2 #include "split-index.h"
5 struct split_index
*init_split_index(struct index_state
*istate
)
7 if (!istate
->split_index
) {
8 if (istate
->sparse_index
)
9 die(_("cannot use split index with a sparse index"));
11 CALLOC_ARRAY(istate
->split_index
, 1);
12 istate
->split_index
->refcount
= 1;
14 return istate
->split_index
;
17 int read_link_extension(struct index_state
*istate
,
18 const void *data_
, unsigned long sz
)
20 const unsigned char *data
= data_
;
21 struct split_index
*si
;
24 if (sz
< the_hash_algo
->rawsz
)
25 return error("corrupt link extension (too short)");
26 si
= init_split_index(istate
);
27 oidread(&si
->base_oid
, data
);
28 data
+= the_hash_algo
->rawsz
;
29 sz
-= the_hash_algo
->rawsz
;
32 si
->delete_bitmap
= ewah_new();
33 ret
= ewah_read_mmap(si
->delete_bitmap
, data
, sz
);
35 return error("corrupt delete bitmap in link extension");
38 si
->replace_bitmap
= ewah_new();
39 ret
= ewah_read_mmap(si
->replace_bitmap
, data
, sz
);
41 return error("corrupt replace bitmap in link extension");
43 return error("garbage at the end of link extension");
47 int write_link_extension(struct strbuf
*sb
,
48 struct index_state
*istate
)
50 struct split_index
*si
= istate
->split_index
;
51 strbuf_add(sb
, si
->base_oid
.hash
, the_hash_algo
->rawsz
);
52 if (!si
->delete_bitmap
&& !si
->replace_bitmap
)
54 ewah_serialize_strbuf(si
->delete_bitmap
, sb
);
55 ewah_serialize_strbuf(si
->replace_bitmap
, sb
);
59 static void mark_base_index_entries(struct index_state
*base
)
63 * To keep track of the shared entries between
64 * istate->base->cache[] and istate->cache[], base entry
65 * position is stored in each base entry. All positions start
66 * from 1 instead of 0, which is reserved to say "this is a new
69 for (i
= 0; i
< base
->cache_nr
; i
++)
70 base
->cache
[i
]->index
= i
+ 1;
73 void move_cache_to_base_index(struct index_state
*istate
)
75 struct split_index
*si
= istate
->split_index
;
79 * If there was a previous base index, then transfer ownership of allocated
80 * entries to the parent index.
83 si
->base
->ce_mem_pool
) {
85 if (!istate
->ce_mem_pool
) {
86 istate
->ce_mem_pool
= xmalloc(sizeof(struct mem_pool
));
87 mem_pool_init(istate
->ce_mem_pool
, 0);
90 mem_pool_combine(istate
->ce_mem_pool
, istate
->split_index
->base
->ce_mem_pool
);
93 ALLOC_ARRAY(si
->base
, 1);
94 index_state_init(si
->base
, istate
->repo
);
95 si
->base
->version
= istate
->version
;
96 /* zero timestamp disables racy test in ce_write_index() */
97 si
->base
->timestamp
= istate
->timestamp
;
98 ALLOC_GROW(si
->base
->cache
, istate
->cache_nr
, si
->base
->cache_alloc
);
99 si
->base
->cache_nr
= istate
->cache_nr
;
102 * The mem_pool needs to move with the allocated entries.
