Merge branch 'jc/utf8-fprintf' into maint
[git] / cache-tree.c
1 #include "cache.h"
2 #include "lockfile.h"
3 #include "tree.h"
4 #include "tree-walk.h"
5 #include "cache-tree.h"
6
7 #ifndef DEBUG
8 #define DEBUG 0
9 #endif
10
11 struct cache_tree *cache_tree(void)
12 {
13         struct cache_tree *it = xcalloc(1, sizeof(struct cache_tree));
14         it->entry_count = -1;
15         return it;
16 }
17
18 void cache_tree_free(struct cache_tree **it_p)
19 {
20         int i;
21         struct cache_tree *it = *it_p;
22
23         if (!it)
24                 return;
25         for (i = 0; i < it->subtree_nr; i++)
26                 if (it->down[i]) {
27                         cache_tree_free(&it->down[i]->cache_tree);
28                         free(it->down[i]);
29                 }
30         free(it->down);
31         free(it);
32         *it_p = NULL;
33 }
34
35 static int subtree_name_cmp(const char *one, int onelen,
36                             const char *two, int twolen)
37 {
38         if (onelen < twolen)
39                 return -1;
40         if (twolen < onelen)
41                 return 1;
42         return memcmp(one, two, onelen);
43 }
44
45 static int subtree_pos(struct cache_tree *it, const char *path, int pathlen)
46 {
47         struct cache_tree_sub **down = it->down;
48         int lo, hi;
49         lo = 0;
50         hi = it->subtree_nr;
51         while (lo < hi) {
52                 int mi = (lo + hi) / 2;
53                 struct cache_tree_sub *mdl = down[mi];
54                 int cmp = subtree_name_cmp(path, pathlen,
55                                            mdl->name, mdl->namelen);
56                 if (!cmp)
57                         return mi;
58                 if (cmp < 0)
59                         hi = mi;
60                 else
61                         lo = mi + 1;
62         }
63         return -lo-1;
64 }
65
66 static struct cache_tree_sub *find_subtree(struct cache_tree *it,
67                                            const char *path,
68                                            int pathlen,
69                                            int create)
70 {
71         struct cache_tree_sub *down;
72         int pos = subtree_pos(it, path, pathlen);
73         if (0 <= pos)
74                 return it->down[pos];
75         if (!create)
76                 return NULL;
77
78         pos = -pos-1;
79         ALLOC_GROW(it->down, it->subtree_nr + 1, it->subtree_alloc);
80         it->subtree_nr++;
81
82         FLEX_ALLOC_MEM(down, name, path, pathlen);
83         down->cache_tree = NULL;
84         down->namelen = pathlen;
85
86         if (pos < it->subtree_nr)
87                 memmove(it->down + pos + 1,
88                         it->down + pos,
89                         sizeof(down) * (it->subtree_nr - pos - 1));
90         it->down[pos] = down;
91         return down;
92 }
93
94 struct cache_tree_sub *cache_tree_sub(struct cache_tree *it, const char *path)
95 {
96         int pathlen = strlen(path);
97         return find_subtree(it, path, pathlen, 1);
98 }
99
100 static int do_invalidate_path(struct cache_tree *it, const char *path)
101 {
102         /* a/b/c
103          * ==> invalidate self
104          * ==> find "a", have it invalidate "b/c"
105          * a
106          * ==> invalidate self
107          * ==> if "a" exists as a subtree, remove it.
