get_sha1: warn about full or short object names that look like refs
[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 typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
13
14 struct disambiguate_state {
15         disambiguate_hint_fn fn;
16         void *cb_data;
17         unsigned char candidate[20];
18         unsigned candidate_exists:1;
19         unsigned candidate_checked:1;
20         unsigned candidate_ok:1;
21         unsigned disambiguate_fn_used:1;
22         unsigned ambiguous:1;
23         unsigned always_call_fn:1;
24 };
25
26 static void update_candidates(struct disambiguate_state *ds, const unsigned char *current)
27 {
28         if (ds->always_call_fn) {
29                 ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
30                 return;
31         }
32         if (!ds->candidate_exists) {
33                 /* this is the first candidate */
34                 hashcpy(ds->candidate, current);
35                 ds->candidate_exists = 1;
36                 return;
37         } else if (!hashcmp(ds->candidate, current)) {
38                 /* the same as what we already have seen */
39                 return;
40         }
41
42         if (!ds->fn) {
43                 /* cannot disambiguate between ds->candidate and current */
44                 ds->ambiguous = 1;
45                 return;
46         }
47
48         if (!ds->candidate_checked) {
49                 ds->candidate_ok = ds->fn(ds->candidate, ds->cb_data);
50                 ds->disambiguate_fn_used = 1;
51                 ds->candidate_checked = 1;
52         }
53
54         if (!ds->candidate_ok) {
55                 /* discard the candidate; we know it does not satisify fn */
56                 hashcpy(ds->candidate, current);
57                 ds->candidate_checked = 0;
58                 return;
59         }
60
61         /* if we reach this point, we know ds->candidate satisfies fn */
62         if (ds->fn(current, ds->cb_data)) {
63                 /*
64                  * if both current and candidate satisfy fn, we cannot
65                  * disambiguate.
66                  */
67                 ds->candidate_ok = 0;
68                 ds->ambiguous = 1;
69         }
70
71         /* otherwise, current can be discarded and candidate is still good */
72 }
73
74 static void find_short_object_filename(int len, const char *hex_pfx, struct disambiguate_state *ds)
75 {
76         struct alternate_object_database *alt;
77         char hex[40];
78         static struct alternate_object_database *fakeent;
79
80         if (!fakeent) {
81                 /*
82                  * Create a "fake" alternate object database that
83                  * points to our own object database, to make it
84                  * easier to get a temporary working space in
85                  * alt->name/alt->base while iterating over the
86                  * object databases including our own.
87                  */
88                 const char *objdir = get_object_directory();
89                 int objdir_len = strlen(objdir);
90                 int entlen = objdir_len + 43;
91                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
92                 memcpy(fakeent->base, objdir, objdir_len);
93                 fakeent->name = fakeent->base + objdir_len + 1;
94                 fakeent->name[-1] = '/';
95         }
96         fakeent->next = alt_odb_list;
97
98         sprintf(hex, "%.2s", hex_pfx);
99         for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
100                 struct dirent *de;
101                 DIR *dir;
102                 sprintf(alt->name, "%.2s/", hex_pfx);
103                 dir = opendir(alt->base);
104                 if (!dir)
105                         continue;
106
107                 while (!ds->ambiguous && (de = readdir(dir)) != NULL) {
108                         unsigned char sha1[20];
109
110                         if (strlen(de->d_name) != 38)
111                                 continue;
112                         if (memcmp(de->d_name, hex_pfx + 2, len - 2))
113                                 continue;
114                         memcpy(hex + 2, de->d_name, 38);
115                         if (!get_sha1_hex(hex, sha1))
116                                 update_candidates(ds, sha1);
117                 }
118                 closedir(dir);
119         }
120 }
121
122 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
123 {
124         do {
125                 if (*a != *b)
126                         return 0;
127                 a++;
128                 b++;
129                 len -= 2;
130         } while (len > 1);
131         if (len)
132                 if ((*a ^ *b) & 0xf0)
133                         return 0;
134         return 1;
135 }
136
137 static void unique_in_pack(int len,
138                           const unsigned char *bin_pfx,
139                            struct packed_git *p,
140                            struct disambiguate_state *ds)
141 {
142         uint32_t num, last, i, first = 0;
143         const unsigned char *current = NULL;
144
145         open_pack_index(p);
146         num = p->num_objects;
147         last = num;
148         while (first < last) {
149                 uint32_t mid = (first + last) / 2;
150                 const unsigned char *current;
151                 int cmp;
152
153                 current = nth_packed_object_sha1(p, mid);
154                 cmp = hashcmp(bin_pfx, current);
155                 if (!cmp) {
156                         first = mid;
157                         break;
158                 }
159                 if (cmp > 0) {
160                         first = mid+1;
161                         continue;
162                 }
163                 last = mid;
164         }
165
166         /*
167          * At this point, "first" is the location of the lowest object
168          * with an object name that could match "bin_pfx".  See if we have
169          * 0, 1 or more objects that actually match(es).
