Make hash-object more robust against malformed objects
[git] / sha1_name.c
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"
8 #include "remote.h"
9
10 static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
11
12 static int find_short_object_filename(int len, const char *name, unsigned char *sha1)
13 {
14         struct alternate_object_database *alt;
15         char hex[40];
16         int found = 0;
17         static struct alternate_object_database *fakeent;
18
19         if (!fakeent) {
20                 const char *objdir = get_object_directory();
21                 int objdir_len = strlen(objdir);
22                 int entlen = objdir_len + 43;
23                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
24                 memcpy(fakeent->base, objdir, objdir_len);
25                 fakeent->name = fakeent->base + objdir_len + 1;
26                 fakeent->name[-1] = '/';
27         }
28         fakeent->next = alt_odb_list;
29
30         sprintf(hex, "%.2s", name);
31         for (alt = fakeent; alt && found < 2; alt = alt->next) {
32                 struct dirent *de;
33                 DIR *dir;
34                 sprintf(alt->name, "%.2s/", name);
35                 dir = opendir(alt->base);
36                 if (!dir)
37                         continue;
38                 while ((de = readdir(dir)) != NULL) {
39                         if (strlen(de->d_name) != 38)
40                                 continue;
41                         if (memcmp(de->d_name, name + 2, len - 2))
42                                 continue;
43                         if (!found) {
44                                 memcpy(hex + 2, de->d_name, 38);
45                                 found++;
46                         }
47                         else if (memcmp(hex + 2, de->d_name, 38)) {
48                                 found = 2;
49                                 break;
50                         }
51                 }
52                 closedir(dir);
53         }
54         if (found == 1)
55                 return get_sha1_hex(hex, sha1) == 0;
56         return found;
57 }
58
59 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
60 {
61         do {
62                 if (*a != *b)
63                         return 0;
64                 a++;
65                 b++;
66                 len -= 2;
67         } while (len > 1);
68         if (len)
69                 if ((*a ^ *b) & 0xf0)
70                         return 0;
71         return 1;
72 }
73
74 static int find_short_packed_object(int len, const unsigned char *match, unsigned char *sha1)
75 {
76         struct packed_git *p;
77         const unsigned char *found_sha1 = NULL;
78         int found = 0;
79
80         prepare_packed_git();
81         for (p = packed_git; p && found < 2; p = p->next) {
82                 uint32_t num, last;
83                 uint32_t first = 0;
84                 open_pack_index(p);
85                 num = p->num_objects;
86                 last = num;
87                 while (first < last) {
88                         uint32_t mid = (first + last) / 2;
89                         const unsigned char *now;
90                         int cmp;
91
92                         now = nth_packed_object_sha1(p, mid);
93                         cmp = hashcmp(match, now);
94                         if (!cmp) {
95                                 first = mid;
96                                 break;
97                         }
98                         if (cmp > 0) {
99                                 first = mid+1;
100                                 continue;
101                         }
102                         last = mid;
103                 }
104                 if (first < num) {
105                         const unsigned char *now, *next;
106                        now = nth_packed_object_sha1(p, first);
107                         if (match_sha(len, match, now)) {
108                                 next = nth_packed_object_sha1(p, first+1);
109                                if (!next|| !match_sha(len, match, next)) {
110                                         /* unique within this pack */
111                                         if (!found) {
112                                                 found_sha1 = now;
113                                                 found++;
114                                         }
115                                         else if (hashcmp(found_sha1, now)) {
116                                                 found = 2;
117                                                 break;
118                                         }
119                                 }
120                                 else {
121                                         /* not even unique within this pack */
122                                         found = 2;
123                                         break;
124                                 }
125                         }
126                 }
127         }
128         if (found == 1)
129                 hashcpy(sha1, found_sha1);
130         return found;
131 }
132
133 #define SHORT_NAME_NOT_FOUND (-1)
134 #define SHORT_NAME_AMBIGUOUS (-2)
135
136 static int find_unique_short_object(int len, char *canonical,
137                                     unsigned char *res, unsigned char *sha1)
138 {
139         int has_unpacked, has_packed;
140         unsigned char unpacked_sha1[20], packed_sha1[20];
141
142         prepare_alt_odb();
143         has_unpacked = find_short_object_filename(len, canonical, unpacked_sha1);
144         has_packed = find_short_packed_object(len, res, packed_sha1);
145         if (!has_unpacked && !has_packed)
146                 return SHORT_NAME_NOT_FOUND;
147         if (1 < has_unpacked || 1 < has_packed)
148                 return SHORT_NAME_AMBIGUOUS;
149         if (has_unpacked != has_packed) {
