Merge branch 'tb/complete-word-diff-regex' into maint
[git] / cache-tree.c
1 #include "cache.h"
2 #include "lockfile.h"
3 #include "tree.h"
4 #include "tree-walk.h"
5 #include "cache-tree.h"
6
7 #ifndef DEBUG
8 #define DEBUG 0
9 #endif
10
11 struct cache_tree *cache_tree(void)
12 {
13         struct cache_tree *it = xcalloc(1, sizeof(struct cache_tree));
14         it->entry_count = -1;
15         return it;
16 }
17
18 void cache_tree_free(struct cache_tree **it_p)
19 {
20         int i;
21         struct cache_tree *it = *it_p;
22
23         if (!it)
24                 return;
25         for (i = 0; i < it->subtree_nr; i++)
26                 if (it->down[i]) {
27                         cache_tree_free(&it->down[i]->cache_tree);
28                         free(it->down[i]);
29                 }
30         free(it->down);
31         free(it);
32         *it_p = NULL;
33 }
34
35 static int subtree_name_cmp(const char *one, int onelen,
36                             const char *two, int twolen)
37 {
38         if (onelen < twolen)
39                 return -1;
40         if (twolen < onelen)
41                 return 1;
42         return memcmp(one, two, onelen);
43 }
44
45 static int subtree_pos(struct cache_tree *it, const char *path, int pathlen)
46 {
47         struct cache_tree_sub **down = it->down;
48         int lo, hi;
49         lo = 0;
50         hi = it->subtree_nr;
51         while (lo < hi) {
52                 int mi = (lo + hi) / 2;
53                 struct cache_tree_sub *mdl = down[mi];
54                 int cmp = subtree_name_cmp(path, pathlen,
55                                            mdl->name, mdl->namelen);
56                 if (!cmp)
57                         return mi;
58                 if (cmp < 0)
59                         hi = mi;
60                 else
61                         lo = mi + 1;
62         }
63         return -lo-1;
64 }
65
66 static struct cache_tree_sub *find_subtree(struct cache_tree *it,
67                                            const char *path,
68                                            int pathlen,
69                                            int create)
70 {
71         struct cache_tree_sub *down;
72         int pos = subtree_pos(it, path, pathlen);
73         if (0 <= pos)
74                 return it->down[pos];
75         if (!create)
76                 return NULL;
77
78         pos = -pos-1;
79         ALLOC_GROW(it->down, it->subtree_nr + 1, it->subtree_alloc);
80         it->subtree_nr++;
81
82         down = xmalloc(sizeof(*down) + pathlen + 1);
83         down->cache_tree = NULL;
84         down->namelen = pathlen;
85         memcpy(down->name, path, pathlen);
86         down->name[pathlen] = 0;
87
88         if (pos < it->subtree_nr)
89                 memmove(it->down + pos + 1,
90                         it->down + pos,
91                         sizeof(down) * (it->subtree_nr - pos - 1));
92         it->down[pos] = down;
93         return down;
94 }
95
96 struct cache_tree_sub *cache_tree_sub(struct cache_tree *it, const char *path)
97 {
98         int pathlen = strlen(path);
99         return find_subtree(it, path, pathlen, 1);
100 }
101
102 static int do_invalidate_path(struct cache_tree *it, const char *path)
103 {
104         /* a/b/c
105          * ==> invalidate self
106          * ==> find "a", have it invalidate "b/c"
107          * a
108          * ==> invalidate self
109          * ==> if "a" exists as a subtree, remove it.
