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