150                 hashcpy(sha1, (has_packed ? packed_sha1 : unpacked_sha1));
151                 return 0;
152         }
153         /* Both have unique ones -- do they match? */
154         if (hashcmp(packed_sha1, unpacked_sha1))
155                 return SHORT_NAME_AMBIGUOUS;
156         hashcpy(sha1, packed_sha1);
157         return 0;
158 }
159
160 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
161                           int quietly)
162 {
163         int i, status;
164         char canonical[40];
165         unsigned char res[20];
166
167         if (len < MINIMUM_ABBREV || len > 40)
168                 return -1;
169         hashclr(res);
170         memset(canonical, 'x', 40);
171         for (i = 0; i < len ;i++) {
172                 unsigned char c = name[i];
173                 unsigned char val;
174                 if (c >= '0' && c <= '9')
175                         val = c - '0';
176                 else if (c >= 'a' && c <= 'f')
177                         val = c - 'a' + 10;
178                 else if (c >= 'A' && c <='F') {
179                         val = c - 'A' + 10;
180                         c -= 'A' - 'a';
181                 }
182                 else
183                         return -1;
184                 canonical[i] = c;
185                 if (!(i & 1))
186                         val <<= 4;
187                 res[i >> 1] |= val;
188         }
189
190         status = find_unique_short_object(i, canonical, res, sha1);
191         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
192                 return error("short SHA1 %.*s is ambiguous.", len, canonical);
193         return status;
194 }
195
196 const char *find_unique_abbrev(const unsigned char *sha1, int len)
197 {
198         int status, exists;
199         static char hex[41];
200
201         exists = has_sha1_file(sha1);
202         memcpy(hex, sha1_to_hex(sha1), 40);
203         if (len == 40 || !len)
204                 return hex;
205         while (len < 40) {
206                 unsigned char sha1_ret[20];
207                 status = get_short_sha1(hex, len, sha1_ret, 1);
208                 if (exists
209                     ? !status
210                     : status == SHORT_NAME_NOT_FOUND) {
211                         int cut_at = len + unique_abbrev_extra_length;
212                         cut_at = (cut_at < 40) ? cut_at : 40;
213                         hex[cut_at] = 0;
214                         return hex;
215                 }
216                 len++;
217         }
218         return hex;
219 }
220
221 static int ambiguous_path(const char *path, int len)
222 {
223         int slash = 1;
224         int cnt;
225
226         for (cnt = 0; cnt < len; cnt++) {
227                 switch (*path++) {
228                 case '\0':
229                         break;
230                 case '/':
231                         if (slash)
232                                 break;
233                         slash = 1;
234                         continue;
235                 case '.':
236                         continue;
237                 default:
238                         slash = 0;
239                         continue;
240                 }
241                 break;
242         }
243         return slash;
244 }
245
246 /*
247  * *string and *len will only be substituted, and *string returned (for
248  * later free()ing) if the string passed in is a magic short-hand form
249  * to name a branch.
250  */
251 static char *substitute_branch_name(const char **string, int *len)
252 {
253         struct strbuf buf = STRBUF_INIT;
254         int ret = interpret_branch_name(*string, &buf);
255
256         if (ret == *len) {
257                 size_t size;
258                 *string = strbuf_detach(&buf, &size);
259                 *len = size;
260                 return (char *)*string;
261         }
262
263         return NULL;
264 }
265
266 int dwim_ref(const char *str, int len, unsigned char *sha1, char **ref)
267 {
268         char *last_branch = substitute_branch_name(&str, &len);
269         const char **p, *r;
270         int refs_found = 0;
271
272         *ref = NULL;
273         for (p = ref_rev_parse_rules; *p; p++) {
274                 char fullref[PATH_MAX];
275                 unsigned char sha1_from_ref[20];
276                 unsigned char *this_result;
277                 int flag;
278
279                 this_result = refs_found ? sha1_from_ref : sha1;
280                 mksnpath(fullref, sizeof(fullref), *p, len, str);
281                 r = resolve_ref(fullref, this_result, 1, &flag);
282                 if (r) {
283                         if (!refs_found++)
284                                 *ref = xstrdup(r);
285                         if (!warn_ambiguous_refs)
286                                 break;
287                 } else if ((flag & REF_ISSYMREF) && strcmp(fullref, "HEAD"))
288                         warning("ignoring dangling symref %s.", fullref);
289         }
290         free(last_branch);
291         return refs_found;
292 }
293
294 int dwim_log(const char *str, int len, unsigned char *sha1, char **log)
295 {
296         char *last_branch = substitute_branch_name(&str, &len);
297         const char **p;
298         int logs_found = 0;
299
300         *log = NULL;
301         for (p = ref_rev_parse_rules; *p; p++) {
302                 struct stat st;
303                 unsigned char hash[20];
304                 char path[PATH_MAX];
305                 const char *ref, *it;
306
307                 mksnpath(path, sizeof(path), *p, len, str);
308                 ref = resolve_ref(path, hash, 1, NULL);
309                 if (!ref)
310                         continue;
311                 if (!stat(git_path("logs/%s", path), &st) &&
312                     S_ISREG(st.st_mode))
313                         it = path;
314                 else if (strcmp(ref, path) &&
315                          !stat(git_path("logs/%s", ref), &st) &&