110          */
111         const char *slash;
112         int namelen;
113         struct cache_tree_sub *down;
114
115 #if DEBUG
116         fprintf(stderr, "cache-tree invalidate <%s>\n", path);
117 #endif
118
119         if (!it)
120                 return 0;
121         slash = strchrnul(path, '/');
122         namelen = slash - path;
123         it->entry_count = -1;
124         if (!*slash) {
125                 int pos;
126                 pos = subtree_pos(it, path, namelen);
127                 if (0 <= pos) {
128                         cache_tree_free(&it->down[pos]->cache_tree);
129                         free(it->down[pos]);
130                         /* 0 1 2 3 4 5
131                          *       ^     ^subtree_nr = 6
132                          *       pos
133                          * move 4 and 5 up one place (2 entries)
134                          * 2 = 6 - 3 - 1 = subtree_nr - pos - 1
135                          */
136                         memmove(it->down+pos, it->down+pos+1,
137                                 sizeof(struct cache_tree_sub *) *
138                                 (it->subtree_nr - pos - 1));
139                         it->subtree_nr--;
140                 }
141                 return 1;
142         }
143         down = find_subtree(it, path, namelen, 0);
144         if (down)
145                 do_invalidate_path(down->cache_tree, slash + 1);
146         return 1;
147 }
148
149 void cache_tree_invalidate_path(struct index_state *istate, const char *path)
150 {
151         if (do_invalidate_path(istate->cache_tree, path))
152                 istate->cache_changed |= CACHE_TREE_CHANGED;
153 }
154
155 static int verify_cache(struct cache_entry **cache,
156                         int entries, int flags)
157 {
158         int i, funny;
159         int silent = flags & WRITE_TREE_SILENT;
160
161         /* Verify that the tree is merged */
162         funny = 0;
163         for (i = 0; i < entries; i++) {
164                 const struct cache_entry *ce = cache[i];
165                 if (ce_stage(ce)) {
166                         if (silent)
167                                 return -1;
168                         if (10 < ++funny) {
169                                 fprintf(stderr, "...\n");
170                                 break;
171                         }
172                         fprintf(stderr, "%s: unmerged (%s)\n",
173                                 ce->name, sha1_to_hex(ce->sha1));
174                 }
175         }
176         if (funny)
177                 return -1;
178
179         /* Also verify that the cache does not have path and path/file
180          * at the same time.  At this point we know the cache has only
181          * stage 0 entries.
182          */
183         funny = 0;
184         for (i = 0; i < entries - 1; i++) {
185                 /* path/file always comes after path because of the way
186                  * the cache is sorted.  Also path can appear only once,
187                  * which means conflicting one would immediately follow.
188                  */
189                 const char *this_name = cache[i]->name;
190                 const char *next_name = cache[i+1]->name;
191                 int this_len = strlen(this_name);
192                 if (this_len < strlen(next_name) &&
193                     strncmp(this_name, next_name, this_len) == 0 &&
194                     next_name[this_len] == '/') {
195                         if (10 < ++funny) {
196                                 fprintf(stderr, "...\n");
197                                 break;
198                         }
199                         fprintf(stderr, "You have both %s and %s\n",
200                                 this_name, next_name);
201                 }
202         }
203         if (funny)
204                 return -1;
205         return 0;
206 }
207
208 static void discard_unused_subtrees(struct cache_tree *it)
209 {
210         struct cache_tree_sub **down = it->down;
211         int nr = it->subtree_nr;
212         int dst, src;
213         for (dst = src = 0; src < nr; src++) {
214                 struct cache_tree_sub *s = down[src];
215                 if (s->used)
216                         down[dst++] = s;
217                 else {
218                         cache_tree_free(&s->cache_tree);
219                         free(s);
220                         it->subtree_nr--;
221                 }
222         }
223 }
224
225 int cache_tree_fully_valid(struct cache_tree *it)
226 {
227         int i;
228         if (!it)
229                 return 0;
230         if (it->entry_count < 0 || !has_sha1_file(it->sha1))
231                 return 0;
232         for (i = 0; i < it->subtree_nr; i++) {
233                 if (!cache_tree_fully_valid(it->down[i]->cache_tree))
234                         return 0;
235         }
236         return 1;
237 }
238
239 static int update_one(struct cache_tree *it,
240                       struct cache_entry **cache,
241                       int entries,
242                       const char *base,
243                       int baselen,
244                       int *skip_count,
245                       int flags)
246 {
247         struct strbuf buffer;
248         int missing_ok = flags & WRITE_TREE_MISSING_OK;
249         int dryrun = flags & WRITE_TREE_DRY_RUN;
250         int repair = flags & WRITE_TREE_REPAIR;
251         int to_invalidate = 0;
252         int i;
253
254         assert(!(dryrun && repair));
255
256         *skip_count = 0;
257
258         if (0 <= it->entry_count && has_sha1_file(it->sha1))
259                 return it->entry_count;
260
261         /*
262          * We first scan for subtrees and update them; we start by
263          * marking existing subtrees -- the ones that are unmarked
264          * should not be in the result.