170          */
171         for (i = first; i < num && !ds->ambiguous; i++) {
172                 current = nth_packed_object_sha1(p, i);
173                 if (!match_sha(len, bin_pfx, current))
174                         break;
175                 update_candidates(ds, current);
176         }
177 }
178
179 static void find_short_packed_object(int len, const unsigned char *bin_pfx,
180                                      struct disambiguate_state *ds)
181 {
182         struct packed_git *p;
183
184         prepare_packed_git();
185         for (p = packed_git; p && !ds->ambiguous; p = p->next)
186                 unique_in_pack(len, bin_pfx, p, ds);
187 }
188
189 #define SHORT_NAME_NOT_FOUND (-1)
190 #define SHORT_NAME_AMBIGUOUS (-2)
191
192 static int finish_object_disambiguation(struct disambiguate_state *ds,
193                                         unsigned char *sha1)
194 {
195         if (ds->ambiguous)
196                 return SHORT_NAME_AMBIGUOUS;
197
198         if (!ds->candidate_exists)
199                 return SHORT_NAME_NOT_FOUND;
200
201         if (!ds->candidate_checked)
202                 /*
203                  * If this is the only candidate, there is no point
204                  * calling the disambiguation hint callback.
205                  *
206                  * On the other hand, if the current candidate
207                  * replaced an earlier candidate that did _not_ pass
208                  * the disambiguation hint callback, then we do have
209                  * more than one objects that match the short name
210                  * given, so we should make sure this one matches;
211                  * otherwise, if we discovered this one and the one
212                  * that we previously discarded in the reverse order,
213                  * we would end up showing different results in the
214                  * same repository!
