git-cherry: document limit and add diagram
[git] / cache-tree.c
1 #include "cache.h"
2 #include "tree.h"
3 #include "cache-tree.h"
4
5 #define DEBUG 0
6
7 struct cache_tree *cache_tree(void)
8 {
9         struct cache_tree *it = xcalloc(1, sizeof(struct cache_tree));
10         it->entry_count = -1;
11         return it;
12 }
13
14 void cache_tree_free(struct cache_tree **it_p)
15 {
16         int i;
17         struct cache_tree *it = *it_p;
18
19         if (!it)
20                 return;
21         for (i = 0; i < it->subtree_nr; i++)
22                 if (it->down[i])
23                         cache_tree_free(&it->down[i]->cache_tree);
24         free(it->down);
25         free(it);
26         *it_p = NULL;
27 }
28
29 static int subtree_name_cmp(const char *one, int onelen,
30                             const char *two, int twolen)
31 {
32         if (onelen < twolen)
33                 return -1;
34         if (twolen < onelen)
35                 return 1;
36         return memcmp(one, two, onelen);
37 }
38
39 static int subtree_pos(struct cache_tree *it, const char *path, int pathlen)
40 {
41         struct cache_tree_sub **down = it->down;
42         int lo, hi;
43         lo = 0;
44         hi = it->subtree_nr;
45         while (lo < hi) {
46                 int mi = (lo + hi) / 2;
47                 struct cache_tree_sub *mdl = down[mi];
48                 int cmp = subtree_name_cmp(path, pathlen,
49                                            mdl->name, mdl->namelen);
50                 if (!cmp)
51                         return mi;
52                 if (cmp < 0)
53                         hi = mi;
54                 else
55                         lo = mi + 1;
56         }
57         return -lo-1;
58 }
59
60 static struct cache_tree_sub *find_subtree(struct cache_tree *it,
61                                            const char *path,
62                                            int pathlen,
63                                            int create)
64 {
65         struct cache_tree_sub *down;
66         int pos = subtree_pos(it, path, pathlen);
67         if (0 <= pos)
68                 return it->down[pos];
69         if (!create)
70                 return NULL;
71
72         pos = -pos-1;
73         if (it->subtree_alloc <= it->subtree_nr) {
74                 it->subtree_alloc = alloc_nr(it->subtree_alloc);
75                 it->down = xrealloc(it->down, it->subtree_alloc *
76                                     sizeof(*it->down));
77         }
78         it->subtree_nr++;
79
80         down = xmalloc(sizeof(*down) + pathlen + 1);
81         down->cache_tree = NULL;
82         down->namelen = pathlen;
83         memcpy(down->name, path, pathlen);
84         down->name[pathlen] = 0;
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 void cache_tree_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;
119         slash = strchr(path, '/');
120         it->entry_count = -1;
121         if (!slash) {
122                 int pos;
123                 namelen = strlen(path);
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;
140         }
141         namelen = slash - path;
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(struct cache_entry **cache,
148                         int entries)
149 {
150         int i, funny;
151
152         /* Verify that the tree is merged */
153         funny = 0;
154         for (i = 0; i < entries; i++) {
155                 struct cache_entry *ce = cache[i];
156                 if (ce_stage(ce)) {
157                         if (10 < ++funny) {
158                                 fprintf(stderr, "...\n");
159                                 break;
160                         }
161                         fprintf(stderr, "%s: unmerged (%s)\n",
162                                 ce->name, sha1_to_hex(ce->sha1));
163                 }
164         }
165         if (funny)
166                 return -1;
167
168         /* Also verify that the cache does not have path and path/file
169          * at the same time.  At this point we know the cache has only
170          * stage 0 entries.
171          */
172         funny = 0;
173         for (i = 0; i < entries - 1; i++) {
174                 /* path/file always comes after path because of the way
175                  * the cache is sorted.  Also path can appear only once,
176                  * which means conflicting one would immediately follow.
