Merge branch 'jc/repack-threads'
[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
358                 if (is_null_sha1(sha1) ||
359                     (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1))) {
360                         strbuf_release(&buffer);
361                         if (expected_missing)
362                                 return -1;
363                         return error("invalid object %06o %s for '%.*s'",
364                                 mode, sha1_to_hex(sha1), entlen+baselen, path);
365                 }
366
367                 /*
368                  * CE_REMOVE entries are removed before the index is
369                  * written to disk. Skip them to remain consistent
370                  * with the future on-disk index.
371                  */
372                 if (ce->ce_flags & CE_REMOVE) {
373                         *skip_count = *skip_count + 1;
374                         continue;
375                 }
376
377                 /*
378                  * CE_INTENT_TO_ADD entries exist on on-disk index but
379                  * they are not part of generated trees. Invalidate up
380                  * to root to force cache-tree users to read elsewhere.
381                  */
382                 if (!sub && ce_intent_to_add(ce)) {
383                         to_invalidate = 1;
384                         continue;
385                 }
386
387                 /*
388                  * "sub" can be an empty tree if all subentries are i-t-a.
389                  */
390                 if (contains_ita && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN))
391                         continue;
392
393                 strbuf_grow(&buffer, entlen + 100);
394                 strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0');
395                 strbuf_add(&buffer, sha1, 20);
396
397 #if DEBUG
398                 fprintf(stderr, "cache-tree update-one %o %.*s\n",
399                         mode, entlen, path + baselen);
400 #endif
401         }
402
403         if (repair) {
404                 unsigned char sha1[20];
405                 hash_sha1_file(buffer.buf, buffer.len, tree_type, sha1);
406                 if (has_sha1_file(sha1))
407                         hashcpy(it->sha1, sha1);
408                 else
409                         to_invalidate = 1;
410         } else if (dryrun)
411                 hash_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1);
412         else if (write_sha1_file(buffer.buf, buffer.len, tree_type, it->sha1)) {
413                 strbuf_release(&buffer);
414                 return -1;
415         }
416
417         strbuf_release(&buffer);
418         it->entry_count = to_invalidate ? -1 : i - *skip_count;
419 #if DEBUG
420         fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
421                 it->entry_count, it->subtree_nr,
422                 sha1_to_hex(it->sha1));
423 #endif
424         return i;
425 }
426
427 int cache_tree_update(struct index_state *istate, int flags)
428 {
429         struct cache_tree *it = istate->cache_tree;
430         struct cache_entry **cache = istate->cache;
431         int entries = istate->cache_nr;
432         int skip, i = verify_cache(cache, entries, flags);
433
434         if (i)
435                 return i;
436         i = update_one(it, cache, entries, "", 0, &skip, flags);
437         if (i < 0)
438                 return i;
439         istate->cache_changed |= CACHE_TREE_CHANGED;
440         return 0;
441 }
442
443 static void write_one(struct strbuf *buffer, struct cache_tree *it,
444                       const char *path, int pathlen)
445 {
446         int i;
447
448         /* One "cache-tree" entry consists of the following:
449          * path (NUL terminated)
450          * entry_count, subtree_nr ("%d %d\n")
451          * tree-sha1 (missing if invalid)
452          * subtree_nr "cache-tree" entries for subtrees.