316                          S_ISREG(st.st_mode))
317                         it = ref;
318                 else
319                         continue;
320                 if (!logs_found++) {
321                         *log = xstrdup(it);
322                         hashcpy(sha1, hash);
323                 }
324                 if (!warn_ambiguous_refs)
325                         break;
326         }
327         free(last_branch);
328         return logs_found;
329 }
330
331 static inline int upstream_mark(const char *string, int len)
332 {
333         const char *suffix[] = { "@{upstream}", "@{u}" };
334         int i;
335
336         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
337                 int suffix_len = strlen(suffix[i]);
338                 if (suffix_len <= len
339                     && !memcmp(string, suffix[i], suffix_len))
340                         return suffix_len;
341         }
342         return 0;
343 }
344
345 static int get_sha1_1(const char *name, int len, unsigned char *sha1);
346
347 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
348 {
349         static const char *warn_msg = "refname '%.*s' is ambiguous.";
350         char *real_ref = NULL;
351         int refs_found = 0;
352         int at, reflog_len;
353
354         if (len == 40 && !get_sha1_hex(str, sha1))
355                 return 0;
356
357         /* basic@{time or number or -number} format to query ref-log */
358         reflog_len = at = 0;
359         if (len && str[len-1] == '}') {
360                 for (at = len-2; at >= 0; at--) {
361                         if (str[at] == '@' && str[at+1] == '{') {
362                                 if (!upstream_mark(str + at, len - at)) {
363                                         reflog_len = (len-1) - (at+2);
364                                         len = at;
365                                 }
366                                 break;
367                         }
368                 }
369         }
370
371         /* Accept only unambiguous ref paths. */
372         if (len && ambiguous_path(str, len))
373                 return -1;
374
375         if (!len && reflog_len) {
376                 struct strbuf buf = STRBUF_INIT;
377                 int ret;
378                 /* try the @{-N} syntax for n-th checkout */
379                 ret = interpret_branch_name(str+at, &buf);
380                 if (ret > 0) {
381                         /* substitute this branch name and restart */
382                         return get_sha1_1(buf.buf, buf.len, sha1);
383                 } else if (ret == 0) {
384                         return -1;
385                 }
386                 /* allow "@{...}" to mean the current branch reflog */
387                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
388         } else if (reflog_len)
389                 refs_found = dwim_log(str, len, sha1, &real_ref);
390         else
391                 refs_found = dwim_ref(str, len, sha1, &real_ref);
392
393         if (!refs_found)
394                 return -1;
395
396         if (warn_ambiguous_refs && refs_found > 1)
397                 warning(warn_msg, len, str);
398
399         if (reflog_len) {
400                 int nth, i;
401                 unsigned long at_time;
402                 unsigned long co_time;
403                 int co_tz, co_cnt;
404
405                 /* a @{-N} placed anywhere except the start is an error */
406                 if (str[at+2] == '-')
407                         return -1;
408
409                 /* Is it asking for N-th entry, or approxidate? */
410                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
411                         char ch = str[at+2+i];
412                         if ('0' <= ch && ch <= '9')
413                                 nth = nth * 10 + ch - '0';
414                         else
415                                 nth = -1;
416                 }
417                 if (100000000 <= nth) {
418                         at_time = nth;
419                         nth = -1;
420                 } else if (0 <= nth)
421                         at_time = 0;
422                 else {
423                         int errors = 0;
424                         char *tmp = xstrndup(str + at + 2, reflog_len);
425                         at_time = approxidate_careful(tmp, &errors);
426                         free(tmp);
427                         if (errors)
428                                 return -1;
429                 }
430                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
431                                 &co_time, &co_tz, &co_cnt)) {
432                         if (at_time)
433                                 warning("Log for '%.*s' only goes "
434                                         "back to %s.", len, str,
435                                         show_date(co_time, co_tz, DATE_RFC2822));
436                         else {
437                                 free(real_ref);
438                                 die("Log for '%.*s' only has %d entries.",
439                                     len, str, co_cnt);
440                         }
441                 }
442         }
443
444         free(real_ref);
445         return 0;
446 }
447
448 static int get_parent(const char *name, int len,
449                       unsigned char *result, int idx)
450 {
451         unsigned char sha1[20];
452         int ret = get_sha1_1(name, len, sha1);
453         struct commit *commit;
454         struct commit_list *p;
455
456         if (ret)
457                 return ret;
458         commit = lookup_commit_reference(sha1);
459         if (!commit)
460                 return -1;
461         if (parse_commit(commit))
462                 return -1;
463         if (!idx) {
464                 hashcpy(result, commit->object.sha1);
465                 return 0;
466         }
467         p = commit->parents;