177                  */
178                 const char *this_name = cache[i]->name;
179                 const char *next_name = cache[i+1]->name;
180                 int this_len = strlen(this_name);
181                 if (this_len < strlen(next_name) &&
182                     strncmp(this_name, next_name, this_len) == 0 &&
183                     next_name[this_len] == '/') {
184                         if (10 < ++funny) {
185                                 fprintf(stderr, "...\n");
186                                 break;
187                         }
188                         fprintf(stderr, "You have both %s and %s\n",
189                                 this_name, next_name);
190                 }
191         }
192         if (funny)
193                 return -1;
194         return 0;
195 }
196
197 static void discard_unused_subtrees(struct cache_tree *it)
198 {
199         struct cache_tree_sub **down = it->down;
200         int nr = it->subtree_nr;
201         int dst, src;
202         for (dst = src = 0; src < nr; src++) {
203                 struct cache_tree_sub *s = down[src];
204                 if (s->used)
205                         down[dst++] = s;
206                 else {
207                         cache_tree_free(&s->cache_tree);
208                         free(s);
209                         it->subtree_nr--;
210                 }
211         }
212 }
213
214 int cache_tree_fully_valid(struct cache_tree *it)
215 {
216         int i;
217         if (!it)
218                 return 0;
219         if (it->entry_count < 0 || !has_sha1_file(it->sha1))
220                 return 0;
221         for (i = 0; i < it->subtree_nr; i++) {
222                 if (!cache_tree_fully_valid(it->down[i]->cache_tree))
223                         return 0;
224         }
225         return 1;
226 }
227
228 static int update_one(struct cache_tree *it,
229                       struct cache_entry **cache,
230                       int entries,
231                       const char *base,
232                       int baselen,
233                       int missing_ok,
234                       int dryrun)
235 {
236         unsigned long size, offset;
237         char *buffer;
238         int i;
239
240         if (0 <= it->entry_count && has_sha1_file(it->sha1))
241                 return it->entry_count;
242
243         /*
244          * We first scan for subtrees and update them; we start by
245          * marking existing subtrees -- the ones that are unmarked
246          * should not be in the result.
247          */
248         for (i = 0; i < it->subtree_nr; i++)
249                 it->down[i]->used = 0;
250
251         /*
252          * Find the subtrees and update them.
253          */
254         for (i = 0; i < entries; i++) {
255                 struct cache_entry *ce = cache[i];
256                 struct cache_tree_sub *sub;
257                 const char *path, *slash;
258                 int pathlen, sublen, subcnt;
259
260                 path = ce->name;
261                 pathlen = ce_namelen(ce);
262                 if (pathlen <= baselen || memcmp(base, path, baselen))
263                         break; /* at the end of this level */
264
265                 slash = strchr(path + baselen, '/');
266                 if (!slash)
267                         continue;
268                 /*
269                  * a/bbb/c (base = a/, slash = /c)
270                  * ==>
271                  * path+baselen = bbb/c, sublen = 3
272                  */
273                 sublen = slash - (path + baselen);
274                 sub = find_subtree(it, path + baselen, sublen, 1);
275                 if (!sub->cache_tree)
276                         sub->cache_tree = cache_tree();
277                 subcnt = update_one(sub->cache_tree,
278                                     cache + i, entries - i,
279                                     path,
280                                     baselen + sublen + 1,
281                                     missing_ok,
282                                     dryrun);
283                 i += subcnt - 1;
284                 sub->used = 1;
285         }
286
287         discard_unused_subtrees(it);
288
289         /*
290          * Then write out the tree object for this level.
