add a howto document about corrupted blob recovery
[git/debian.git] / sha1_name.c
blob2d727d54dc43c23cb2fe19369d58e57e2a020ca9
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "tree-walk.h"
7 #include "refs.h"
9 static int find_short_object_filename(int len, const char *name, unsigned char *sha1)
11 struct alternate_object_database *alt;
12 char hex[40];
13 int found = 0;
14 static struct alternate_object_database *fakeent;
16 if (!fakeent) {
17 const char *objdir = get_object_directory();
18 int objdir_len = strlen(objdir);
19 int entlen = objdir_len + 43;
20 fakeent = xmalloc(sizeof(*fakeent) + entlen);
21 memcpy(fakeent->base, objdir, objdir_len);
22 fakeent->name = fakeent->base + objdir_len + 1;
23 fakeent->name[-1] = '/';
25 fakeent->next = alt_odb_list;
27 sprintf(hex, "%.2s", name);
28 for (alt = fakeent; alt && found < 2; alt = alt->next) {
29 struct dirent *de;
30 DIR *dir;
31 sprintf(alt->name, "%.2s/", name);
32 dir = opendir(alt->base);
33 if (!dir)
34 continue;
35 while ((de = readdir(dir)) != NULL) {
36 if (strlen(de->d_name) != 38)
37 continue;
38 if (memcmp(de->d_name, name + 2, len - 2))
39 continue;
40 if (!found) {
41 memcpy(hex + 2, de->d_name, 38);
42 found++;
44 else if (memcmp(hex + 2, de->d_name, 38)) {
45 found = 2;
46 break;
49 closedir(dir);
51 if (found == 1)
52 return get_sha1_hex(hex, sha1) == 0;
53 return found;
56 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
58 do {
59 if (*a != *b)
60 return 0;
61 a++;
62 b++;
63 len -= 2;
64 } while (len > 1);
65 if (len)
66 if ((*a ^ *b) & 0xf0)
67 return 0;
68 return 1;
71 static int find_short_packed_object(int len, const unsigned char *match, unsigned char *sha1)
73 struct packed_git *p;
74 const unsigned char *found_sha1 = NULL;
75 int found = 0;
77 prepare_packed_git();
78 for (p = packed_git; p && found < 2; p = p->next) {
79 uint32_t num, last;
80 uint32_t first = 0;
81 open_pack_index(p);
82 num = p->num_objects;
83 last = num;
84 while (first < last) {
85 uint32_t mid = (first + last) / 2;
86 const unsigned char *now;
87 int cmp;
89 now = nth_packed_object_sha1(p, mid);
90 cmp = hashcmp(match, now);
91 if (!cmp) {
92 first = mid;
93 break;
95 if (cmp > 0) {
96 first = mid+1;
97 continue;
99 last = mid;
101 if (first < num) {
102 const unsigned char *now, *next;
103 now = nth_packed_object_sha1(p, first);
104 if (match_sha(len, match, now)) {
105 next = nth_packed_object_sha1(p, first+1);
106 if (!next|| !match_sha(len, match, next)) {
107 /* unique within this pack */
108 if (!found) {
109 found_sha1 = now;
110 found++;
112 else if (hashcmp(found_sha1, now)) {
113 found = 2;
114 break;
117 else {
118 /* not even unique within this pack */
119 found = 2;
120 break;
125 if (found == 1)
126 hashcpy(sha1, found_sha1);
127 return found;
130 #define SHORT_NAME_NOT_FOUND (-1)
131 #define SHORT_NAME_AMBIGUOUS (-2)
133 static int find_unique_short_object(int len, char *canonical,
134 unsigned char *res, unsigned char *sha1)
136 int has_unpacked, has_packed;
137 unsigned char unpacked_sha1[20], packed_sha1[20];
139 prepare_alt_odb();
140 has_unpacked = find_short_object_filename(len, canonical, unpacked_sha1);
141 has_packed = find_short_packed_object(len, res, packed_sha1);
142 if (!has_unpacked && !has_packed)
143 return SHORT_NAME_NOT_FOUND;
144 if (1 < has_unpacked || 1 < has_packed)
145 return SHORT_NAME_AMBIGUOUS;
146 if (has_unpacked != has_packed) {