468         while (p) {
469                 if (!--idx) {
470                         hashcpy(result, p->item->object.sha1);
471                         return 0;
472                 }
473                 p = p->next;
474         }
475         return -1;
476 }
477
478 static int get_nth_ancestor(const char *name, int len,
479                             unsigned char *result, int generation)
480 {
481         unsigned char sha1[20];
482         struct commit *commit;
483         int ret;
484
485         ret = get_sha1_1(name, len, sha1);
486         if (ret)
487                 return ret;
488         commit = lookup_commit_reference(sha1);
489         if (!commit)
490                 return -1;
491
492         while (generation--) {
493                 if (parse_commit(commit) || !commit->parents)
494                         return -1;
495                 commit = commit->parents->item;
496         }
497         hashcpy(result, commit->object.sha1);
498         return 0;
499 }
500
501 struct object *peel_to_type(const char *name, int namelen,
502                             struct object *o, enum object_type expected_type)
503 {
504         if (name && !namelen)
505                 namelen = strlen(name);
506         if (!o) {
507                 unsigned char sha1[20];
508                 if (get_sha1_1(name, namelen, sha1))
509                         return NULL;
510                 o = parse_object(sha1);
511         }
512         while (1) {
513                 if (!o || (!o->parsed && !parse_object(o->sha1)))
514                         return NULL;
515                 if (o->type == expected_type)
516                         return o;
517                 if (o->type == OBJ_TAG)
518                         o = ((struct tag*) o)->tagged;
519                 else if (o->type == OBJ_COMMIT)
520                         o = &(((struct commit *) o)->tree->object);
521                 else {
522                         if (name)
523                                 error("%.*s: expected %s type, but the object "
524                                       "dereferences to %s type",
525                                       namelen, name, typename(expected_type),
526                                       typename(o->type));
527                         return NULL;
528                 }
529         }
530 }
531
532 static int peel_onion(const char *name, int len, unsigned char *sha1)
533 {
534         unsigned char outer[20];
535         const char *sp;
536         unsigned int expected_type = 0;
537         struct object *o;
538
539         /*
540          * "ref^{type}" dereferences ref repeatedly until you cannot
541          * dereference anymore, or you get an object of given type,
542          * whichever comes first.  "ref^{}" means just dereference
543          * tags until you get a non-tag.  "ref^0" is a shorthand for
544          * "ref^{commit}".  "commit^{tree}" could be used to find the
545          * top-level tree of the given commit.
546          */
547         if (len < 4 || name[len-1] != '}')
548                 return -1;
549
550         for (sp = name + len - 1; name <= sp; sp--) {
551                 int ch = *sp;
552                 if (ch == '{' && name < sp && sp[-1] == '^')
553                         break;
554         }
555         if (sp <= name)
556                 return -1;
557
558         sp++; /* beginning of type name, or closing brace for empty */
559         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
560                 expected_type = OBJ_COMMIT;
561         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
562                 expected_type = OBJ_TREE;
563         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
564                 expected_type = OBJ_BLOB;
565         else if (sp[0] == '}')
566                 expected_type = OBJ_NONE;
567         else if (sp[0] == '/')
568                 expected_type = OBJ_COMMIT;
569         else
570                 return -1;
571
572         if (get_sha1_1(name, sp - name - 2, outer))
573                 return -1;
574
575         o = parse_object(outer);
576         if (!o)
577                 return -1;
578         if (!expected_type) {
579                 o = deref_tag(o, name, sp - name - 2);
580                 if (!o || (!o->parsed && !parse_object(o->sha1)))
581                         return -1;
582                 hashcpy(sha1, o->sha1);
583                 return 0;
584         }
585
586         /*
587          * At this point, the syntax look correct, so
588          * if we do not get the needed object, we should
589          * barf.
590          */
591         o = peel_to_type(name, len, o, expected_type);
592         if (!o)
593                 return -1;
594
595         hashcpy(sha1, o->sha1);
596         if (sp[0] == '/') {
597                 /* "$commit^{/foo}" */
598                 char *prefix;
599                 int ret;
600                 struct commit_list *list = NULL;
601
602                 /*
603                  * $commit^{/}. Some regex implementation may reject.
604                  * We don't need regex anyway. '' pattern always matches.
605                  */
606                 if (sp[1] == '}')
607                         return 0;
608
609                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
610                 commit_list_insert((struct commit *)o, &list);
611                 ret = get_sha1_oneline(prefix, sha1, list);
612                 free(prefix);
613                 return ret;
614         }
615         return 0;
616 }
617
618 static int get_describe_name(const char *name, int len, unsigned char *sha1)
619 {
620         const char *cp;
621
622         for (cp = name + len - 1; name + 2 <= cp; cp--) {
623                 char ch = *cp;
624                 if (hexval(ch) & ~0377) {
625                         /* We must be looking at g in "SOMETHING-g"
626                          * for it to be describe output.