291          */
292         size = 8192;
293         buffer = xmalloc(size);
294         offset = 0;
295
296         for (i = 0; i < entries; i++) {
297                 struct cache_entry *ce = cache[i];
298                 struct cache_tree_sub *sub;
299                 const char *path, *slash;
300                 int pathlen, entlen;
301                 const unsigned char *sha1;
302                 unsigned mode;
303
304                 path = ce->name;
305                 pathlen = ce_namelen(ce);
306                 if (pathlen <= baselen || memcmp(base, path, baselen))
307                         break; /* at the end of this level */
308
309                 slash = strchr(path + baselen, '/');
310                 if (slash) {
311                         entlen = slash - (path + baselen);
312                         sub = find_subtree(it, path + baselen, entlen, 0);
313                         if (!sub)
314                                 die("cache-tree.c: '%.*s' in '%s' not found",
315                                     entlen, path + baselen, path);
316                         i += sub->cache_tree->entry_count - 1;
317                         sha1 = sub->cache_tree->sha1;
318                         mode = S_IFDIR;
319                 }
320                 else {
321                         sha1 = ce->sha1;
322                         mode = ntohl(ce->ce_mode);
323                         entlen = pathlen - baselen;
324                 }
325                 if (!missing_ok && !has_sha1_file(sha1))
326                         return error("invalid object %s", sha1_to_hex(sha1));
327
328                 if (!ce->ce_mode)
329                         continue; /* entry being removed */
330
331                 if (size < offset + entlen + 100) {
332                         size = alloc_nr(offset + entlen + 100);
333                         buffer = xrealloc(buffer, size);
334                 }
335                 offset += sprintf(buffer + offset,
336                                   "%o %.*s", mode, entlen, path + baselen);
337                 buffer[offset++] = 0;
338                 hashcpy((unsigned char*)buffer + offset, sha1);
339                 offset += 20;
340
341 #if DEBUG
342                 fprintf(stderr, "cache-tree update-one %o %.*s\n",
343                         mode, entlen, path + baselen);
344 #endif
345         }
346
347         if (dryrun)
348                 hash_sha1_file(buffer, offset, tree_type, it->sha1);
349         else
350                 write_sha1_file(buffer, offset, tree_type, it->sha1);
351         free(buffer);
352         it->entry_count = i;
353 #if DEBUG
354         fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
355                 it->entry_count, it->subtree_nr,
356                 sha1_to_hex(it->sha1));
357 #endif
358         return i;
359 }
360
361 int cache_tree_update(struct cache_tree *it,
362                       struct cache_entry **cache,
363                       int entries,
364                       int missing_ok,
365                       int dryrun)
366 {
367         int i;
368         i = verify_cache(cache, entries);
369         if (i)
370                 return i;
371         i = update_one(it, cache, entries, "", 0, missing_ok, dryrun);
372         if (i < 0)
373                 return i;
374         return 0;
375 }
376
377 static void *write_one(struct cache_tree *it,
378                        char *path,
379                        int pathlen,
380                        char *buffer,
381                        unsigned long *size,
382                        unsigned long *offset)
383 {
384         int i;
385
386         /* One "cache-tree" entry consists of the following:
387          * path (NUL terminated)
388          * entry_count, subtree_nr ("%d %d\n")
389          * tree-sha1 (missing if invalid)
390          * subtree_nr "cache-tree" entries for subtrees.