147 hashcpy(sha1, (has_packed ? packed_sha1 : unpacked_sha1));
148 return 0;
150 /* Both have unique ones -- do they match? */
151 if (hashcmp(packed_sha1, unpacked_sha1))
152 return SHORT_NAME_AMBIGUOUS;
153 hashcpy(sha1, packed_sha1);
154 return 0;
157 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
158 int quietly)
160 int i, status;
161 char canonical[40];
162 unsigned char res[20];
164 if (len < MINIMUM_ABBREV || len > 40)
165 return -1;
166 hashclr(res);
167 memset(canonical, 'x', 40);
168 for (i = 0; i < len ;i++) {
169 unsigned char c = name[i];
170 unsigned char val;
171 if (c >= '0' && c <= '9')
172 val = c - '0';
173 else if (c >= 'a' && c <= 'f')
174 val = c - 'a' + 10;
175 else if (c >= 'A' && c <='F') {
176 val = c - 'A' + 10;
177 c -= 'A' - 'a';
179 else
180 return -1;
181 canonical[i] = c;
182 if (!(i & 1))
183 val <<= 4;
184 res[i >> 1] |= val;
187 status = find_unique_short_object(i, canonical, res, sha1);
188 if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
189 return error("short SHA1 %.*s is ambiguous.", len, canonical);
190 return status;
193 const char *find_unique_abbrev(const unsigned char *sha1, int len)
195 int status, is_null;
196 static char hex[41];
198 is_null = is_null_sha1(sha1);
199 memcpy(hex, sha1_to_hex(sha1), 40);
200 if (len == 40 || !len)
201 return hex;
202 while (len < 40) {
203 unsigned char sha1_ret[20];
204 status = get_short_sha1(hex, len, sha1_ret, 1);
205 if (!status ||
206 (is_null && status != SHORT_NAME_AMBIGUOUS)) {
207 hex[len] = 0;
208 return hex;
210 if (status != SHORT_NAME_AMBIGUOUS)
211 return NULL;
212 len++;
214 return NULL;
217 static int ambiguous_path(const char *path, int len)
219 int slash = 1;
220 int cnt;
222 for (cnt = 0; cnt < len; cnt++) {
223 switch (*path++) {
224 case '\0':
225 break;
226 case '/':
227 if (slash)
228 break;
229 slash = 1;
230 continue;
231 case '.':
232 continue;
233 default:
234 slash = 0;
235 continue;
237 break;
239 return slash;
242 static const char *ref_fmt[] = {
243 "%.*s",
244 "refs/%.*s",
245 "refs/tags/%.*s",
246 "refs/heads/%.*s",
247 "refs/remotes/%.*s",
248 "refs/remotes/%.*s/HEAD",
249 NULL
252 int dwim_ref(const char *str, int len, unsigned char *sha1, char **ref)
254 const char **p, *r;
255 int refs_found = 0;
257 *ref = NULL;
258 for (p = ref_fmt; *p; p++) {
259 unsigned char sha1_from_ref[20];
260 unsigned char *this_result;
262 this_result = refs_found ? sha1_from_ref : sha1;
263 r = resolve_ref(mkpath(*p, len, str), this_result, 1, NULL);
264 if (r) {
265 if (!refs_found++)
266 *ref = xstrdup(r);
267 if (!warn_ambiguous_refs)
268 break;
271 return refs_found;
274 int dwim_log(const char *str, int len, unsigned char *sha1, char **log)
276 const char **p;
277 int logs_found = 0;
279 *log = NULL;
280 for (p = ref_fmt; *p; p++) {
281 struct stat st;
282 unsigned char hash[20];
283 char path[PATH_MAX];
284 const char *ref, *it;
286 strcpy(path, mkpath(*p, len, str));
287 ref = resolve_ref(path, hash, 0, NULL);
288 if (!ref)
289 continue;
290 if (!stat(git_path("logs/%s", path), &st) &&
291 S_ISREG(st.st_mode))
292 it = path;
293 else if (strcmp(ref, path) &&
294 !stat(git_path("logs/%s", ref), &st) &&
295 S_ISREG(st.st_mode))
296 it = ref;
297 else
298 continue;
299 if (!logs_found++) {
300 *log = xstrdup(it);
301 hashcpy(sha1, hash);
303 if (!warn_ambiguous_refs)
304 break;
306 return logs_found;
309 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