265          */
266         for (i = 0; i < it->subtree_nr; i++)
267                 it->down[i]->used = 0;
268
269         /*
270          * Find the subtrees and update them.
271          */
272         i = 0;
273         while (i < entries) {
274                 const struct cache_entry *ce = cache[i];
275                 struct cache_tree_sub *sub;
276                 const char *path, *slash;
277                 int pathlen, sublen, subcnt, subskip;
278
279                 path = ce->name;
280                 pathlen = ce_namelen(ce);
281                 if (pathlen <= baselen || memcmp(base, path, baselen))
282                         break; /* at the end of this level */
283
284                 slash = strchr(path + baselen, '/');
285                 if (!slash) {
286                         i++;
287                         continue;
288                 }
289                 /*
290                  * a/bbb/c (base = a/, slash = /c)
291                  * ==>
292                  * path+baselen = bbb/c, sublen = 3
293                  */
294                 sublen = slash - (path + baselen);
295                 sub = find_subtree(it, path + baselen, sublen, 1);
296                 if (!sub->cache_tree)
297                         sub->cache_tree = cache_tree();
298                 subcnt = update_one(sub->cache_tree,
299                                     cache + i, entries - i,
300                                     path,
301                                     baselen + sublen + 1,
302                                     &subskip,
303                                     flags);
304                 if (subcnt < 0)
305                         return subcnt;
306                 if (!subcnt)
307                         die("index cache-tree records empty sub-tree");
308                 i += subcnt;
309                 sub->count = subcnt; /* to be used in the next loop */
310                 *skip_count += subskip;
311                 sub->used = 1;
312         }
313
314         discard_unused_subtrees(it);
315
316         /*
317          * Then write out the tree object for this level.
318          */
319         strbuf_init(&buffer, 8192);
320
321         i = 0;
322         while (i < entries) {
323                 const struct cache_entry *ce = cache[i];
324                 struct cache_tree_sub *sub;
325                 const char *path, *slash;
326                 int pathlen, entlen;
327                 const unsigned char *sha1;
328                 unsigned mode;
329                 int expected_missing = 0;
330
331                 path = ce->name;
332                 pathlen = ce_namelen(ce);
333                 if (pathlen <= baselen || memcmp(base, path, baselen))
334                         break; /* at the end of this level */
335
336                 slash = strchr(path + baselen, '/');
337                 if (slash) {
338                         entlen = slash - (path + baselen);
339                         sub = find_subtree(it, path + baselen, entlen, 0);
340                         if (!sub)
341                                 die("cache-tree.c: '%.*s' in '%s' not found",
342                                     entlen, path + baselen, path);
343                         i += sub->count;
344                         sha1 = sub->cache_tree->sha1;
345                         mode = S_IFDIR;
346                         if (sub->cache_tree->entry_count < 0) {
347                                 to_invalidate = 1;
348                                 expected_missing = 1;
349                         }
350                 }
351                 else {
352                         sha1 = ce->sha1;
353                         mode = ce->ce_mode;
354                         entlen = pathlen - baselen;
355                         i++;
356                 }
357                 if (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1)) {
358                         strbuf_release(&buffer);
359                         if (expected_missing)
360                                 return -1;
361                         return error("invalid object %06o %s for '%.*s'",
362                                 mode, sha1_to_hex(sha1), entlen+baselen, path);
363                 }
364
365                 /*
366                  * CE_REMOVE entries are removed before the index is
367                  * written to disk. Skip them to remain consistent
368                  * with the future on-disk index.
