git-svn: disable _rev_list memoization
[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                 i += subcnt;
307                 sub->count = subcnt; /* to be used in the next loop */
308                 *skip_count += subskip;
309                 sub->used = 1;
310         }
311
312         discard_unused_subtrees(it);
313
314         /*
315          * Then write out the tree object for this level.
316          */
317         strbuf_init(&buffer, 8192);
318
319         i = 0;
320         while (i < entries) {
321                 const struct cache_entry *ce = cache[i];
322                 struct cache_tree_sub *sub;
323                 const char *path, *slash;
324                 int pathlen, entlen;
325                 const unsigned char *sha1;
326                 unsigned mode;
327                 int expected_missing = 0;
328
329                 path = ce->name;
330                 pathlen = ce_namelen(ce);
331                 if (pathlen <= baselen || memcmp(base, path, baselen))
332                         break; /* at the end of this level */
333
334                 slash = strchr(path + baselen, '/');
335                 if (slash) {
336                         entlen = slash - (path + baselen);
337                         sub = find_subtree(it, path + baselen, entlen, 0);
338                         if (!sub)
339                                 die("cache-tree.c: '%.*s' in '%s' not found",
340                                     entlen, path + baselen, path);
341                         i += sub->count;
342                         sha1 = sub->cache_tree->sha1;
343                         mode = S_IFDIR;
344                         if (sub->cache_tree->entry_count < 0) {
345                                 to_invalidate = 1;
346                                 expected_missing = 1;
347                         }
348                 }
349                 else {
350                         sha1 = ce->sha1;
351                         mode = ce->ce_mode;
352                         entlen = pathlen - baselen;
353                         i++;
354                 }
355                 if (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1)) {
356                         strbuf_release(&buffer);
357                         if (expected_missing)
358                                 return -1;
359                         return error("invalid object %06o %s for '%.*s'",
360                                 mode, sha1_to_hex(sha1), entlen+baselen, path);
361                 }
362
363                 /*
364                  * CE_REMOVE entries are removed before the index is
365                  * written to disk. Skip them to remain consistent
366                  * with the future on-disk index.
367                  */
368                 if (ce->ce_flags & CE_REMOVE) {
369                         *skip_count = *skip_count + 1;
370                         continue;
371                 }
372
373                 /*
374                  * CE_INTENT_TO_ADD entries exist on on-disk index but
375                  * they are not part of generated trees. Invalidate up
376                  * to root to force cache-tree users to read elsewhere.
377                  */
378                 if (ce->ce_flags & CE_INTENT_TO_ADD) {
379                         to_invalidate = 1;
380                         continue;
381                 }
382
383                 strbuf_grow(&buffer, entlen + 100);
384                 strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0');
385                 strbuf_add(&buffer, sha1, 20);
386
387 #if DEBUG
388                 fprintf(stderr, "cache-tree update-one %o %.*s\n",
389                         mode, entlen, path + baselen);
390 #endif
391         }
392
393         if (repair) {
394                 unsigned char sha1[20];
395                 hash_sha1_file(buffer.buf, buffer.len, tree_type, sha1);
396                 if (has_sha1_file(sha1))
397                         hashcpy(it->sha1, sha1);
398                 else
399                         to_invalidate = 1;
400         } else if (dryrun)
401                 hash_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1);
402         else if (write_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1)) {
403                 strbuf_release(&buffer);
404                 return -1;
405         }
406
407         strbuf_release(&buffer);
408         it->entry_count = to_invalidate ? -1 : i - *skip_count;
409 #if DEBUG
410         fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
411                 it->entry_count, it->subtree_nr,
412                 sha1_to_hex(it->sha1));
413 #endif
414         return i;
415 }
416
417 int cache_tree_update(struct index_state *istate, int flags)
418 {
419         struct cache_tree *it = istate->cache_tree;
420         struct cache_entry **cache = istate->cache;
421         int entries = istate->cache_nr;
422         int skip, i = verify_cache(cache, entries, flags);
423
424         if (i)
425                 return i;
426         i = update_one(it, cache, entries, "", 0, &skip, flags);
427         if (i < 0)
428                 return i;
429         istate->cache_changed |= CACHE_TREE_CHANGED;
430         return 0;
431 }
432
433 static void write_one(struct strbuf *buffer, struct cache_tree *it,
434                       const char *path, int pathlen)
435 {
436         int i;
437
438         /* One "cache-tree" entry consists of the following:
439          * path (NUL terminated)
440          * entry_count, subtree_nr ("%d %d\n")
441          * tree-sha1 (missing if invalid)
442          * subtree_nr "cache-tree" entries for subtrees.