311 static const char *warning = "warning: refname '%.*s' is ambiguous.\n";
312 char *real_ref = NULL;
313 int refs_found = 0;
314 int at, reflog_len;
316 if (len == 40 && !get_sha1_hex(str, sha1))
317 return 0;
319 /* basic@{time or number} format to query ref-log */
320 reflog_len = at = 0;
321 if (str[len-1] == '}') {
322 for (at = 0; at < len - 1; at++) {
323 if (str[at] == '@' && str[at+1] == '{') {
324 reflog_len = (len-1) - (at+2);
325 len = at;
326 break;
331 /* Accept only unambiguous ref paths. */
332 if (len && ambiguous_path(str, len))
333 return -1;
335 if (!len && reflog_len) {
336 /* allow "@{...}" to mean the current branch reflog */
337 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
338 } else if (reflog_len)
339 refs_found = dwim_log(str, len, sha1, &real_ref);
340 else
341 refs_found = dwim_ref(str, len, sha1, &real_ref);
343 if (!refs_found)
344 return -1;
346 if (warn_ambiguous_refs && refs_found > 1)
347 fprintf(stderr, warning, len, str);
349 if (reflog_len) {
350 int nth, i;
351 unsigned long at_time;
352 unsigned long co_time;
353 int co_tz, co_cnt;
355 /* Is it asking for N-th entry, or approxidate? */
356 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
357 char ch = str[at+2+i];
358 if ('0' <= ch && ch <= '9')
359 nth = nth * 10 + ch - '0';
360 else
361 nth = -1;
363 if (0 <= nth)
364 at_time = 0;
365 else
366 at_time = approxidate(str + at + 2);
367 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
368 &co_time, &co_tz, &co_cnt)) {
369 if (at_time)
370 fprintf(stderr,
371 "warning: Log for '%.*s' only goes "
372 "back to %s.\n", len, str,
373 show_date(co_time, co_tz, DATE_RFC2822));
374 else
375 fprintf(stderr,
376 "warning: Log for '%.*s' only has "
377 "%d entries.\n", len, str, co_cnt);
381 free(real_ref);
382 return 0;
385 static int get_sha1_1(const char *name, int len, unsigned char *sha1);
387 static int get_parent(const char *name, int len,
388 unsigned char *result, int idx)
390 unsigned char sha1[20];
391 int ret = get_sha1_1(name, len, sha1);
392 struct commit *commit;
393 struct commit_list *p;
395 if (ret)
396 return ret;
397 commit = lookup_commit_reference(sha1);
398 if (!commit)
399 return -1;
400 if (parse_commit(commit))
401 return -1;
402 if (!idx) {
403 hashcpy(result, commit->object.sha1);
404 return 0;
406 p = commit->parents;
407 while (p) {
408 if (!--idx) {
409 hashcpy(result, p->item->object.sha1);
410 return 0;
412 p = p->next;
414 return -1;
417 static int get_nth_ancestor(const char *name, int len,
418 unsigned char *result, int generation)
420 unsigned char sha1[20];
421 int ret = get_sha1_1(name, len, sha1);
422 if (ret)
423 return ret;
425 while (generation--) {
426 struct commit *commit = lookup_commit_reference(sha1);
428 if (!commit || parse_commit(commit) || !commit->parents)
429 return -1;
430 hashcpy(sha1, commit->parents->item->object.sha1);
432 hashcpy(result, sha1);
433 return 0;
436 static int peel_onion(const char *name, int len, unsigned char *sha1)
438 unsigned char outer[20];
439 const char *sp;
440 unsigned int expected_type = 0;
441 struct object *o;
444 * "ref^{type}" dereferences ref repeatedly until you cannot
445 * dereference anymore, or you get an object of given type,
446 * whichever comes first. "ref^{}" means just dereference
447 * tags until you get a non-tag. "ref^0" is a shorthand for
448 * "ref^{commit}". "commit^{tree}" could be used to find the
449 * top-level tree of the given commit.