104 si
->base
->ce_mem_pool
= istate
->ce_mem_pool
;
105 istate
->ce_mem_pool
= NULL
;
107 COPY_ARRAY(si
->base
->cache
, istate
->cache
, istate
->cache_nr
);
108 mark_base_index_entries(si
->base
);
109 for (i
= 0; i
< si
->base
->cache_nr
; i
++)
110 si
->base
->cache
[i
]->ce_flags
&= ~CE_UPDATE_IN_BASE
;
113 static void mark_entry_for_delete(size_t pos
, void *data
)
115 struct index_state
*istate
= data
;
116 if (pos
>= istate
->cache_nr
)
117 die("position for delete %d exceeds base index size %d",
118 (int)pos
, istate
->cache_nr
);
119 istate
->cache
[pos
]->ce_flags
|= CE_REMOVE
;
120 istate
->split_index
->nr_deletions
++;
123 static void replace_entry(size_t pos
, void *data
)
125 struct index_state
*istate
= data
;
126 struct split_index
*si
= istate
->split_index
;
127 struct cache_entry
*dst
, *src
;
129 if (pos
>= istate
->cache_nr
)
130 die("position for replacement %d exceeds base index size %d",
131 (int)pos
, istate
->cache_nr
);
132 if (si
->nr_replacements
>= si
->saved_cache_nr
)
133 die("too many replacements (%d vs %d)",
134 si
->nr_replacements
, si
->saved_cache_nr
);
135 dst
= istate
->cache
[pos
];
136 if (dst
->ce_flags
& CE_REMOVE
)
137 die("entry %d is marked as both replaced and deleted",
139 src
= si
->saved_cache
[si
->nr_replacements
];
141 die("corrupt link extension, entry %d should have "
142 "zero length name", (int)pos
);
143 src
->index
= pos
+ 1;
144 src
->ce_flags
|= CE_UPDATE_IN_BASE
;
145 src
->ce_namelen
= dst
->ce_namelen
;
146 copy_cache_entry(dst
, src
);
147 discard_cache_entry(src
);
148 si
->nr_replacements
++;
151 void merge_base_index(struct index_state
*istate
)
153 struct split_index
*si
= istate
->split_index
;
156 mark_base_index_entries(si
->base
);
158 si
->saved_cache
= istate
->cache
;
159 si
->saved_cache_nr
= istate
->cache_nr
;
160 istate
->cache_nr
= si
->base
->cache_nr
;
161 istate
->cache
= NULL
;
162 istate
->cache_alloc
= 0;
163 ALLOC_GROW(istate
->cache
, istate
->cache_nr
, istate
->cache_alloc
);
164 COPY_ARRAY(istate
->cache
, si
->base
->cache
, istate
->cache_nr
);
166 si
->nr_deletions
= 0;
167 si
->nr_replacements
= 0;
168 ewah_each_bit(si
->replace_bitmap
, replace_entry
, istate
);
169 ewah_each_bit(si
->delete_bitmap
, mark_entry_for_delete
, istate
);
170 if (si
->nr_deletions
)
171 remove_marked_cache_entries(istate
, 0);
173 for (i
= si
->nr_replacements
; i
< si
->saved_cache_nr
; i
++) {
174 if (!ce_namelen(si
->saved_cache
[i
]))
175 die("corrupt link extension, entry %d should "
176 "have non-zero length name", i
);
177 add_index_entry(istate
, si
->saved_cache
[i
],
178 ADD_CACHE_OK_TO_ADD
|
179 ADD_CACHE_KEEP_CACHE_TREE
|
181 * we may have to replay what
182 * merge-recursive.c:update_stages()
183 * does, which has this flag on
185 ADD_CACHE_SKIP_DFCHECK
);
186 si
->saved_cache
[i
] = NULL
;
189 ewah_free(si
->delete_bitmap
);
190 ewah_free(si
->replace_bitmap
);
191 FREE_AND_NULL(si
->saved_cache
);
192 si
->delete_bitmap
= NULL
;
193 si
->replace_bitmap
= NULL
;
194 si
->saved_cache_nr
= 0;
198 * Compare most of the fields in two cache entries, i.e. all except the
199 * hashmap_entry and the name.
