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