451 if (len < 4 || name[len-1] != '}')
452 return -1;
454 for (sp = name + len - 1; name <= sp; sp--) {
455 int ch = *sp;
456 if (ch == '{' && name < sp && sp[-1] == '^')
457 break;
459 if (sp <= name)
460 return -1;
462 sp++; /* beginning of type name, or closing brace for empty */
463 if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
464 expected_type = OBJ_COMMIT;
465 else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
466 expected_type = OBJ_TREE;
467 else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
468 expected_type = OBJ_BLOB;
469 else if (sp[0] == '}')
470 expected_type = OBJ_NONE;
471 else
472 return -1;
474 if (get_sha1_1(name, sp - name - 2, outer))
475 return -1;
477 o = parse_object(outer);
478 if (!o)
479 return -1;
480 if (!expected_type) {
481 o = deref_tag(o, name, sp - name - 2);
482 if (!o || (!o->parsed && !parse_object(o->sha1)))
483 return -1;
484 hashcpy(sha1, o->sha1);
486 else {
487 /* At this point, the syntax look correct, so
488 * if we do not get the needed object, we should
489 * barf.
492 while (1) {
493 if (!o || (!o->parsed && !parse_object(o->sha1)))
494 return -1;
495 if (o->type == expected_type) {
496 hashcpy(sha1, o->sha1);
497 return 0;
499 if (o->type == OBJ_TAG)
500 o = ((struct tag*) o)->tagged;
501 else if (o->type == OBJ_COMMIT)
502 o = &(((struct commit *) o)->tree->object);
503 else
504 return error("%.*s: expected %s type, but the object dereferences to %s type",
505 len, name, typename(expected_type),
506 typename(o->type));
507 if (!o->parsed)
508 parse_object(o->sha1);
511 return 0;
514 static int get_describe_name(const char *name, int len, unsigned char *sha1)
516 const char *cp;
518 for (cp = name + len - 1; name + 2 <= cp; cp--) {
519 char ch = *cp;
520 if (hexval(ch) & ~0377) {
521 /* We must be looking at g in "SOMETHING-g"
522 * for it to be describe output.
524 if (ch == 'g' && cp[-1] == '-') {
525 cp++;
526 len -= cp - name;
527 return get_short_sha1(cp, len, sha1, 1);
531 return -1;
534 static int get_sha1_1(const char *name, int len, unsigned char *sha1)
536 int ret, has_suffix;
537 const char *cp;
539 /* "name~3" is "name^^^",
540 * "name~" and "name~0" are name -- not "name^0"!
541 * "name^" is not "name^0"; it is "name^1".
543 has_suffix = 0;
544 for (cp = name + len - 1; name <= cp; cp--) {
545 int ch = *cp;
546 if ('0' <= ch && ch <= '9')
547 continue;
548 if (ch == '~' || ch == '^')
549 has_suffix = ch;
550 break;
553 if (has_suffix) {
554 int num = 0;
555 int len1 = cp - name;
556 cp++;
557 while (cp < name + len)
558 num = num * 10 + *cp++ - '0';
559 if (has_suffix == '^') {
560 if (!num && len1 == len - 1)
561 num = 1;
562 return get_parent(name, len1, sha1, num);
564 /* else if (has_suffix == '~') -- goes without saying */
565 return get_nth_ancestor(name, len1, sha1, num);
568 ret = peel_onion(name, len, sha1);
569 if (!ret)
570 return 0;
572 ret = get_sha1_basic(name, len, sha1);
573 if (!ret)
574 return 0;
576 /* It could be describe output that is "SOMETHING-gXXXX" */
577 ret = get_describe_name(name, len, sha1);
578 if (!ret)
579 return 0;
581 return get_short_sha1(name, len, sha1, 0);
584 static int handle_one_ref(const char *path,
585 const unsigned char *sha1, int flag, void *cb_data)
587 struct commit_list **list = cb_data;
588 struct object *object = parse_object(sha1);
589 if (!object)
590 return 0;
591 if (object->type == OBJ_TAG)
592 object = deref_tag(object, path, strlen(path));
593 if (object->type != OBJ_COMMIT)
594 return 0;
595 insert_by_date((struct commit *)object, list);
596 return 0;
600 * This interprets names like ':/Initial revision of "git"' by searching
601 * through history and returning the first commit whose message starts
602 * with the given string.