215                  */
216                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
217                                     ds->fn(ds->candidate, ds->cb_data));
218
219         if (!ds->candidate_ok)
220                 return SHORT_NAME_AMBIGUOUS;
221
222         hashcpy(sha1, ds->candidate);
223         return 0;
224 }
225
226 static int disambiguate_commit_only(const unsigned char *sha1, void *cb_data_unused)
227 {
228         int kind = sha1_object_info(sha1, NULL);
229         return kind == OBJ_COMMIT;
230 }
231
232 static int disambiguate_committish_only(const unsigned char *sha1, void *cb_data_unused)
233 {
234         struct object *obj;
235         int kind;
236
237         kind = sha1_object_info(sha1, NULL);
238         if (kind == OBJ_COMMIT)
239                 return 1;
240         if (kind != OBJ_TAG)
241                 return 0;
242
243         /* We need to do this the hard way... */
244         obj = deref_tag(lookup_object(sha1), NULL, 0);
245         if (obj && obj->type == OBJ_COMMIT)
246                 return 1;
247         return 0;
248 }
249
250 static int disambiguate_tree_only(const unsigned char *sha1, void *cb_data_unused)
251 {
252         int kind = sha1_object_info(sha1, NULL);
253         return kind == OBJ_TREE;
254 }
255
256 static int disambiguate_treeish_only(const unsigned char *sha1, void *cb_data_unused)
257 {
258         struct object *obj;
259         int kind;
260
261         kind = sha1_object_info(sha1, NULL);
262         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
263                 return 1;
264         if (kind != OBJ_TAG)
265                 return 0;
266
267         /* We need to do this the hard way... */
268         obj = deref_tag(lookup_object(sha1), NULL, 0);
269         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
270                 return 1;
271         return 0;
272 }
273
274 static int disambiguate_blob_only(const unsigned char *sha1, void *cb_data_unused)
275 {
276         int kind = sha1_object_info(sha1, NULL);
277         return kind == OBJ_BLOB;
278 }
279
280 static int prepare_prefixes(const char *name, int len,
281                             unsigned char *bin_pfx,
282                             char *hex_pfx)
283 {
284         int i;
285
286         hashclr(bin_pfx);
287         memset(hex_pfx, 'x', 40);
288         for (i = 0; i < len ;i++) {
289                 unsigned char c = name[i];
290                 unsigned char val;
291                 if (c >= '0' && c <= '9')
292                         val = c - '0';
293                 else if (c >= 'a' && c <= 'f')
294                         val = c - 'a' + 10;
295                 else if (c >= 'A' && c <='F') {
296                         val = c - 'A' + 10;
297                         c -= 'A' - 'a';
298                 }
299                 else
300                         return -1;
301                 hex_pfx[i] = c;
302                 if (!(i & 1))
303                         val <<= 4;
304                 bin_pfx[i >> 1] |= val;
305         }
306         return 0;
307 }
308
309 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
310                           unsigned flags)
311 {
312         int status;
313         char hex_pfx[40];
314         unsigned char bin_pfx[20];
315         struct disambiguate_state ds;
316         int quietly = !!(flags & GET_SHA1_QUIETLY);
317
318         if (len < MINIMUM_ABBREV || len > 40)
319                 return -1;
320         if (prepare_prefixes(name, len, bin_pfx, hex_pfx) < 0)
321                 return -1;
322
323         prepare_alt_odb();
324
325         memset(&ds, 0, sizeof(ds));
326         if (flags & GET_SHA1_COMMIT)
327                 ds.fn = disambiguate_commit_only;
328         else if (flags & GET_SHA1_COMMITTISH)
329                 ds.fn = disambiguate_committish_only;
330         else if (flags & GET_SHA1_TREE)
331                 ds.fn = disambiguate_tree_only;
332         else if (flags & GET_SHA1_TREEISH)
333                 ds.fn = disambiguate_treeish_only;
334         else if (flags & GET_SHA1_BLOB)
335                 ds.fn = disambiguate_blob_only;
336
337         find_short_object_filename(len, hex_pfx, &ds);
338         find_short_packed_object(len, bin_pfx, &ds);
339         status = finish_object_disambiguation(&ds, sha1);
340
341         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
342                 return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
343         return status;
344 }
345
346
347 int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
348 {
349         char hex_pfx[40];
350         unsigned char bin_pfx[20];
351         struct disambiguate_state ds;
352         int len = strlen(prefix);
353
354         if (len < MINIMUM_ABBREV || len > 40)
355                 return -1;
356         if (prepare_prefixes(prefix, len, bin_pfx, hex_pfx) < 0)
357                 return -1;
358
359         prepare_alt_odb();
360
361         memset(&ds, 0, sizeof(ds));
362         ds.always_call_fn = 1;
363         ds.cb_data = cb_data;
364         ds.fn = fn;
365
366         find_short_object_filename(len, hex_pfx, &ds);
367         find_short_packed_object(len, bin_pfx, &ds);
368         return ds.ambiguous;
369 }
370
371 const char *find_unique_abbrev(const unsigned char *sha1, int len)
372 {
373         int status, exists;
374         static char hex[41];