108          */
109         const char *slash;
110         int namelen;
111         struct cache_tree_sub *down;
112
113 #if DEBUG
114         fprintf(stderr, "cache-tree invalidate <%s>\n", path);
115 #endif
116
117         if (!it)
118                 return 0;
119         slash = strchrnul(path, '/');
120         namelen = slash - path;
121         it->entry_count = -1;
122         if (!*slash) {
123                 int pos;
124                 pos = subtree_pos(it, path, namelen);
125                 if (0 <= pos) {
126                         cache_tree_free(&it->down[pos]->cache_tree);
127                         free(it->down[pos]);
128                         /* 0 1 2 3 4 5
129                          *       ^     ^subtree_nr = 6
130                          *       pos
131                          * move 4 and 5 up one place (2 entries)
132                          * 2 = 6 - 3 - 1 = subtree_nr - pos - 1
133                          */
134                         memmove(it->down+pos, it->down+pos+1,
135                                 sizeof(struct cache_tree_sub *) *
136                                 (it->subtree_nr - pos - 1));
137                         it->subtree_nr--;
138                 }
139                 return 1;
140         }
141         down = find_subtree(it, path, namelen, 0);
142         if (down)
143                 do_invalidate_path(down->cache_tree, slash + 1);
144         return 1;
145 }
146
147 void cache_tree_invalidate_path(struct index_state *istate, const char *path)
148 {
149         if (do_invalidate_path(istate->cache_tree, path))
150                 istate->cache_changed |= CACHE_TREE_CHANGED;
151 }
152
153 static int verify_cache(struct cache_entry **cache,
154                         int entries, int flags)
155 {
156         int i, funny;
157         int silent = flags & WRITE_TREE_SILENT;
158
159         /* Verify that the tree is merged */
160         funny = 0;
161         for (i = 0; i < entries; i++) {
162                 const struct cache_entry *ce = cache[i];
163                 if (ce_stage(ce)) {
164                         if (silent)
165                                 return -1;
166                         if (10 < ++funny) {
167                                 fprintf(stderr, "...\n");
168                                 break;
169                         }
170                         fprintf(stderr, "%s: unmerged (%s)\n",
171                                 ce->name, oid_to_hex(&ce->oid));
172                 }
173         }
174         if (funny)
175                 return -1;
176
177         /* Also verify that the cache does not have path and path/file
178          * at the same time.  At this point we know the cache has only
179          * stage 0 entries.
180          */
181         funny = 0;
182         for (i = 0; i < entries - 1; i++) {
183                 /* path/file always comes after path because of the way
184                  * the cache is sorted.  Also path can appear only once,
185                  * which means conflicting one would immediately follow.
186                  */
187                 const char *this_name = cache[i]->name;
188                 const char *next_name = cache[i+1]->name;
189                 int this_len = strlen(this_name);
190                 if (this_len < strlen(next_name) &&
191                     strncmp(this_name, next_name, this_len) == 0 &&
192                     next_name[this_len] == '/') {
193                         if (10 < ++funny) {
194                                 fprintf(stderr, "...\n");
195                                 break;
196                         }
197                         fprintf(stderr, "You have both %s and %s\n",
198                                 this_name, next_name);
199                 }
200         }
201         if (funny)
202                 return -1;
203         return 0;
204 }
205
206 static void discard_unused_subtrees(struct cache_tree *it)
207 {
208         struct cache_tree_sub **down = it->down;
209         int nr = it->subtree_nr;
210         int dst, src;
211         for (dst = src = 0; src < nr; src++) {
212                 struct cache_tree_sub *s = down[src];
213                 if (s->used)
214                         down[dst++] = s;
215                 else {
216                         cache_tree_free(&s->cache_tree);
217                         free(s);
218                         it->subtree_nr--;
219                 }
220         }
221 }
222
223 int cache_tree_fully_valid(struct cache_tree *it)
224 {
225         int i;
226         if (!it)
227                 return 0;
228         if (it->entry_count < 0 || !has_sha1_file(it->sha1))
229                 return 0;
230         for (i = 0; i < it->subtree_nr; i++) {
231                 if (!cache_tree_fully_valid(it->down[i]->cache_tree))
232                         return 0;
233         }
234         return 1;
235 }
236
237 static int update_one(struct cache_tree *it,
238                       struct cache_entry **cache,
239                       int entries,
240                       const char *base,
241                       int baselen,
242                       int *skip_count,
243                       int flags)
244 {
245         struct strbuf buffer;
246         int missing_ok = flags & WRITE_TREE_MISSING_OK;
247         int dryrun = flags & WRITE_TREE_DRY_RUN;
248         int repair = flags & WRITE_TREE_REPAIR;
249         int to_invalidate = 0;
250         int i;
251
252         assert(!(dryrun && repair));
253
254         *skip_count = 0;
255
256         if (0 <= it->entry_count && has_sha1_file(it->sha1))
257                 return it->entry_count;
258
259         /*
260          * We first scan for subtrees and update them; we start by
261          * marking existing subtrees -- the ones that are unmarked
262          * should not be in the result.