369                  */
370                 if (ce->ce_flags & CE_REMOVE) {
371                         *skip_count = *skip_count + 1;
372                         continue;
373                 }
374
375                 /*
376                  * CE_INTENT_TO_ADD entries exist on on-disk index but
377                  * they are not part of generated trees. Invalidate up
378                  * to root to force cache-tree users to read elsewhere.
379                  */
380                 if (ce->ce_flags & CE_INTENT_TO_ADD) {
381                         to_invalidate = 1;
382                         continue;
383                 }
384
385                 strbuf_grow(&buffer, entlen + 100);
386                 strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0');
387                 strbuf_add(&buffer, sha1, 20);
388
389 #if DEBUG
390                 fprintf(stderr, "cache-tree update-one %o %.*s\n",
391                         mode, entlen, path + baselen);
392 #endif
393         }
394
395         if (repair) {
396                 unsigned char sha1[20];
397                 hash_sha1_file(buffer.buf, buffer.len, tree_type, sha1);
398                 if (has_sha1_file(sha1))
399                         hashcpy(it->sha1, sha1);
400                 else
401                         to_invalidate = 1;
402         } else if (dryrun)
403                 hash_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1);
404         else if (write_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1)) {
405                 strbuf_release(&buffer);
406                 return -1;
407         }
408
409         strbuf_release(&buffer);
410         it->entry_count = to_invalidate ? -1 : i - *skip_count;
411 #if DEBUG
412         fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
413                 it->entry_count, it->subtree_nr,
414                 sha1_to_hex(it->sha1));
415 #endif
416         return i;
417 }
418
419 int cache_tree_update(struct index_state *istate, int flags)
420 {
421         struct cache_tree *it = istate->cache_tree;
422         struct cache_entry **cache = istate->cache;
423         int entries = istate->cache_nr;
424         int skip, i = verify_cache(cache, entries, flags);
425
426         if (i)
427                 return i;
428         i = update_one(it, cache, entries, "", 0, &skip, flags);
429         if (i < 0)
430                 return i;
431         istate->cache_changed |= CACHE_TREE_CHANGED;
432         return 0;
433 }
434
435 static void write_one(struct strbuf *buffer, struct cache_tree *it,
436                       const char *path, int pathlen)
437 {
438         int i;
439
440         /* One "cache-tree" entry consists of the following:
441          * path (NUL terminated)
442          * entry_count, subtree_nr ("%d %d\n")
443          * tree-sha1 (missing if invalid)
444          * subtree_nr "cache-tree" entries for subtrees.