627                          */
628                         if (ch == 'g' && cp[-1] == '-') {
629                                 cp++;
630                                 len -= cp - name;
631                                 return get_short_sha1(cp, len, sha1, 1);
632                         }
633                 }
634         }
635         return -1;
636 }
637
638 static int get_sha1_1(const char *name, int len, unsigned char *sha1)
639 {
640         int ret, has_suffix;
641         const char *cp;
642
643         /*
644          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
645          */
646         has_suffix = 0;
647         for (cp = name + len - 1; name <= cp; cp--) {
648                 int ch = *cp;
649                 if ('0' <= ch && ch <= '9')
650                         continue;
651                 if (ch == '~' || ch == '^')
652                         has_suffix = ch;
653                 break;
654         }
655
656         if (has_suffix) {
657                 int num = 0;
658                 int len1 = cp - name;
659                 cp++;
660                 while (cp < name + len)
661                         num = num * 10 + *cp++ - '0';
662                 if (!num && len1 == len - 1)
663                         num = 1;
664                 if (has_suffix == '^')
665                         return get_parent(name, len1, sha1, num);
666                 /* else if (has_suffix == '~') -- goes without saying */
667                 return get_nth_ancestor(name, len1, sha1, num);
668         }
669
670         ret = peel_onion(name, len, sha1);
671         if (!ret)
672                 return 0;
673
674         ret = get_sha1_basic(name, len, sha1);
675         if (!ret)
676                 return 0;
677
678         /* It could be describe output that is "SOMETHING-gXXXX" */
679         ret = get_describe_name(name, len, sha1);
680         if (!ret)
681                 return 0;
682
683         return get_short_sha1(name, len, sha1, 0);
684 }
685
686 /*
687  * This interprets names like ':/Initial revision of "git"' by searching
688  * through history and returning the first commit whose message starts
689  * the given regular expression.
690  *
691  * For future extension, ':/!' is reserved. If you want to match a message
692  * beginning with a '!', you have to repeat the exclamation mark.
693  */
694 #define ONELINE_SEEN (1u<<20)
695
696 static int handle_one_ref(const char *path,
697                 const unsigned char *sha1, int flag, void *cb_data)
698 {
699         struct commit_list **list = cb_data;
700         struct object *object = parse_object(sha1);
701         if (!object)
702                 return 0;
703         if (object->type == OBJ_TAG) {
704                 object = deref_tag(object, path, strlen(path));
705                 if (!object)
706                         return 0;
707         }
708         if (object->type != OBJ_COMMIT)
709                 return 0;
710         commit_list_insert_by_date((struct commit *)object, list);
711         return 0;
712 }
713
714 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
715                             struct commit_list *list)
716 {
717         struct commit_list *backup = NULL, *l;
718         int found = 0;
719         regex_t regex;
720
721         if (prefix[0] == '!') {
722                 if (prefix[1] != '!')
723                         die ("Invalid search pattern: %s", prefix);
724                 prefix++;
725         }
726
727         if (regcomp(&regex, prefix, REG_EXTENDED))
728                 die("Invalid search pattern: %s", prefix);
729
730         for (l = list; l; l = l->next) {
731                 l->item->object.flags |= ONELINE_SEEN;
732                 commit_list_insert(l->item, &backup);
733         }
734         while (list) {
735                 char *p, *to_free = NULL;
736                 struct commit *commit;
737                 enum object_type type;
738                 unsigned long size;
739                 int matches;
740
741                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
742                 if (!parse_object(commit->object.sha1))
743                         continue;
744                 if (commit->buffer)
745                         p = commit->buffer;
746                 else {
747                         p = read_sha1_file(commit->object.sha1, &type, &size);
748                         if (!p)
749                                 continue;
750                         to_free = p;
751                 }
752
753                 p = strstr(p, "\n\n");
754                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
755                 free(to_free);
756
757                 if (matches) {
758                         hashcpy(sha1, commit->object.sha1);
759                         found = 1;
760                         break;
761                 }
762         }
763         regfree(&regex);
764         free_commit_list(list);
765         for (l = backup; l; l = l->next)
766                 clear_commit_marks(l->item, ONELINE_SEEN);
767         free_commit_list(backup);
768         return found ? 0 : -1;
769 }
770
771 struct grab_nth_branch_switch_cbdata {
772         long cnt, alloc;
773         struct strbuf *buf;
774 };
775
776 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
777                                   const char *email, unsigned long timestamp, int tz,
778                                   const char *message, void *cb_data)
779 {
780         struct grab_nth_branch_switch_cbdata *cb = cb_data;
781         const char *match = NULL, *target = NULL;
782         size_t len;
783         int nth;
784
785         if (!prefixcmp(message, "checkout: moving from ")) {
786                 match = message + strlen("checkout: moving from ");
787                 target = strstr(match, " to ");
788         }
789
790         if (!match || !target)
791                 return 0;
792
793         len = target - match;
794         nth = cb->cnt++ % cb->alloc;
795         strbuf_reset(&cb->buf[nth]);
796         strbuf_add(&cb->buf[nth], match, len);
797         return 0;
798 }
799
800 /*
801  * Parse @{-N} syntax, return the number of characters parsed
802  * if successful; otherwise signal an error with negative value.