604 * For future extension, ':/!' is reserved. If you want to match a message
605 * beginning with a '!', you have to repeat the exclamation mark.
608 #define ONELINE_SEEN (1u<<20)
609 static int get_sha1_oneline(const char *prefix, unsigned char *sha1)
611 struct commit_list *list = NULL, *backup = NULL, *l;
612 int retval = -1;
614 if (prefix[0] == '!') {
615 if (prefix[1] != '!')
616 die ("Invalid search pattern: %s", prefix);
617 prefix++;
619 if (!save_commit_buffer)
620 return error("Could not expand oneline-name.");
621 for_each_ref(handle_one_ref, &list);
622 for (l = list; l; l = l->next)
623 commit_list_insert(l->item, &backup);
624 while (list) {
625 char *p;
626 struct commit *commit;
628 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
629 parse_object(commit->object.sha1);
630 if (!commit->buffer || !(p = strstr(commit->buffer, "\n\n")))
631 continue;
632 if (!prefixcmp(p + 2, prefix)) {
633 hashcpy(sha1, commit->object.sha1);
634 retval = 0;
635 break;
638 free_commit_list(list);
639 for (l = backup; l; l = l->next)
640 clear_commit_marks(l->item, ONELINE_SEEN);
641 return retval;
645 * This is like "get_sha1_basic()", except it allows "sha1 expressions",
646 * notably "xyz^" for "parent of xyz"
648 int get_sha1(const char *name, unsigned char *sha1)
650 unsigned unused;
651 return get_sha1_with_mode(name, sha1, &unused);
654 int get_sha1_with_mode(const char *name, unsigned char *sha1, unsigned *mode)
656 int ret, bracket_depth;
657 int namelen = strlen(name);
658 const char *cp;
660 *mode = S_IFINVALID;
661 ret = get_sha1_1(name, namelen, sha1);
662 if (!ret)
663 return ret;
664 /* sha1:path --> object name of path in ent sha1
665 * :path -> object name of path in index
666 * :[0-3]:path -> object name of path in index at stage
668 if (name[0] == ':') {
669 int stage = 0;
670 struct cache_entry *ce;
671 int pos;
672 if (namelen > 2 && name[1] == '/')
673 return get_sha1_oneline(name + 2, sha1);
674 if (namelen < 3 ||
675 name[2] != ':' ||
676 name[1] < '0' || '3' < name[1])
677 cp = name + 1;
678 else {
679 stage = name[1] - '0';
680 cp = name + 3;
682 namelen = namelen - (cp - name);
683 if (!active_cache)
684 read_cache();
685 pos = cache_name_pos(cp, namelen);
686 if (pos < 0)
687 pos = -pos - 1;
688 while (pos < active_nr) {
689 ce = active_cache[pos];
690 if (ce_namelen(ce) != namelen ||
691 memcmp(ce->name, cp, namelen))
692 break;
693 if (ce_stage(ce) == stage) {
694 hashcpy(sha1, ce->sha1);
695 *mode = ntohl(ce->ce_mode);
696 return 0;
698 pos++;
700 return -1;
702 for (cp = name, bracket_depth = 0; *cp; cp++) {
703 if (*cp == '{')
704 bracket_depth++;
705 else if (bracket_depth && *cp == '}')
706 bracket_depth--;
707 else if (!bracket_depth && *cp == ':')
708 break;
710 if (*cp == ':') {
711 unsigned char tree_sha1[20];
712 if (!get_sha1_1(name, cp-name, tree_sha1))
713 return get_tree_entry(tree_sha1, cp+1, sha1,
714 mode);
716 return ret;