445          */
446         strbuf_grow(buffer, pathlen + 100);
447         strbuf_add(buffer, path, pathlen);
448         strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr);
449
450 #if DEBUG
451         if (0 <= it->entry_count)
452                 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
453                         pathlen, path, it->entry_count, it->subtree_nr,
454                         sha1_to_hex(it->sha1));
455         else
456                 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
457                         pathlen, path, it->subtree_nr);
458 #endif
459
460         if (0 <= it->entry_count) {
461                 strbuf_add(buffer, it->sha1, 20);
462         }
463         for (i = 0; i < it->subtree_nr; i++) {
464                 struct cache_tree_sub *down = it->down[i];
465                 if (i) {
466                         struct cache_tree_sub *prev = it->down[i-1];
467                         if (subtree_name_cmp(down->name, down->namelen,
468                                              prev->name, prev->namelen) <= 0)
469                                 die("fatal - unsorted cache subtree");
470                 }
471                 write_one(buffer, down->cache_tree, down->name, down->namelen);
472         }
473 }
474
475 void cache_tree_write(struct strbuf *sb, struct cache_tree *root)
476 {
477         write_one(sb, root, "", 0);
478 }
479
480 static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
481 {
482         const char *buf = *buffer;
483         unsigned long size = *size_p;
484         const char *cp;
485         char *ep;
486         struct cache_tree *it;
487         int i, subtree_nr;
488
489         it = NULL;
490         /* skip name, but make sure name exists */
491         while (size && *buf) {
492                 size--;
493                 buf++;
494         }
495         if (!size)
496                 goto free_return;
497         buf++; size--;
498         it = cache_tree();
499
500         cp = buf;
501         it->entry_count = strtol(cp, &ep, 10);
502         if (cp == ep)
503                 goto free_return;
504         cp = ep;
505         subtree_nr = strtol(cp, &ep, 10);
506         if (cp == ep)
507                 goto free_return;
508         while (size && *buf && *buf != '\n') {
509                 size--;
510                 buf++;
511         }
512         if (!size)
513                 goto free_return;
514         buf++; size--;
515         if (0 <= it->entry_count) {
516                 if (size < 20)
517                         goto free_return;
518                 hashcpy(it->sha1, (const unsigned char*)buf);
519                 buf += 20;
520                 size -= 20;
521         }
522
523 #if DEBUG
524         if (0 <= it->entry_count)
525                 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
526                         *buffer, it->entry_count, subtree_nr,
527                         sha1_to_hex(it->sha1));
528         else
529                 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
530                         *buffer, subtree_nr);
531 #endif
532
533         /*
534          * Just a heuristic -- we do not add directories that often but
535          * we do not want to have to extend it immediately when we do,
536          * hence +2.
537          */
538         it->subtree_alloc = subtree_nr + 2;
539         it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
540         for (i = 0; i < subtree_nr; i++) {
541                 /* read each subtree */
542                 struct cache_tree *sub;
543                 struct cache_tree_sub *subtree;
544                 const char *name = buf;
545
546                 sub = read_one(&buf, &size);
547                 if (!sub)
548                         goto free_return;
549                 subtree = cache_tree_sub(it, name);
550                 subtree->cache_tree = sub;
551         }
552         if (subtree_nr != it->subtree_nr)
553                 die("cache-tree: internal error");
554         *buffer = buf;
555         *size_p = size;
556         return it;
557
558  free_return:
559         cache_tree_free(&it);
560         return NULL;
561 }
562
563 struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
564 {
565         if (buffer[0])
566                 return NULL; /* not the whole tree */
567         return read_one(&buffer, &size);
568 }
569
570 static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
571 {
572         if (!it)
573                 return NULL;
574         while (*path) {
575                 const char *slash;
576                 struct cache_tree_sub *sub;
577
578                 slash = strchrnul(path, '/');
579                 /*
580                  * Between path and slash is the name of the subtree
581                  * to look for.
582                  */
583                 sub = find_subtree(it, path, slash - path, 0);
584                 if (!sub)
585                         return NULL;
586                 it = sub->cache_tree;
587
588                 path = slash;
589                 while (*path == '/')
590                         path++;
591         }
592         return it;
593 }
594
595 int write_index_as_tree(unsigned char *sha1, struct index_state *index_state, const char *index_path, int flags, const char *prefix)
596 {
597         int entries, was_valid, newfd;
598         struct lock_file *lock_file;
599
600         /*
601          * We can't free this memory, it becomes part of a linked list
602          * parsed atexit()
603          */
604         lock_file = xcalloc(1, sizeof(struct lock_file));
605
606         newfd = hold_lock_file_for_update(lock_file, index_path, LOCK_DIE_ON_ERROR);
607
608         entries = read_index_from(index_state, index_path);
609         if (entries < 0)
610                 return WRITE_TREE_UNREADABLE_INDEX;
611         if (flags & WRITE_TREE_IGNORE_CACHE_TREE)
612                 cache_tree_free(&index_state->cache_tree);
613
614         if (!index_state->cache_tree)
615                 index_state->cache_tree = cache_tree();
616
617         was_valid = cache_tree_fully_valid(index_state->cache_tree);
618         if (!was_valid) {
619                 if (cache_tree_update(index_state, flags) < 0)
620                         return WRITE_TREE_UNMERGED_INDEX;
621                 if (0 <= newfd) {
622                         if (!write_locked_index(index_state, lock_file, COMMIT_LOCK))
623                                 newfd = -1;
624                 }
625                 /* Not being able to write is fine -- we are only interested
626                  * in updating the cache-tree part, and if the next caller
627                  * ends up using the old index with unupdated cache-tree part
628                  * it misses the work we did here, but that is just a
629                  * performance penalty and not a big deal.