263          */
264         for (i = 0; i < it->subtree_nr; i++)
265                 it->down[i]->used = 0;
266
267         /*
268          * Find the subtrees and update them.
269          */
270         i = 0;
271         while (i < entries) {
272                 const struct cache_entry *ce = cache[i];
273                 struct cache_tree_sub *sub;
274                 const char *path, *slash;
275                 int pathlen, sublen, subcnt, subskip;
276
277                 path = ce->name;
278                 pathlen = ce_namelen(ce);
279                 if (pathlen <= baselen || memcmp(base, path, baselen))
280                         break; /* at the end of this level */
281
282                 slash = strchr(path + baselen, '/');
283                 if (!slash) {
284                         i++;
285                         continue;
286                 }
287                 /*
288                  * a/bbb/c (base = a/, slash = /c)
289                  * ==>
290                  * path+baselen = bbb/c, sublen = 3
291                  */
292                 sublen = slash - (path + baselen);
293                 sub = find_subtree(it, path + baselen, sublen, 1);
294                 if (!sub->cache_tree)
295                         sub->cache_tree = cache_tree();
296                 subcnt = update_one(sub->cache_tree,
297                                     cache + i, entries - i,
298                                     path,
299                                     baselen + sublen + 1,
300                                     &subskip,
301                                     flags);
302                 if (subcnt < 0)
303                         return subcnt;
304                 if (!subcnt)
305                         die("index cache-tree records empty sub-tree");
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 = NULL;
323                 const char *path, *slash;
324                 int pathlen, entlen;
325                 const unsigned char *sha1;
326                 unsigned mode;
327                 int expected_missing = 0;
328                 int contains_ita = 0;
329
330                 path = ce->name;
331                 pathlen = ce_namelen(ce);
332                 if (pathlen <= baselen || memcmp(base, path, baselen))
333                         break; /* at the end of this level */
334
335                 slash = strchr(path + baselen, '/');
336                 if (slash) {
337                         entlen = slash - (path + baselen);
338                         sub = find_subtree(it, path + baselen, entlen, 0);
339                         if (!sub)
340                                 die("cache-tree.c: '%.*s' in '%s' not found",
341                                     entlen, path + baselen, path);
342                         i += sub->count;
343                         sha1 = sub->cache_tree->sha1;
344                         mode = S_IFDIR;
345                         contains_ita = sub->cache_tree->entry_count < 0;
346                         if (contains_ita) {
347                                 to_invalidate = 1;
348                                 expected_missing = 1;
349                         }
350                 }
351                 else {
352                         sha1 = ce->oid.hash;
353                         mode = ce->ce_mode;
354                         entlen = pathlen - baselen;
355                         i++;
356                 }
357                 if (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1)) {
358                         strbuf_release(&buffer);
359                         if (expected_missing)
360                                 return -1;
361                         return error("invalid object %06o %s for '%.*s'",
362                                 mode, sha1_to_hex(sha1), entlen+baselen, path);
363                 }
364
365                 /*
366                  * CE_REMOVE entries are removed before the index is
367                  * written to disk. Skip them to remain consistent
368                  * with the future on-disk index.