453          */
454         strbuf_grow(buffer, pathlen + 100);
455         strbuf_add(buffer, path, pathlen);
456         strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr);
457
458 #if DEBUG
459         if (0 <= it->entry_count)
460                 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
461                         pathlen, path, it->entry_count, it->subtree_nr,
462                         sha1_to_hex(it->sha1));
463         else
464                 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
465                         pathlen, path, it->subtree_nr);
466 #endif
467
468         if (0 <= it->entry_count) {
469                 strbuf_add(buffer, it->sha1, 20);
470         }
471         for (i = 0; i < it->subtree_nr; i++) {
472                 struct cache_tree_sub *down = it->down[i];
473                 if (i) {
474                         struct cache_tree_sub *prev = it->down[i-1];
475                         if (subtree_name_cmp(down->name, down->namelen,
476                                              prev->name, prev->namelen) <= 0)
477                                 die("fatal - unsorted cache subtree");
478                 }
479                 write_one(buffer, down->cache_tree, down->name, down->namelen);
480         }
481 }
482
483 void cache_tree_write(struct strbuf *sb, struct cache_tree *root)
484 {
485         write_one(sb, root, "", 0);
486 }
487
488 static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
489 {
490         const char *buf = *buffer;
491         unsigned long size = *size_p;
492         const char *cp;
493         char *ep;
494         struct cache_tree *it;
495         int i, subtree_nr;
496
497         it = NULL;
498         /* skip name, but make sure name exists */
499         while (size && *buf) {
500                 size--;
501                 buf++;
502         }
503         if (!size)
504                 goto free_return;
505         buf++; size--;
506         it = cache_tree();
507
508         cp = buf;
509         it->entry_count = strtol(cp, &ep, 10);
510         if (cp == ep)
511                 goto free_return;
512         cp = ep;
513         subtree_nr = strtol(cp, &ep, 10);
514         if (cp == ep)
515                 goto free_return;
516         while (size && *buf && *buf != '\n') {
517                 size--;
518                 buf++;
519         }
520         if (!size)
521                 goto free_return;
522         buf++; size--;
523         if (0 <= it->entry_count) {
524                 if (size < 20)
525                         goto free_return;
526                 hashcpy(it->sha1, (const unsigned char*)buf);
527                 buf += 20;
528                 size -= 20;
529         }
530
531 #if DEBUG
532         if (0 <= it->entry_count)
533                 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
534                         *buffer, it->entry_count, subtree_nr,
535                         sha1_to_hex(it->sha1));
536         else
537                 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
538                         *buffer, subtree_nr);
539 #endif
540
541         /*
542          * Just a heuristic -- we do not add directories that often but
543          * we do not want to have to extend it immediately when we do,
544          * hence +2.
545          */
546         it->subtree_alloc = subtree_nr + 2;
547         it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
548         for (i = 0; i < subtree_nr; i++) {
549                 /* read each subtree */
550                 struct cache_tree *sub;
551                 struct cache_tree_sub *subtree;
552                 const char *name = buf;
553
554                 sub = read_one(&buf, &size);
555                 if (!sub)
556                         goto free_return;
557                 subtree = cache_tree_sub(it, name);
558                 subtree->cache_tree = sub;
559         }
560         if (subtree_nr != it->subtree_nr)
561                 die("cache-tree: internal error");
562         *buffer = buf;
563         *size_p = size;
564         return it;
565
566  free_return:
567         cache_tree_free(&it);
568         return NULL;
569 }
570
571 struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
572 {
573         if (buffer[0])
574                 return NULL; /* not the whole tree */
575         return read_one(&buffer, &size);
576 }
577
578 static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
579 {
580         if (!it)
581                 return NULL;
582         while (*path) {
583                 const char *slash;
584                 struct cache_tree_sub *sub;
585
586                 slash = strchrnul(path, '/');
587                 /*
588                  * Between path and slash is the name of the subtree
589                  * to look for.
590                  */
591                 sub = find_subtree(it, path, slash - path, 0);
592                 if (!sub)
593                         return NULL;
594                 it = sub->cache_tree;
595
596                 path = slash;
597                 while (*path == '/')
598                         path++;
599         }
600         return it;
601 }
602
603 int write_index_as_tree(unsigned char *sha1, struct index_state *index_state, const char *index_path, int flags, const char *prefix)
604 {
605         int entries, was_valid, newfd;
606         struct lock_file *lock_file;
607
608         /*
609          * We can't free this memory, it becomes part of a linked list
610          * parsed atexit()
611          */
612         lock_file = xcalloc(1, sizeof(struct lock_file));
613
614         newfd = hold_lock_file_for_update(lock_file, index_path, LOCK_DIE_ON_ERROR);
615
616         entries = read_index_from(index_state, index_path);
617         if (entries < 0)
618                 return WRITE_TREE_UNREADABLE_INDEX;
619         if (flags & WRITE_TREE_IGNORE_CACHE_TREE)
620                 cache_tree_free(&index_state->cache_tree);
621
622         if (!index_state->cache_tree)
623                 index_state->cache_tree = cache_tree();
624
625         was_valid = cache_tree_fully_valid(index_state->cache_tree);
626         if (!was_valid) {
627                 if (cache_tree_update(index_state, flags) < 0)
628                         return WRITE_TREE_UNMERGED_INDEX;
629                 if (0 <= newfd) {
630                         if (!write_locked_index(index_state, lock_file, COMMIT_LOCK))
631                                 newfd = -1;
632                 }
633                 /* Not being able to write is fine -- we are only interested
634                  * in updating the cache-tree part, and if the next caller
635                  * ends up using the old index with unupdated cache-tree part
636                  * it misses the work we did here, but that is just a
637                  * performance penalty and not a big deal.
