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