369                  */
370                 if (ce->ce_flags & CE_REMOVE) {
371                         *skip_count = *skip_count + 1;
372                         continue;
373                 }
374
375                 /*
376                  * CE_INTENT_TO_ADD entries exist on on-disk index but
377                  * they are not part of generated trees. Invalidate up
378                  * to root to force cache-tree users to read elsewhere.
379                  */
380                 if (!sub && ce_intent_to_add(ce)) {
381                         to_invalidate = 1;
382                         continue;
383                 }
384
385                 /*
386                  * "sub" can be an empty tree if all subentries are i-t-a.
387                  */
388                 if (contains_ita && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN))
389                         continue;
390
391                 strbuf_grow(&buffer, entlen + 100);
392                 strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0');
393                 strbuf_add(&buffer, sha1, 20);
394
395 #if DEBUG
396                 fprintf(stderr, "cache-tree update-one %o %.*s\n",
397                         mode, entlen, path + baselen);
398 #endif
399         }
400
401         if (repair) {
402                 unsigned char sha1[20];
403                 hash_sha1_file(buffer.buf, buffer.len, tree_type, sha1);
404                 if (has_sha1_file(sha1))
405                         hashcpy(it->sha1, sha1);
406                 else
407                         to_invalidate = 1;
408         } else if (dryrun)
409                 hash_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1);
410         else if (write_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1)) {
411                 strbuf_release(&buffer);
412                 return -1;
413         }
414
415         strbuf_release(&buffer);
416         it->entry_count = to_invalidate ? -1 : i - *skip_count;
417 #if DEBUG
418         fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
419                 it->entry_count, it->subtree_nr,
420                 sha1_to_hex(it->sha1));
421 #endif
422         return i;
423 }
424
425 int cache_tree_update(struct index_state *istate, int flags)
426 {
427         struct cache_tree *it = istate->cache_tree;
428         struct cache_entry **cache = istate->cache;
429         int entries = istate->cache_nr;
430         int skip, i = verify_cache(cache, entries, flags);
431
432         if (i)
433                 return i;
434         i = update_one(it, cache, entries, "", 0, &skip, flags);
435         if (i < 0)
436                 return i;
437         istate->cache_changed |= CACHE_TREE_CHANGED;
438         return 0;
439 }
440
441 static void write_one(struct strbuf *buffer, struct cache_tree *it,
442                       const char *path, int pathlen)
443 {
444         int i;
445
446         /* One "cache-tree" entry consists of the following:
447          * path (NUL terminated)
448          * entry_count, subtree_nr ("%d %d\n")
449          * tree-sha1 (missing if invalid)
450          * subtree_nr "cache-tree" entries for subtrees.
451          */
452         strbuf_grow(buffer, pathlen + 100);
453         strbuf_add(buffer, path, pathlen);
454         strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr);
455
456 #if DEBUG
457         if (0 <= it->entry_count)
458                 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
459                         pathlen, path, it->entry_count, it->subtree_nr,
460                         sha1_to_hex(it->sha1));
461         else
462                 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
463                         pathlen, path, it->subtree_nr);
464 #endif
465
466         if (0 <= it->entry_count) {
467                 strbuf_add(buffer, it->sha1, 20);
468         }
469         for (i = 0; i < it->subtree_nr; i++) {
470                 struct cache_tree_sub *down = it->down[i];
471                 if (i) {
472                         struct cache_tree_sub *prev = it->down[i-1];
473                         if (subtree_name_cmp(down->name, down->namelen,
474                                              prev->name, prev->namelen) <= 0)
475                                 die("fatal - unsorted cache subtree");
476                 }
477                 write_one(buffer, down->cache_tree, down->name, down->namelen);
478         }
479 }
480
481 void cache_tree_write(struct strbuf *sb, struct cache_tree *root)
482 {
483         write_one(sb, root, "", 0);
484 }
485
486 static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
487 {
488         const char *buf = *buffer;
489         unsigned long size = *size_p;
490         const char *cp;
491         char *ep;
492         struct cache_tree *it;
493         int i, subtree_nr;
494
495         it = NULL;
496         /* skip name, but make sure name exists */
497         while (size && *buf) {
498                 size--;
499                 buf++;
500         }