630                  */
631         }
632
633         if (prefix) {
634                 struct cache_tree *subtree;
635                 subtree = cache_tree_find(index_state->cache_tree, prefix);
636                 if (!subtree)
637                         return WRITE_TREE_PREFIX_ERROR;
638                 hashcpy(sha1, subtree->sha1);
639         }
640         else
641                 hashcpy(sha1, index_state->cache_tree->sha1);
642
643         if (0 <= newfd)
644                 rollback_lock_file(lock_file);
645
646         return 0;
647 }
648
649 int write_cache_as_tree(unsigned char *sha1, int flags, const char *prefix)
650 {
651         return write_index_as_tree(sha1, &the_index, get_index_file(), flags, prefix);
652 }
653
654 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
655 {
656         struct tree_desc desc;
657         struct name_entry entry;
658         int cnt;
659
660         hashcpy(it->sha1, tree->object.oid.hash);
661         init_tree_desc(&desc, tree->buffer, tree->size);
662         cnt = 0;
663         while (tree_entry(&desc, &entry)) {
664                 if (!S_ISDIR(entry.mode))
665                         cnt++;
666                 else {
667                         struct cache_tree_sub *sub;
668                         struct tree *subtree = lookup_tree(entry.sha1);
669                         if (!subtree->object.parsed)
670                                 parse_tree(subtree);
671                         sub = cache_tree_sub(it, entry.path);
672                         sub->cache_tree = cache_tree();
673                         prime_cache_tree_rec(sub->cache_tree, subtree);
674                         cnt += sub->cache_tree->entry_count;
675                 }
676         }
677         it->entry_count = cnt;
678 }
679
680 void prime_cache_tree(struct index_state *istate, struct tree *tree)
681 {
682         cache_tree_free(&istate->cache_tree);
683         istate->cache_tree = cache_tree();
684         prime_cache_tree_rec(istate->cache_tree, tree);
685         istate->cache_changed |= CACHE_TREE_CHANGED;
686 }
687
688 /*
689  * find the cache_tree that corresponds to the current level without
690  * exploding the full path into textual form.  The root of the
691  * cache tree is given as "root", and our current level is "info".
692  * (1) When at root level, info->prev is NULL, so it is "root" itself.
693  * (2) Otherwise, find the cache_tree that corresponds to one level
694  *     above us, and find ourselves in there.
695  */
696 static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root,
697                                                          struct traverse_info *info)
698 {
699         struct cache_tree *our_parent;
700
701         if (!info->prev)
702                 return root;
703         our_parent = find_cache_tree_from_traversal(root, info->prev);
704         return cache_tree_find(our_parent, info->name.path);
705 }
706
707 int cache_tree_matches_traversal(struct cache_tree *root,
708                                  struct name_entry *ent,
709                                  struct traverse_info *info)
710 {
711         struct cache_tree *it;
712
713         it = find_cache_tree_from_traversal(root, info);
714         it = cache_tree_find(it, ent->path);
715         if (it && it->entry_count > 0 && !hashcmp(ent->sha1, it->sha1))
716                 return it->entry_count;
717         return 0;
718 }
719
720 int update_main_cache_tree(int flags)
721 {
722         if (!the_index.cache_tree)
723                 the_index.cache_tree = cache_tree();
724         return cache_tree_update(&the_index, flags);
725 }