201 static int compare_ce_content(struct cache_entry
*a
, struct cache_entry
*b
)
203 const unsigned int ondisk_flags
= CE_STAGEMASK
| CE_VALID
|
205 unsigned int ce_flags
= a
->ce_flags
;
206 unsigned int base_flags
= b
->ce_flags
;
209 /* only on-disk flags matter */
210 a
->ce_flags
&= ondisk_flags
;
211 b
->ce_flags
&= ondisk_flags
;
212 ret
= memcmp(&a
->ce_stat_data
, &b
->ce_stat_data
,
213 offsetof(struct cache_entry
, name
) -
214 offsetof(struct cache_entry
, oid
)) ||
215 !oideq(&a
->oid
, &b
->oid
);
216 a
->ce_flags
= ce_flags
;
217 b
->ce_flags
= base_flags
;
222 void prepare_to_write_split_index(struct index_state
*istate
)
224 struct split_index
*si
= init_split_index(istate
);
225 struct cache_entry
**entries
= NULL
, *ce
;
226 int i
, nr_entries
= 0, nr_alloc
= 0;
228 si
->delete_bitmap
= ewah_new();
229 si
->replace_bitmap
= ewah_new();
232 /* Go through istate->cache[] and mark CE_MATCHED to
233 * entry with positive index. We'll go through
234 * base->cache[] later to delete all entries in base
235 * that are not marked with either CE_MATCHED or
236 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
237 * duplicate, deduplicate it.
239 for (i
= 0; i
< istate
->cache_nr
; i
++) {
240 struct cache_entry
*base
;
241 ce
= istate
->cache
[i
];
244 * During simple update index operations this
245 * is a cache entry that is not present in
246 * the shared index. It will be added to the
249 * However, it might also represent a file
250 * that already has a cache entry in the
251 * shared index, but a new index has just
252 * been constructed by unpack_trees(), and
253 * this entry now refers to different content
254 * than what was recorded in the original
255 * index, e.g. during 'read-tree -m HEAD^' or
256 * 'checkout HEAD^'. In this case the
257 * original entry in the shared index will be
258 * marked as deleted, and this entry will be
259 * added to the split index.
263 if (ce
->index
> si
->base
->cache_nr
) {
264 BUG("ce refers to a shared ce at %d, which is beyond the shared index size %d",
265 ce
->index
, si
->base
->cache_nr
);
267 ce
->ce_flags
|= CE_MATCHED
; /* or "shared" */
268 base
= si
->base
->cache
[ce
->index
- 1];
270 /* The entry is present in the shared index. */
271 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
273 * Already marked for inclusion in
274 * the split index, either because
275 * the corresponding file was
276 * modified and the cached stat data
277 * was refreshed, or because there
278 * is already a replacement entry in
280 * Nothing more to do here.
282 } else if (!ce_uptodate(ce
) &&
283 is_racy_timestamp(istate
, ce
)) {
285 * A racily clean cache entry stored
286 * only in the shared index: it must
287 * be added to the split index, so
288 * the subsequent do_write_index()
289 * can smudge its stat data.
291 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
294 * The entry is only present in the
295 * shared index and it was not
297 * Just leave it there.
302 if (ce
->ce_namelen
!= base
->ce_namelen
||
303 strcmp(ce
->name
, base
->name
)) {
308 * This is the copy of a cache entry that is present
309 * in the shared index, created by unpack_trees()
310 * while it constructed a new index.
312 if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
314 * Already marked for inclusion in the split
315 * index, either because the corresponding
316 * file was modified and the cached stat data
317 * was refreshed, or because the original
318 * entry already had a replacement entry in
322 } else if (!ce_uptodate(ce
) &&
323 is_racy_timestamp(istate
, ce
)) {
325 * A copy of a racily clean cache entry from
326 * the shared index. It must be added to
327 * the split index, so the subsequent
328 * do_write_index() can smudge its stat data.
330 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
333 * Thoroughly compare the cached data to see
334 * whether it should be marked for inclusion
335 * in the split index.
337 * This comparison might be unnecessary, as
338 * code paths modifying the cached data do
339 * set CE_UPDATE_IN_BASE as well.