501         if (!size)
502                 goto free_return;
503         buf++; size--;
504         it = cache_tree();
505
506         cp = buf;
507         it->entry_count = strtol(cp, &ep, 10);
508         if (cp == ep)
509                 goto free_return;
510         cp = ep;
511         subtree_nr = strtol(cp, &ep, 10);
512         if (cp == ep)
513                 goto free_return;
514         while (size && *buf && *buf != '\n') {
515                 size--;
516                 buf++;
517         }
518         if (!size)
519                 goto free_return;
520         buf++; size--;
521         if (0 <= it->entry_count) {
522                 if (size < 20)
523                         goto free_return;
524                 hashcpy(it->sha1, (const unsigned char*)buf);
525                 buf += 20;
526                 size -= 20;
527         }
528
529 #if DEBUG
530         if (0 <= it->entry_count)
531                 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
532                         *buffer, it->entry_count, subtree_nr,
533                         sha1_to_hex(it->sha1));
534         else
535                 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
536                         *buffer, subtree_nr);
537 #endif
538
539         /*
540          * Just a heuristic -- we do not add directories that often but
541          * we do not want to have to extend it immediately when we do,
542          * hence +2.
543          */
544         it->subtree_alloc = subtree_nr + 2;
545         it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
546         for (i = 0; i < subtree_nr; i++) {
547                 /* read each subtree */
548                 struct cache_tree *sub;
549                 struct cache_tree_sub *subtree;
550                 const char *name = buf;
551
552                 sub = read_one(&buf, &size);
553                 if (!sub)
554                         goto free_return;
555                 subtree = cache_tree_sub(it, name);
556                 subtree->cache_tree = sub;
557         }
558         if (subtree_nr != it->subtree_nr)
559                 die("cache-tree: internal error");
560         *buffer = buf;
561         *size_p = size;
562         return it;
563
564  free_return:
565         cache_tree_free(&it);
566         return NULL;
567 }
568
569 struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
570 {
571         if (buffer[0])
572                 return NULL; /* not the whole tree */
573         return read_one(&buffer, &size);
574 }
575
576 static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
577 {
578         if (!it)
579                 return NULL;
580         while (*path) {
581                 const char *slash;
582                 struct cache_tree_sub *sub;
583
584                 slash = strchrnul(path, '/');
585                 /*
586                  * Between path and slash is the name of the subtree
587                  * to look for.
588                  */
589                 sub = find_subtree(it, path, slash - path, 0);
590                 if (!sub)
591                         return NULL;
592                 it = sub->cache_tree;
593
594                 path = slash;
595                 while (*path == '/')
596                         path++;
597         }
598         return it;
599 }
600
601 int write_index_as_tree(unsigned char *sha1, struct index_state *index_state, const char *index_path, int flags, const char *prefix)
602 {
603         int entries, was_valid, newfd;
604         struct lock_file *lock_file;
605
606         /*
607          * We can't free this memory, it becomes part of a linked list
608          * parsed atexit()
609          */
610         lock_file = xcalloc(1, sizeof(struct lock_file));
611
612         newfd = hold_lock_file_for_update(lock_file, index_path, LOCK_DIE_ON_ERROR);
613
614         entries = read_index_from(index_state, index_path);
615         if (entries < 0)
616                 return WRITE_TREE_UNREADABLE_INDEX;
617         if (flags & WRITE_TREE_IGNORE_CACHE_TREE)
618                 cache_tree_free(&index_state->cache_tree);
619
620         if (!index_state->cache_tree)
621                 index_state->cache_tree = cache_tree();
622
623         was_valid = cache_tree_fully_valid(index_state->cache_tree);
624         if (!was_valid) {
625                 if (cache_tree_update(index_state, flags) < 0)
626                         return WRITE_TREE_UNMERGED_INDEX;
627                 if (0 <= newfd) {
628                         if (!write_locked_index(index_state, lock_file, COMMIT_LOCK))
629                                 newfd = -1;
630                 }
631                 /* Not being able to write is fine -- we are only interested
632                  * in updating the cache-tree part, and if the next caller
633                  * ends up using the old index with unupdated cache-tree part
634                  * it misses the work we did here, but that is just a
635                  * performance penalty and not a big deal.