803  */
804 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
805 {
806         long nth;
807         int i, retval;
808         struct grab_nth_branch_switch_cbdata cb;
809         const char *brace;
810         char *num_end;
811
812         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
813                 return -1;
814         brace = strchr(name, '}');
815         if (!brace)
816                 return -1;
817         nth = strtol(name+3, &num_end, 10);
818         if (num_end != brace)
819                 return -1;
820         if (nth <= 0)
821                 return -1;
822         cb.alloc = nth;
823         cb.buf = xmalloc(nth * sizeof(struct strbuf));
824         for (i = 0; i < nth; i++)
825                 strbuf_init(&cb.buf[i], 20);
826         cb.cnt = 0;
827         retval = 0;
828         for_each_recent_reflog_ent("HEAD", grab_nth_branch_switch, 40960, &cb);
829         if (cb.cnt < nth) {
830                 cb.cnt = 0;
831                 for_each_reflog_ent("HEAD", grab_nth_branch_switch, &cb);
832         }
833         if (cb.cnt < nth)
834                 goto release_return;
835         i = cb.cnt % nth;
836         strbuf_reset(buf);
837         strbuf_add(buf, cb.buf[i].buf, cb.buf[i].len);
838         retval = brace-name+1;
839
840 release_return:
841         for (i = 0; i < nth; i++)
842                 strbuf_release(&cb.buf[i]);
843         free(cb.buf);
844
845         return retval;
846 }
847
848 int get_sha1_mb(const char *name, unsigned char *sha1)
849 {
850         struct commit *one, *two;
851         struct commit_list *mbs;
852         unsigned char sha1_tmp[20];
853         const char *dots;
854         int st;
855
856         dots = strstr(name, "...");
857         if (!dots)
858                 return get_sha1(name, sha1);
859         if (dots == name)
860                 st = get_sha1("HEAD", sha1_tmp);
861         else {
862                 struct strbuf sb;
863                 strbuf_init(&sb, dots - name);
864                 strbuf_add(&sb, name, dots - name);
865                 st = get_sha1(sb.buf, sha1_tmp);
866                 strbuf_release(&sb);
867         }
868         if (st)
869                 return st;
870         one = lookup_commit_reference_gently(sha1_tmp, 0);
871         if (!one)
872                 return -1;
873
874         if (get_sha1(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
875                 return -1;
876         two = lookup_commit_reference_gently(sha1_tmp, 0);
877         if (!two)
878                 return -1;
879         mbs = get_merge_bases(one, two, 1);
880         if (!mbs || mbs->next)
881                 st = -1;
882         else {
883                 st = 0;
884                 hashcpy(sha1, mbs->item->object.sha1);
885         }
886         free_commit_list(mbs);
887         return st;
888 }
889
890 /*
891  * This reads short-hand syntax that not only evaluates to a commit
892  * object name, but also can act as if the end user spelled the name
893  * of the branch from the command line.
894  *
895  * - "@{-N}" finds the name of the Nth previous branch we were on, and
896  *   places the name of the branch in the given buf and returns the
897  *   number of characters parsed if successful.
898  *
899  * - "<branch>@{upstream}" finds the name of the other ref that
900  *   <branch> is configured to merge with (missing <branch> defaults
901  *   to the current branch), and places the name of the branch in the
902  *   given buf and returns the number of characters parsed if
903  *   successful.
904  *
905  * If the input is not of the accepted format, it returns a negative
906  * number to signal an error.
907  *
908  * If the input was ok but there are not N branch switches in the
909  * reflog, it returns 0.