638                  */
639         }
640
641         if (prefix) {
642                 struct cache_tree *subtree;
643                 subtree = cache_tree_find(index_state->cache_tree, prefix);
644                 if (!subtree)
645                         return WRITE_TREE_PREFIX_ERROR;
646                 hashcpy(sha1, subtree->sha1);
647         }
648         else
649                 hashcpy(sha1, index_state->cache_tree->sha1);
650
651         if (0 <= newfd)
652                 rollback_lock_file(lock_file);
653
654         return 0;
655 }
656
657 int write_cache_as_tree(unsigned char *sha1, int flags, const char *prefix)
658 {
659         return write_index_as_tree(sha1, &the_index, get_index_file(), flags, prefix);
660 }
661
662 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
663 {
664         struct tree_desc desc;
665         struct name_entry entry;
666         int cnt;
667
668         hashcpy(it->sha1, tree->object.oid.hash);
669         init_tree_desc(&desc, tree->buffer, tree->size);
670         cnt = 0;
671         while (tree_entry(&desc, &entry)) {
672                 if (!S_ISDIR(entry.mode))
673                         cnt++;
674                 else {
675                         struct cache_tree_sub *sub;
676                         struct tree *subtree = lookup_tree(entry.oid->hash);
677                         if (!subtree->object.parsed)
678                                 parse_tree(subtree);
679                         sub = cache_tree_sub(it, entry.path);
680                         sub->cache_tree = cache_tree();
681                         prime_cache_tree_rec(sub->cache_tree, subtree);
682                         cnt += sub->cache_tree->entry_count;
683                 }
684         }
685         it->entry_count = cnt;
686 }
687
688 void prime_cache_tree(struct index_state *istate, struct tree *tree)
689 {
690         cache_tree_free(&istate->cache_tree);
691         istate->cache_tree = cache_tree();
692         prime_cache_tree_rec(istate->cache_tree, tree);
693         istate->cache_changed |= CACHE_TREE_CHANGED;
694 }
695
696 /*
697  * find the cache_tree that corresponds to the current level without
698  * exploding the full path into textual form.  The root of the
699  * cache tree is given as "root", and our current level is "info".
700  * (1) When at root level, info->prev is NULL, so it is "root" itself.
701  * (2) Otherwise, find the cache_tree that corresponds to one level
702  *     above us, and find ourselves in there.
703  */
704 static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root,
705                                                          struct traverse_info *info)
706 {
707         struct cache_tree *our_parent;
708
709         if (!info->prev)
710                 return root;
711         our_parent = find_cache_tree_from_traversal(root, info->prev);
712         return cache_tree_find(our_parent, info->name.path);
713 }
714
715 int cache_tree_matches_traversal(struct cache_tree *root,
716                                  struct name_entry *ent,
717                                  struct traverse_info *info)
718 {
719         struct cache_tree *it;
720
721         it = find_cache_tree_from_traversal(root, info);
722         it = cache_tree_find(it, ent->path);
723         if (it && it->entry_count > 0 && !hashcmp(ent->oid->hash, it->sha1))
724                 return it->entry_count;
725         return 0;
726 }
727
728 int update_main_cache_tree(int flags)
729 {
730         if (!the_index.cache_tree)
731                 the_index.cache_tree = cache_tree();
732         return cache_tree_update(&the_index, flags);
733 }