375
376         exists = has_sha1_file(sha1);
377         memcpy(hex, sha1_to_hex(sha1), 40);
378         if (len == 40 || !len)
379                 return hex;
380         while (len < 40) {
381                 unsigned char sha1_ret[20];
382                 status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
383                 if (exists
384                     ? !status
385                     : status == SHORT_NAME_NOT_FOUND) {
386                         hex[len] = 0;
387                         return hex;
388                 }
389                 len++;
390         }
391         return hex;
392 }
393
394 static int ambiguous_path(const char *path, int len)
395 {
396         int slash = 1;
397         int cnt;
398
399         for (cnt = 0; cnt < len; cnt++) {
400                 switch (*path++) {
401                 case '\0':
402                         break;
403                 case '/':
404                         if (slash)
405                                 break;
406                         slash = 1;
407                         continue;
408                 case '.':
409                         continue;
410                 default:
411                         slash = 0;
412                         continue;
413                 }
414                 break;
415         }
416         return slash;
417 }
418
419 static inline int upstream_mark(const char *string, int len)
420 {
421         const char *suffix[] = { "@{upstream}", "@{u}" };
422         int i;
423
424         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
425                 int suffix_len = strlen(suffix[i]);
426                 if (suffix_len <= len
427                     && !memcmp(string, suffix[i], suffix_len))
428                         return suffix_len;
429         }
430         return 0;
431 }
432
433 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
434
435 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
436 {
437         static const char *warn_msg = "refname '%.*s' is ambiguous.";
438         static const char *object_name_msg = N_(
439         "Git normally never creates a ref that ends with 40 hex characters\n"
440         "because it will be ignored when you just specify 40-hex. These refs\n"
441         "may be created by mistake. For example,\n"
442         "\n"
443         "  git checkout -b $br $(git rev-parse ...)\n"
444         "\n"
445         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
446         "examine these refs and maybe delete them. Turn this message off by\n"
447         "running \"git config advice.object_name_warning false\"");
448         unsigned char tmp_sha1[20];
449         char *real_ref = NULL;
450         int refs_found = 0;
451         int at, reflog_len;
452
453         if (len == 40 && !get_sha1_hex(str, sha1)) {
454                 refs_found = dwim_ref(str, len, tmp_sha1, &real_ref);
455                 if (refs_found > 0 && warn_ambiguous_refs) {
456                         warning(warn_msg, len, str);
457                         if (advice_object_name_warning)
458                                 fprintf(stderr, "%s\n", _(object_name_msg));
459                 }
460                 free(real_ref);
461                 return 0;
462         }
463
464         /* basic@{time or number or -number} format to query ref-log */
465         reflog_len = at = 0;
466         if (len && str[len-1] == '}') {
467                 for (at = len-2; at >= 0; at--) {
468                         if (str[at] == '@' && str[at+1] == '{') {
469                                 if (!upstream_mark(str + at, len - at)) {
470                                         reflog_len = (len-1) - (at+2);
471                                         len = at;
472                                 }
473                                 break;
474                         }
475                 }
476         }
477
478         /* Accept only unambiguous ref paths. */
479         if (len && ambiguous_path(str, len))
480                 return -1;
481
482         if (!len && reflog_len) {
483                 struct strbuf buf = STRBUF_INIT;
484                 int ret;
485                 /* try the @{-N} syntax for n-th checkout */
486                 ret = interpret_branch_name(str+at, &buf);
487                 if (ret > 0) {
488                         /* substitute this branch name and restart */
489                         return get_sha1_1(buf.buf, buf.len, sha1, 0);
490                 } else if (ret == 0) {
491                         return -1;
492                 }
493                 /* allow "@{...}" to mean the current branch reflog */
494                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
495         } else if (reflog_len)
496                 refs_found = dwim_log(str, len, sha1, &real_ref);
497         else
498                 refs_found = dwim_ref(str, len, sha1, &real_ref);
499
500         if (!refs_found)
501                 return -1;
502
503         if (warn_ambiguous_refs &&
504             (refs_found > 1 ||
505              !get_short_sha1(str, len, tmp_sha1, GET_SHA1_QUIETLY)))
506                 warning(warn_msg, len, str);
507
508         if (reflog_len) {
509                 int nth, i;
510                 unsigned long at_time;
511                 unsigned long co_time;
512                 int co_tz, co_cnt;
513