341 if (compare_ce_content(ce
, base
))
342 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
344 discard_cache_entry(base
);
345 si
->base
->cache
[ce
->index
- 1] = ce
;
347 for (i
= 0; i
< si
->base
->cache_nr
; i
++) {
348 ce
= si
->base
->cache
[i
];
349 if ((ce
->ce_flags
& CE_REMOVE
) ||
350 !(ce
->ce_flags
& CE_MATCHED
))
351 ewah_set(si
->delete_bitmap
, i
);
352 else if (ce
->ce_flags
& CE_UPDATE_IN_BASE
) {
353 ewah_set(si
->replace_bitmap
, i
);
354 ce
->ce_flags
|= CE_STRIP_NAME
;
355 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
356 entries
[nr_entries
++] = ce
;
358 if (is_null_oid(&ce
->oid
))
359 istate
->drop_cache_tree
= 1;
363 for (i
= 0; i
< istate
->cache_nr
; i
++) {
364 ce
= istate
->cache
[i
];
365 if ((!si
->base
|| !ce
->index
) && !(ce
->ce_flags
& CE_REMOVE
)) {
366 assert(!(ce
->ce_flags
& CE_STRIP_NAME
));
367 ALLOC_GROW(entries
, nr_entries
+1, nr_alloc
);
368 entries
[nr_entries
++] = ce
;
370 ce
->ce_flags
&= ~CE_MATCHED
;
374 * take cache[] out temporarily, put entries[] in its place
377 si
->saved_cache
= istate
->cache
;
378 si
->saved_cache_nr
= istate
->cache_nr
;
379 istate
->cache
= entries
;
380 istate
->cache_nr
= nr_entries
;
383 void finish_writing_split_index(struct index_state
*istate
)
385 struct split_index
*si
= init_split_index(istate
);
387 ewah_free(si
->delete_bitmap
);
388 ewah_free(si
->replace_bitmap
);
389 si
->delete_bitmap
= NULL
;
390 si
->replace_bitmap
= NULL
;
392 istate
->cache
= si
->saved_cache
;
393 istate
->cache_nr
= si
->saved_cache_nr
;
396 void discard_split_index(struct index_state
*istate
)
398 struct split_index
*si
= istate
->split_index
;
401 istate
->split_index
= NULL
;
406 discard_index(si
->base
);
412 void save_or_free_index_entry(struct index_state
*istate
, struct cache_entry
*ce
)
415 istate
->split_index
&&
416 istate
->split_index
->base
&&
417 ce
->index
<= istate
->split_index
->base
->cache_nr
&&
418 ce
== istate
->split_index
->base
->cache
[ce
->index
- 1])
419 ce
->ce_flags
|= CE_REMOVE
;
421 discard_cache_entry(ce
);
424 void replace_index_entry_in_base(struct index_state
*istate
,
425 struct cache_entry
*old_entry
,
426 struct cache_entry
*new_entry
)
428 if (old_entry
->index
&&
429 istate
->split_index
&&
430 istate
->split_index
->base
&&
431 old_entry
->index
<= istate
->split_index
->base
->cache_nr
) {
432 new_entry
->index
= old_entry
->index
;
433 if (old_entry
!= istate
->split_index
->base
->cache
[new_entry
->index
- 1])
434 discard_cache_entry(istate
->split_index
->base
->cache
[new_entry
->index
- 1]);
435 istate
->split_index
->base
->cache
[new_entry
->index
- 1] = new_entry
;
439 void add_split_index(struct index_state
*istate
)
441 if (!istate
->split_index
) {
442 init_split_index(istate
);
443 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
447 void remove_split_index(struct index_state
*istate
)
449 if (istate
->split_index
) {
450 if (istate
->split_index
->base
) {
452 * When removing the split index, we need to move
453 * ownership of the mem_pool associated with the
454 * base index to the main index. There may be cache entries
455 * allocated from the base's memory pool that are shared with
458 mem_pool_combine(istate
->ce_mem_pool
,
459 istate
->split_index
->base
->ce_mem_pool
);
462 * The split index no longer owns the mem_pool backing
463 * its cache array. As we are discarding this index,
464 * mark the index as having no cache entries, so it
465 * will not attempt to clean up the cache entries or
468 istate
->split_index
->base
->cache_nr
= 0;
472 * We can discard the split index because its
473 * memory pool has been incorporated into the
474 * memory pool associated with the the_index.
476 discard_split_index(istate
);
478 istate
->cache_changed
|= SOMETHING_CHANGED
;