391          */
392         if (*size < *offset + pathlen + 100) {
393                 *size = alloc_nr(*offset + pathlen + 100);
394                 buffer = xrealloc(buffer, *size);
395         }
396         *offset += sprintf(buffer + *offset, "%.*s%c%d %d\n",
397                            pathlen, path, 0,
398                            it->entry_count, it->subtree_nr);
399
400 #if DEBUG
401         if (0 <= it->entry_count)
402                 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
403                         pathlen, path, it->entry_count, it->subtree_nr,
404                         sha1_to_hex(it->sha1));
405         else
406                 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
407                         pathlen, path, it->subtree_nr);
408 #endif
409
410         if (0 <= it->entry_count) {
411                 hashcpy((unsigned char*)buffer + *offset, it->sha1);
412                 *offset += 20;
413         }
414         for (i = 0; i < it->subtree_nr; i++) {
415                 struct cache_tree_sub *down = it->down[i];
416                 if (i) {
417                         struct cache_tree_sub *prev = it->down[i-1];
418                         if (subtree_name_cmp(down->name, down->namelen,
419                                              prev->name, prev->namelen) <= 0)
420                                 die("fatal - unsorted cache subtree");
421                 }
422                 buffer = write_one(down->cache_tree, down->name, down->namelen,
423                                    buffer, size, offset);
424         }
425         return buffer;
426 }
427
428 void *cache_tree_write(struct cache_tree *root, unsigned long *size_p)
429 {
430         char path[PATH_MAX];
431         unsigned long size = 8192;
432         char *buffer = xmalloc(size);
433
434         *size_p = 0;
435         path[0] = 0;
436         return write_one(root, path, 0, buffer, &size, size_p);
437 }
438
439 static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
440 {
441         const char *buf = *buffer;
442         unsigned long size = *size_p;
443         const char *cp;
444         char *ep;
445         struct cache_tree *it;
446         int i, subtree_nr;
447
448         it = NULL;
449         /* skip name, but make sure name exists */
450         while (size && *buf) {
451                 size--;
452                 buf++;
453         }
454         if (!size)
455                 goto free_return;
456         buf++; size--;
457         it = cache_tree();
458
459         cp = buf;
460         it->entry_count = strtol(cp, &ep, 10);
461         if (cp == ep)
462                 goto free_return;
463         cp = ep;
464         subtree_nr = strtol(cp, &ep, 10);
465         if (cp == ep)
466                 goto free_return;
467         while (size && *buf && *buf != '\n') {
468                 size--;
469                 buf++;
470         }
471         if (!size)
472                 goto free_return;
473         buf++; size--;
474         if (0 <= it->entry_count) {
475                 if (size < 20)
476                         goto free_return;
477                 hashcpy(it->sha1, (unsigned char*)buf);
478                 buf += 20;
479                 size -= 20;
480         }
481
482 #if DEBUG
483         if (0 <= it->entry_count)
484                 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
485                         *buffer, it->entry_count, subtree_nr,
486                         sha1_to_hex(it->sha1));
487         else
488                 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
489                         *buffer, subtree_nr);
490 #endif
491
492         /*
493          * Just a heuristic -- we do not add directories that often but
494          * we do not want to have to extend it immediately when we do,
495          * hence +2.
496          */
497         it->subtree_alloc = subtree_nr + 2;
498         it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
499         for (i = 0; i < subtree_nr; i++) {
500                 /* read each subtree */
501                 struct cache_tree *sub;
502                 struct cache_tree_sub *subtree;
503                 const char *name = buf;
504
505                 sub = read_one(&buf, &size);
506                 if (!sub)
507                         goto free_return;
508                 subtree = cache_tree_sub(it, name);
509                 subtree->cache_tree = sub;
510         }
511         if (subtree_nr != it->subtree_nr)
512                 die("cache-tree: internal error");
513         *buffer = buf;
514         *size_p = size;
515         return it;
516
517  free_return:
518         cache_tree_free(&it);
519         return NULL;
520 }
521
522 struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
523 {
524         if (buffer[0])
525                 return NULL; /* not the whole tree */
526         return read_one(&buffer, &size);
527 }
528
529 struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
530 {
531         while (*path) {
532                 const char *slash;
533                 struct cache_tree_sub *sub;
534
535                 slash = strchr(path, '/');
536                 if (!slash)
537                         slash = path + strlen(path);
538                 /* between path and slash is the name of the
539                  * subtree to look for.
540                  */
541                 sub = find_subtree(it, path, slash - path, 0);
542                 if (!sub)
543                         return NULL;
544                 it = sub->cache_tree;
545                 if (slash)
546                         while (*slash && *slash == '/')
547                                 slash++;
548                 if (!slash || !*slash)
549                         return it; /* prefix ended with slashes */
550                 path = slash;
551         }
552         return it;
553 }