514                 /* a @{-N} placed anywhere except the start is an error */
515                 if (str[at+2] == '-')
516                         return -1;
517
518                 /* Is it asking for N-th entry, or approxidate? */
519                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
520                         char ch = str[at+2+i];
521                         if ('0' <= ch && ch <= '9')
522                                 nth = nth * 10 + ch - '0';
523                         else
524                                 nth = -1;
525                 }
526                 if (100000000 <= nth) {
527                         at_time = nth;
528                         nth = -1;
529                 } else if (0 <= nth)
530                         at_time = 0;
531                 else {
532                         int errors = 0;
533                         char *tmp = xstrndup(str + at + 2, reflog_len);
534                         at_time = approxidate_careful(tmp, &errors);
535                         free(tmp);
536                         if (errors)
537                                 return -1;
538                 }
539                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
540                                 &co_time, &co_tz, &co_cnt)) {
541                         if (at_time)
542                                 warning("Log for '%.*s' only goes "
543                                         "back to %s.", len, str,
544                                         show_date(co_time, co_tz, DATE_RFC2822));
545                         else {
546                                 free(real_ref);
547                                 die("Log for '%.*s' only has %d entries.",
548                                     len, str, co_cnt);
549                         }
550                 }
551         }
552
553         free(real_ref);
554         return 0;
555 }
556
557 static int get_parent(const char *name, int len,
558                       unsigned char *result, int idx)
559 {
560         unsigned char sha1[20];
561         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
562         struct commit *commit;
563         struct commit_list *p;
564
565         if (ret)
566                 return ret;
567         commit = lookup_commit_reference(sha1);
568         if (!commit)
569                 return -1;
570         if (parse_commit(commit))
571                 return -1;
572         if (!idx) {
573                 hashcpy(result, commit->object.sha1);
574                 return 0;
575         }
576         p = commit->parents;
577         while (p) {
578                 if (!--idx) {
579                         hashcpy(result, p->item->object.sha1);
580                         return 0;
581                 }
582                 p = p->next;
583         }
584         return -1;
585 }
586
587 static int get_nth_ancestor(const char *name, int len,
588                             unsigned char *result, int generation)
589 {
590         unsigned char sha1[20];
591         struct commit *commit;
592         int ret;
593
594         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
595         if (ret)
596                 return ret;
597         commit = lookup_commit_reference(sha1);
598         if (!commit)
599                 return -1;
600
601         while (generation--) {
602                 if (parse_commit(commit) || !commit->parents)
603                         return -1;
604                 commit = commit->parents->item;
605         }
606         hashcpy(result, commit->object.sha1);
607         return 0;
608 }
609
610 struct object *peel_to_type(const char *name, int namelen,
611                             struct object *o, enum object_type expected_type)
612 {
613         if (name && !namelen)
614                 namelen = strlen(name);
615         while (1) {
616                 if (!o || (!o->parsed && !parse_object(o->sha1)))
617                         return NULL;
618                 if (o->type == expected_type)
619                         return o;
620                 if (o->type == OBJ_TAG)
621                         o = ((struct tag*) o)->tagged;
622                 else if (o->type == OBJ_COMMIT)
623                         o = &(((struct commit *) o)->tree->object);
624                 else {
625                         if (name)
626                                 error("%.*s: expected %s type, but the object "
627                                       "dereferences to %s type",
628                                       namelen, name, typename(expected_type),
629                                       typename(o->type));
630                         return NULL;
631                 }
632         }
633 }
634
635 static int peel_onion(const char *name, int len, unsigned char *sha1)
636 {
637         unsigned char outer[20];
638         const char *sp;
639         unsigned int expected_type = 0;
640         unsigned lookup_flags = 0;
641         struct object *o;
642
643         /*
644          * "ref^{type}" dereferences ref repeatedly until you cannot
645          * dereference anymore, or you get an object of given type,
646          * whichever comes first.  "ref^{}" means just dereference
647          * tags until you get a non-tag.  "ref^0" is a shorthand for
648          * "ref^{commit}".  "commit^{tree}" could be used to find the
649          * top-level tree of the given commit.