910  */
911 int interpret_branch_name(const char *name, struct strbuf *buf)
912 {
913         char *cp;
914         struct branch *upstream;
915         int namelen = strlen(name);
916         int len = interpret_nth_prior_checkout(name, buf);
917         int tmp_len;
918
919         if (!len)
920                 return len; /* syntax Ok, not enough switches */
921         if (0 < len && len == namelen)
922                 return len; /* consumed all */
923         else if (0 < len) {
924                 /* we have extra data, which might need further processing */
925                 struct strbuf tmp = STRBUF_INIT;
926                 int used = buf->len;
927                 int ret;
928
929                 strbuf_add(buf, name + len, namelen - len);
930                 ret = interpret_branch_name(buf->buf, &tmp);
931                 /* that data was not interpreted, remove our cruft */
932                 if (ret < 0) {
933                         strbuf_setlen(buf, used);
934                         return len;
935                 }
936                 strbuf_reset(buf);
937                 strbuf_addbuf(buf, &tmp);
938                 strbuf_release(&tmp);
939                 /* tweak for size of {-N} versus expanded ref name */
940                 return ret - used + len;
941         }
942
943         cp = strchr(name, '@');
944         if (!cp)
945                 return -1;
946         tmp_len = upstream_mark(cp, namelen - (cp - name));
947         if (!tmp_len)
948                 return -1;
949         len = cp + tmp_len - name;
950         cp = xstrndup(name, cp - name);
951         upstream = branch_get(*cp ? cp : NULL);
952         if (!upstream
953             || !upstream->merge
954             || !upstream->merge[0]->dst)
955                 return error("No upstream branch found for '%s'", cp);
956         free(cp);
957         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
958         strbuf_reset(buf);
959         strbuf_addstr(buf, cp);
960         free(cp);
961         return len;
962 }
963
964 int strbuf_branchname(struct strbuf *sb, const char *name)
965 {
966         int len = strlen(name);
967         if (interpret_branch_name(name, sb) == len)
968                 return 0;
969         strbuf_add(sb, name, len);
970         return len;
971 }
972
973 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
974 {
975         strbuf_branchname(sb, name);
976         if (name[0] == '-')
977                 return CHECK_REF_FORMAT_ERROR;
978         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
979         return check_ref_format(sb->buf);
980 }
981
982 /*
983  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
984  * notably "xyz^" for "parent of xyz"
985  */
986 int get_sha1(const char *name, unsigned char *sha1)
987 {
988         struct object_context unused;
989         return get_sha1_with_context(name, sha1, &unused);
990 }
991
992 /* Must be called only when object_name:filename doesn't exist. */
993 static void diagnose_invalid_sha1_path(const char *prefix,
994                                        const char *filename,
995                                        const unsigned char *tree_sha1,
996                                        const char *object_name)
997 {
998         struct stat st;
999         unsigned char sha1[20];
1000         unsigned mode;
1001
1002         if (!prefix)
1003                 prefix = "";
1004
1005         if (!lstat(filename, &st))
1006                 die("Path '%s' exists on disk, but not in '%s'.",
1007                     filename, object_name);
1008         if (errno == ENOENT || errno == ENOTDIR) {
1009                 char *fullname = xmalloc(strlen(filename)
1010                                              + strlen(prefix) + 1);
1011                 strcpy(fullname, prefix);
1012                 strcat(fullname, filename);
1013
1014                 if (!get_tree_entry(tree_sha1, fullname,
1015                                     sha1, &mode)) {
1016                         die("Path '%s' exists, but not '%s'.\n"
1017                             "Did you mean '%s:%s'?",
1018                             fullname,
1019                             filename,
1020                             object_name,
1021                             fullname);
1022                 }
1023                 die("Path '%s' does not exist in '%s'",
1024                     filename, object_name);
1025         }
1026 }
1027
1028 /* Must be called only when :stage:filename doesn't exist. */
1029 static void diagnose_invalid_index_path(int stage,
1030                                         const char *prefix,
1031                                         const char *filename)
1032 {
1033         struct stat st;
1034         struct cache_entry *ce;
1035         int pos;
1036         unsigned namelen = strlen(filename);
1037         unsigned fullnamelen;
1038         char *fullname;
1039
1040         if (!prefix)
1041                 prefix = "";
1042
1043         /* Wrong stage number? */
1044         pos = cache_name_pos(filename, namelen);
1045         if (pos < 0)
1046                 pos = -pos - 1;
1047         if (pos < active_nr) {
1048                 ce = active_cache[pos];
1049                 if (ce_namelen(ce) == namelen &&
1050                     !memcmp(ce->name, filename, namelen))
1051                         die("Path '%s' is in the index, but not at stage %d.\n"
1052                             "Did you mean ':%d:%s'?",
1053                             filename, stage,
1054                             ce_stage(ce), filename);
1055         }
1056
1057         /* Confusion between relative and absolute filenames? */
1058         fullnamelen = namelen + strlen(prefix);
1059         fullname = xmalloc(fullnamelen + 1);
1060         strcpy(fullname, prefix);
1061         strcat(fullname, filename);
1062         pos = cache_name_pos(fullname, fullnamelen);
1063         if (pos < 0)
1064                 pos = -pos - 1;
1065         if (pos < active_nr) {
1066                 ce = active_cache[pos];
1067                 if (ce_namelen(ce) == fullnamelen &&
1068                     !memcmp(ce->name, fullname, fullnamelen))
1069                         die("Path '%s' is in the index, but not '%s'.\n"
1070                             "Did you mean ':%d:%s'?",
1071                             fullname, filename,
1072                             ce_stage(ce), fullname);
1073         }
1074
1075         if (!lstat(filename, &st))