443          */
444         strbuf_grow(buffer, pathlen + 100);
445         strbuf_add(buffer, path, pathlen);
446         strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr);
447
448 #if DEBUG
449         if (0 <= it->entry_count)
450                 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
451                         pathlen, path, it->entry_count, it->subtree_nr,
452                         sha1_to_hex(it->sha1));
453         else
454                 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
455                         pathlen, path, it->subtree_nr);
456 #endif
457
458         if (0 <= it->entry_count) {
459                 strbuf_add(buffer, it->sha1, 20);
460         }
461         for (i = 0; i < it->subtree_nr; i++) {
462                 struct cache_tree_sub *down = it->down[i];
463                 if (i) {
464                         struct cache_tree_sub *prev = it->down[i-1];
465                         if (subtree_name_cmp(down->name, down->namelen,
466                                              prev->name, prev->namelen) <= 0)
467                                 die("fatal - unsorted cache subtree");
468                 }
469                 write_one(buffer, down->cache_tree, down->name, down->namelen);
470         }
471 }
472
473 void cache_tree_write(struct strbuf *sb, struct cache_tree *root)
474 {
475         write_one(sb, root, "", 0);
476 }
477
478 static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
479 {
480         const char *buf = *buffer;
481         unsigned long size = *size_p;
482         const char *cp;
483         char *ep;
484         struct cache_tree *it;
485         int i, subtree_nr;
486
487         it = NULL;
488         /* skip name, but make sure name exists */
489         while (size && *buf) {
490                 size--;
491                 buf++;
492         }
493         if (!size)
494                 goto free_return;
495         buf++; size--;
496         it = cache_tree();
497
498         cp = buf;
499         it->entry_count = strtol(cp, &ep, 10);
500         if (cp == ep)
501                 goto free_return;
502         cp = ep;
503         subtree_nr = strtol(cp, &ep, 10);
504         if (cp == ep)
505                 goto free_return;
506         while (size && *buf && *buf != '\n') {
507                 size--;
508                 buf++;
509         }
510         if (!size)
511                 goto free_return;
512         buf++; size--;
513         if (0 <= it->entry_count) {
514                 if (size < 20)
515                         goto free_return;
516                 hashcpy(it->sha1, (const unsigned char*)buf);
517                 buf += 20;
518                 size -= 20;
519         }
520
521 #if DEBUG
522         if (0 <= it->entry_count)
523                 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
524                         *buffer, it->entry_count, subtree_nr,
525                         sha1_to_hex(it->sha1));
526         else
527                 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
528                         *buffer, subtree_nr);
529 #endif
530
531         /*
532          * Just a heuristic -- we do not add directories that often but
533          * we do not want to have to extend it immediately when we do,
534          * hence +2.
535          */
536         it->subtree_alloc = subtree_nr + 2;
537         it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
538         for (i = 0; i < subtree_nr; i++) {
539                 /* read each subtree */
540                 struct cache_tree *sub;
541                 struct cache_tree_sub *subtree;
542                 const char *name = buf;
543
544                 sub = read_one(&buf, &size);
545                 if (!sub)
546                         goto free_return;
547                 subtree = cache_tree_sub(it, name);
548                 subtree->cache_tree = sub;
549         }
550         if (subtree_nr != it->subtree_nr)
551                 die("cache-tree: internal error");
552         *buffer = buf;
553         *size_p = size;
554         return it;
555
556  free_return:
557         cache_tree_free(&it);
558         return NULL;
559 }
560
561 struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
562 {
563         if (buffer[0])
564                 return NULL; /* not the whole tree */
565         return read_one(&buffer, &size);
566 }
567
568 static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
569 {
570         if (!it)
571                 return NULL;
572         while (*path) {
573                 const char *slash;
574                 struct cache_tree_sub *sub;
575
576                 slash = strchrnul(path, '/');
577                 /*
578                  * Between path and slash is the name of the subtree
579                  * to look for.
