of a racily clean cache entry from
328 * the shared index. It must be added to
329 * the split index, so the subsequent
330 * do_write_index() can smudge its stat data.
332 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
335 * Thoroughly compare the cached data to see
336 * whether it should be marked for inclusion
337 * in the split index.
339 * This comparison might be unnecessary, as
340 * code paths modifying the cached data do
341 * set CE_UPDATE_IN_BASE as well.
343 if (compare_ce_content(ce
, base
))
344 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
346 discard_cache_entry(base
);
347 si
->base
->cache
[ce
->index
- 1] = ce
;
349 for (i
= 0; i
< si
->base
->cache_nr
; i
++) {
350 ce
= si
->base
->cache
[i
];
351 if ((ce
->ce_flags
& CE_REMOVE
) ||
352 !(ce
->ce_flags
& CE_MATCHED
))
353 ewah_set(si
->delete_bitmap
, i
);
354 else if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
355 ewah_set(si
->replace_bitmap
, i
);
356 ce
->ce_flags
|= CE_STRIP_NAME
;
357 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
358 entries
[nr_entries
++] = ce
;
360 if (is_null_oid(&ce
->oid
))
361 istate
->drop_cache_tree
= 1;
365 for (i
= 0; i
< istate
->cache_nr
; i
++) {
366 ce
= istate
->cache
[i
];
367 if ((!si
->base
|| !ce
->index
) && !(ce
->ce_flags
& CE_REMOVE
)) {
368 assert(!(ce
->ce_flags
& CE_STRIP_NAME
));
369 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
370 entries
[nr_entries
++] = ce
;
372 ce
->ce_flags
&= ~CE_MATCHED
;
376 * take cache[] out temporarily, put entries[] in its place
379 si
->saved_cache
= istate
->cache
;
380 si
->saved_cache_nr
= istate
->cache_nr
;
381 istate
->cache
= entries
;
382 istate
->cache_nr
= nr_entries
;
385 void finish_writing_split_index(struct index_state
*istate
)
387 struct split_index
*si
= init_split_index(istate
);
389 ewah_free(si
->delete_bitmap
);
390 ewah_free(si
->replace_bitmap
);
391 si
->delete_bitmap
= NULL
;
392 si
->replace_bitmap
= NULL
;
394 istate
->cache
= si
->saved_cache
;
395 istate
->cache_nr
= si
->saved_cache_nr
;
398 void discard_split_index(struct index_state
*istate
)
400 struct split_index
*si
= istate
->split_index
;
403 istate
->split_index
= NULL
;
408 discard_index(si
->base
);
414 void save_or_free_index_entry(struct index_state
*istate
, struct cache_entry
*ce
)
417 istate
->split_index
&&
418 istate
->split_index
->base
&&
419 ce
->index
<= istate
->split_index
->base
->cache_nr
&&
420 ce
== istate
->split_index
->base
->cache
[ce
->index
- 1])
421 ce
->ce_flags
|= CE_REMOVE
;
423 discard_cache_entry(ce
);
426 void replace_index_entry_in_base(struct index_state
*istate
,
427 struct cache_entry
*old_entry
,
428 struct cache_entry
*new_entry
)
430 if (old_entry
->index
&&
431 istate
->split_index
&&
432 istate
->split_index
->base
&&
433 old_entry
->index
<= istate
->split_index
->base
->cache_nr
) {
434 new_entry
->index
= old_entry
->index
;
435 if (old_entry
!= istate
->split_index
->base
->cache
[new_entry
->index
- 1])
436 discard_cache_entry(istate
->split_index
->base
->cache
[new_entry
->index
- 1]);
437 istate
->split_index
->base
->cache
[new_entry
->index
- 1] = new_entry
;
441 void add_split_index(struct index_state
*istate
)
443 if (!istate
->split_index
) {
444 init_split_index(istate
);
445 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
449 void remove_split_index(struct index_state
*istate
)
451 if (istate
->split_index
) {
452 if (istate
->split_index
->base
) {
454 * When removing the split index, we need to move
455 * ownership of the mem_pool associated with the
456 * base index to the main index. There may be cache entries
457 * allocated from the base's memory pool that are shared with
460 mem_pool_combine(istate
->ce_mem_pool
,
461 istate
->split_index
->base
->ce_mem_pool
);
464 * The split index no longer owns the mem_pool backing
465 * its cache array. As we are discarding this index,
466 * mark the index as having no cache entries, so it
467 * will not attempt to clean up the cache entries or
470 istate
->split_index
->base
->cache_nr
= 0;
474 * We can discard the split index because its
475 * memory pool has been incorporated into the
476 * memory pool associated with the the_index.
478 discard_split_index(istate
);
480 istate
->cache_changed
|= SOMETHING_CHANGED
;