636                  */
637         }
638
639         if (prefix) {
640                 struct cache_tree *subtree;
641                 subtree = cache_tree_find(index_state->cache_tree, prefix);
642                 if (!subtree)
643                         return WRITE_TREE_PREFIX_ERROR;
644                 hashcpy(sha1, subtree->sha1);
645         }
646         else
647                 hashcpy(sha1, index_state->cache_tree->sha1);
648
649         if (0 <= newfd)
650                 rollback_lock_file(lock_file);
651
652         return 0;
653 }
654
655 int write_cache_as_tree(unsigned char *sha1, int flags, const char *prefix)
656 {
657         return write_index_as_tree(sha1, &the_index, get_index_file(), flags, prefix);
658 }
659
660 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
661 {
662         struct tree_desc desc;
663         struct name_entry entry;
664         int cnt;
665
666         hashcpy(it->sha1, tree->object.oid.hash);
667         init_tree_desc(&desc, tree->buffer, tree->size);
668         cnt = 0;
669         while (tree_entry(&desc, &entry)) {
670                 if (!S_ISDIR(entry.mode))
671                         cnt++;
672                 else {
673                         struct cache_tree_sub *sub;
674                         struct tree *subtree = lookup_tree(entry.oid->hash);
675                         if (!subtree->object.parsed)
676                                 parse_tree(subtree);
677                         sub = cache_tree_sub(it, entry.path);
678                         sub->cache_tree = cache_tree();
679                         prime_cache_tree_rec(sub->cache_tree, subtree);
680                         cnt += sub->cache_tree->entry_count;
681                 }
682         }
683         it->entry_count = cnt;
684 }
685
686 void prime_cache_tree(struct index_state *istate, struct tree *tree)
687 {
688         cache_tree_free(&istate->cache_tree);
689         istate->cache_tree = cache_tree();
690         prime_cache_tree_rec(istate->cache_tree, tree);
691         istate->cache_changed |= CACHE_TREE_CHANGED;
692 }
693
694 /*
695  * find the cache_tree that corresponds to the current level without
696  * exploding the full path into textual form.  The root of the
697  * cache tree is given as "root", and our current level is "info".
698  * (1) When at root level, info->prev is NULL, so it is "root" itself.
699  * (2) Otherwise, find the cache_tree that corresponds to one level
700  *     above us, and find ourselves in there.
701  */
702 static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root,
703                                                          struct traverse_info *info)
704 {
705         struct cache_tree *our_parent;
706
707         if (!info->prev)
708                 return root;
709         our_parent = find_cache_tree_from_traversal(root, info->prev);
710         return cache_tree_find(our_parent, info->name.path);
711 }
712
713 int cache_tree_matches_traversal(struct cache_tree *root,
714                                  struct name_entry *ent,
715                                  struct traverse_info *info)
716 {
717         struct cache_tree *it;
718
719         it = find_cache_tree_from_traversal(root, info);
720         it = cache_tree_find(it, ent->path);
721         if (it && it->entry_count > 0 && !hashcmp(ent->oid->hash, it->sha1))
722                 return it->entry_count;
723         return 0;
724 }
725
726 int update_main_cache_tree(int flags)
727 {
728         if (!the_index.cache_tree)
729                 the_index.cache_tree = cache_tree();
730         return cache_tree_update(&the_index, flags);
731 }