580                  */
581                 sub = find_subtree(it, path, slash - path, 0);
582                 if (!sub)
583                         return NULL;
584                 it = sub->cache_tree;
585
586                 path = slash;
587                 while (*path == '/')
588                         path++;
589         }
590         return it;
591 }
592
593 int write_cache_as_tree(unsigned char *sha1, int flags, const char *prefix)
594 {
595         int entries, was_valid, newfd;
596         struct lock_file *lock_file;
597
598         /*
599          * We can't free this memory, it becomes part of a linked list
600          * parsed atexit()
601          */
602         lock_file = xcalloc(1, sizeof(struct lock_file));
603
604         newfd = hold_locked_index(lock_file, 1);
605
606         entries = read_cache();
607         if (entries < 0)
608                 return WRITE_TREE_UNREADABLE_INDEX;
609         if (flags & WRITE_TREE_IGNORE_CACHE_TREE)
610                 cache_tree_free(&(active_cache_tree));
611
612         if (!active_cache_tree)
613                 active_cache_tree = cache_tree();
614
615         was_valid = cache_tree_fully_valid(active_cache_tree);
616         if (!was_valid) {
617                 if (cache_tree_update(&the_index, flags) < 0)
618                         return WRITE_TREE_UNMERGED_INDEX;
619                 if (0 <= newfd) {
620                         if (!write_locked_index(&the_index, lock_file, COMMIT_LOCK))
621                                 newfd = -1;
622                 }
623                 /* Not being able to write is fine -- we are only interested
624                  * in updating the cache-tree part, and if the next caller
625                  * ends up using the old index with unupdated cache-tree part
626                  * it misses the work we did here, but that is just a
627                  * performance penalty and not a big deal.
628                  */
629         }
630
631         if (prefix) {
632                 struct cache_tree *subtree =
633                         cache_tree_find(active_cache_tree, prefix);
634                 if (!subtree)
635                         return WRITE_TREE_PREFIX_ERROR;
636                 hashcpy(sha1, subtree->sha1);
637         }
638         else
639                 hashcpy(sha1, active_cache_tree->sha1);
640
641         if (0 <= newfd)
642                 rollback_lock_file(lock_file);
643
644         return 0;
645 }
646
647 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
648 {
649         struct tree_desc desc;
650         struct name_entry entry;
651         int cnt;
652
653         hashcpy(it->sha1, tree->object.sha1);
654         init_tree_desc(&desc, tree->buffer, tree->size);
655         cnt = 0;
656         while (tree_entry(&desc, &entry)) {
657                 if (!S_ISDIR(entry.mode))
658                         cnt++;
659                 else {
660                         struct cache_tree_sub *sub;
661                         struct tree *subtree = lookup_tree(entry.sha1);
662                         if (!subtree->object.parsed)
663                                 parse_tree(subtree);
664                         sub = cache_tree_sub(it, entry.path);
665                         sub->cache_tree = cache_tree();
666                         prime_cache_tree_rec(sub->cache_tree, subtree);
667                         cnt += sub->cache_tree->entry_count;
668                 }
669         }
670         it->entry_count = cnt;
671 }
672
673 void prime_cache_tree(struct index_state *istate, struct tree *tree)
674 {
675         cache_tree_free(&istate->cache_tree);
676         istate->cache_tree = cache_tree();
677         prime_cache_tree_rec(istate->cache_tree, tree);
678         istate->cache_changed |= CACHE_TREE_CHANGED;
679 }
680
681 /*
682  * find the cache_tree that corresponds to the current level without
683  * exploding the full path into textual form.  The root of the
684  * cache tree is given as "root", and our current level is "info".
685  * (1) When at root level, info->prev is NULL, so it is "root" itself.
686  * (2) Otherwise, find the cache_tree that corresponds to one level
687  *     above us, and find ourselves in there.
688  */
689 static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root,
690                                                          struct traverse_info *info)
691 {
692         struct cache_tree *our_parent;
693
694         if (!info->prev)
695                 return root;
696         our_parent = find_cache_tree_from_traversal(root, info->prev);
697         return cache_tree_find(our_parent, info->name.path);
698 }
699
700 int cache_tree_matches_traversal(struct cache_tree *root,
701                                  struct name_entry *ent,
702                                  struct traverse_info *info)
703 {
704         struct cache_tree *it;
705
706         it = find_cache_tree_from_traversal(root, info);
707         it = cache_tree_find(it, ent->path);
708         if (it && it->entry_count > 0 && !hashcmp(ent->sha1, it->sha1))
709                 return it->entry_count;
710         return 0;
711 }
712
713 int update_main_cache_tree(int flags)
714 {
715         if (!the_index.cache_tree)
716                 the_index.cache_tree = cache_tree();
717         return cache_tree_update(&the_index, flags);
718 }