1076                 die("Path '%s' exists on disk, but not in the index.", filename);
1077         if (errno == ENOENT || errno == ENOTDIR)
1078                 die("Path '%s' does not exist (neither on disk nor in the index).",
1079                     filename);
1080
1081         free(fullname);
1082 }
1083
1084
1085 int get_sha1_with_mode_1(const char *name, unsigned char *sha1, unsigned *mode, int gently, const char *prefix)
1086 {
1087         struct object_context oc;
1088         int ret;
1089         ret = get_sha1_with_context_1(name, sha1, &oc, gently, prefix);
1090         *mode = oc.mode;
1091         return ret;
1092 }
1093
1094 static char *resolve_relative_path(const char *rel)
1095 {
1096         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1097                 return NULL;
1098
1099         if (!startup_info)
1100                 die("BUG: startup_info struct is not initialized.");
1101
1102         if (!is_inside_work_tree())
1103                 die("relative path syntax can't be used outside working tree.");
1104
1105         /* die() inside prefix_path() if resolved path is outside worktree */
1106         return prefix_path(startup_info->prefix,
1107                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1108                            rel);
1109 }
1110
1111 int get_sha1_with_context_1(const char *name, unsigned char *sha1,
1112                             struct object_context *oc,
1113                             int gently, const char *prefix)
1114 {
1115         int ret, bracket_depth;
1116         int namelen = strlen(name);
1117         const char *cp;
1118
1119         memset(oc, 0, sizeof(*oc));
1120         oc->mode = S_IFINVALID;
1121         ret = get_sha1_1(name, namelen, sha1);
1122         if (!ret)
1123                 return ret;
1124         /* sha1:path --> object name of path in ent sha1
1125          * :path -> object name of absolute path in index
1126          * :./path -> object name of path relative to cwd in index
1127          * :[0-3]:path -> object name of path in index at stage
1128          * :/foo -> recent commit matching foo
1129          */
1130         if (name[0] == ':') {
1131                 int stage = 0;
1132                 struct cache_entry *ce;
1133                 char *new_path = NULL;
1134                 int pos;
1135                 if (namelen > 2 && name[1] == '/') {
1136                         struct commit_list *list = NULL;
1137                         for_each_ref(handle_one_ref, &list);
1138                         return get_sha1_oneline(name + 2, sha1, list);
1139                 }
1140                 if (namelen < 3 ||
1141                     name[2] != ':' ||
1142                     name[1] < '0' || '3' < name[1])
1143                         cp = name + 1;
1144                 else {
1145                         stage = name[1] - '0';
1146                         cp = name + 3;
1147                 }
1148                 new_path = resolve_relative_path(cp);
1149                 if (!new_path) {
1150                         namelen = namelen - (cp - name);
1151                 } else {
1152                         cp = new_path;
1153                         namelen = strlen(cp);
1154                 }
1155
1156                 strncpy(oc->path, cp,
1157                         sizeof(oc->path));
1158                 oc->path[sizeof(oc->path)-1] = '\0';
1159
1160                 if (!active_cache)
1161                         read_cache();
1162                 pos = cache_name_pos(cp, namelen);
1163                 if (pos < 0)
1164                         pos = -pos - 1;
1165                 while (pos < active_nr) {
1166                         ce = active_cache[pos];
1167                         if (ce_namelen(ce) != namelen ||
1168                             memcmp(ce->name, cp, namelen))
1169                                 break;
1170                         if (ce_stage(ce) == stage) {
1171                                 hashcpy(sha1, ce->sha1);
1172                                 oc->mode = ce->ce_mode;
1173                                 free(new_path);
1174                                 return 0;
1175                         }
1176                         pos++;
1177                 }
1178                 if (!gently)
1179                         diagnose_invalid_index_path(stage, prefix, cp);
1180                 free(new_path);
1181                 return -1;
1182         }
1183         for (cp = name, bracket_depth = 0; *cp; cp++) {
1184                 if (*cp == '{')
1185                         bracket_depth++;
1186                 else if (bracket_depth && *cp == '}')
1187                         bracket_depth--;
1188                 else if (!bracket_depth && *cp == ':')
1189                         break;
1190         }
1191         if (*cp == ':') {
1192                 unsigned char tree_sha1[20];
1193                 char *object_name = NULL;
1194                 if (!gently) {
1195                         object_name = xmalloc(cp-name+1);
1196                         strncpy(object_name, name, cp-name);
1197                         object_name[cp-name] = '\0';
1198                 }
1199                 if (!get_sha1_1(name, cp-name, tree_sha1)) {
1200                         const char *filename = cp+1;
1201                         char *new_filename = NULL;
1202
1203                         new_filename = resolve_relative_path(filename);
1204                         if (new_filename)
1205                                 filename = new_filename;
1206                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1207                         if (!gently) {
1208                                 diagnose_invalid_sha1_path(prefix, filename,
1209                                                            tree_sha1, object_name);
1210                                 free(object_name);
1211                         }
1212                         hashcpy(oc->tree, tree_sha1);
1213                         strncpy(oc->path, filename,
1214                                 sizeof(oc->path));
1215                         oc->path[sizeof(oc->path)-1] = '\0';
1216
1217                         free(new_filename);
1218                         return ret;
1219                 } else {
1220                         if (!gently)
1221                                 die("Invalid object name '%s'.", object_name);
1222                 }
1223         }
1224         return ret;
1225 }