650          */
651         if (len < 4 || name[len-1] != '}')
652                 return -1;
653
654         for (sp = name + len - 1; name <= sp; sp--) {
655                 int ch = *sp;
656                 if (ch == '{' && name < sp && sp[-1] == '^')
657                         break;
658         }
659         if (sp <= name)
660                 return -1;
661
662         sp++; /* beginning of type name, or closing brace for empty */
663         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
664                 expected_type = OBJ_COMMIT;
665         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
666                 expected_type = OBJ_TREE;
667         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
668                 expected_type = OBJ_BLOB;
669         else if (sp[0] == '}')
670                 expected_type = OBJ_NONE;
671         else if (sp[0] == '/')
672                 expected_type = OBJ_COMMIT;
673         else
674                 return -1;
675
676         if (expected_type == OBJ_COMMIT)
677                 lookup_flags = GET_SHA1_COMMITTISH;
678
679         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
680                 return -1;
681
682         o = parse_object(outer);
683         if (!o)
684                 return -1;
685         if (!expected_type) {
686                 o = deref_tag(o, name, sp - name - 2);
687                 if (!o || (!o->parsed && !parse_object(o->sha1)))
688                         return -1;
689                 hashcpy(sha1, o->sha1);
690                 return 0;
691         }
692
693         /*
694          * At this point, the syntax look correct, so
695          * if we do not get the needed object, we should
696          * barf.
697          */
698         o = peel_to_type(name, len, o, expected_type);
699         if (!o)
700                 return -1;
701
702         hashcpy(sha1, o->sha1);
703         if (sp[0] == '/') {
704                 /* "$commit^{/foo}" */
705                 char *prefix;
706                 int ret;
707                 struct commit_list *list = NULL;
708
709                 /*
710                  * $commit^{/}. Some regex implementation may reject.
711                  * We don't need regex anyway. '' pattern always matches.
712                  */
713                 if (sp[1] == '}')
714                         return 0;
715
716                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
717                 commit_list_insert((struct commit *)o, &list);
718                 ret = get_sha1_oneline(prefix, sha1, list);
719                 free(prefix);
720                 return ret;
721         }
722         return 0;
723 }
724
725 static int get_describe_name(const char *name, int len, unsigned char *sha1)
726 {
727         const char *cp;
728         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
729
730         for (cp = name + len - 1; name + 2 <= cp; cp--) {
731                 char ch = *cp;
732                 if (hexval(ch) & ~0377) {
733                         /* We must be looking at g in "SOMETHING-g"
734                          * for it to be describe output.
735                          */
736                         if (ch == 'g' && cp[-1] == '-') {
737                                 cp++;
738                                 len -= cp - name;
739                                 return get_short_sha1(cp, len, sha1, flags);
740                         }
741                 }
742         }
743         return -1;
744 }
745
746 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
747 {
748         int ret, has_suffix;
749         const char *cp;
750
751         /*
752          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
753          */
754         has_suffix = 0;
755         for (cp = name + len - 1; name <= cp; cp--) {
756                 int ch = *cp;
757                 if ('0' <= ch && ch <= '9')
758                         continue;
759                 if (ch == '~' || ch == '^')
760                         has_suffix = ch;
761                 break;
762         }
763
764         if (has_suffix) {
765                 int num = 0;
766                 int len1 = cp - name;
767                 cp++;
768                 while (cp < name + len)
769                         num = num * 10 + *cp++ - '0';
770                 if (!num && len1 == len - 1)
771                         num = 1;
772                 if (has_suffix == '^')
773                         return get_parent(name, len1, sha1, num);
774                 /* else if (has_suffix == '~') -- goes without saying */
775                 return get_nth_ancestor(name, len1, sha1, num);
776         }
777
778         ret = peel_onion(name, len, sha1);
779         if (!ret)
780                 return 0;
781
782         ret = get_sha1_basic(name, len, sha1);
783         if (!ret)
784                 return 0;
785
786         /* It could be describe output that is "SOMETHING-gXXXX" */
787         ret = get_describe_name(name, len, sha1);
788         if (!ret)
789                 return 0;
790
791         return get_short_sha1(name, len, sha1, lookup_flags);
792 }
793
794 /*
795  * This interprets names like ':/Initial revision of "git"' by searching
796  * through history and returning the first commit whose message starts
797  * the given regular expression.
798  *
799  * For future extension, ':/!' is reserved. If you want to match a message
800  * beginning with a '!', you have to repeat the exclamation mark.
801  */
802 #define ONELINE_SEEN (1u<<20)
803
804 static int handle_one_ref(const char *path,
805                 const unsigned char *sha1, int flag, void *cb_data)
806 {
807         struct commit_list **list = cb_data;
808         struct object *object = parse_object(sha1);
809         if (!object)
810                 return 0;
811         if (object->type == OBJ_TAG) {
812                 object = deref_tag(object, path, strlen(path));
813                 if (!object)
814                         return 0;
815         }
816         if (object->type != OBJ_COMMIT)
817                 return 0;
818         commit_list_insert_by_date((struct commit *)object, list);
819         return 0;
820 }
821
822 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
823                             struct commit_list *list)
824 {
825         struct commit_list *backup = NULL, *l;
826         int found = 0;
827         regex_t regex;
828
829         if (prefix[0] == '!') {
830                 if (prefix[1] != '!')
831                         die ("Invalid search pattern: %s", prefix);
832                 prefix++;
833         }
834
835         if (regcomp(&regex, prefix, REG_EXTENDED))
836                 die("Invalid search pattern: %s", prefix);
837
838         for (l = list; l; l = l->next) {
839                 l->item->object.flags |= ONELINE_SEEN;
840                 commit_list_insert(l->item, &backup);
841         }
842         while (list) {
843                 char *p, *to_free = NULL;
844                 struct commit *commit;
845                 enum object_type type;
846                 unsigned long size;
847                 int matches;
848
849                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
850                 if (!parse_object(commit->object.sha1))
851                         continue;
852                 if (commit->buffer)
853                         p = commit->buffer;
854                 else {
855                         p = read_sha1_file(commit->object.sha1, &type, &size);
856                         if (!p)
857                                 continue;
858                         to_free = p;
859                 }
860
861                 p = strstr(p, "\n\n");
862                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
863                 free(to_free);
864
865                 if (matches) {
866                         hashcpy(sha1, commit->object.sha1);
867                         found = 1;
868                         break;
869                 }
870         }
871         regfree(&regex);
872         free_commit_list(list);
873         for (l = backup; l; l = l->next)
874                 clear_commit_marks(l->item, ONELINE_SEEN);
875         free_commit_list(backup);
876         return found ? 0 : -1;
877 }
878
879 struct grab_nth_branch_switch_cbdata {
880         long cnt, alloc;
881         struct strbuf *buf;
882 };
883
884 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
885                                   const char *email, unsigned long timestamp, int tz,
886                                   const char *message, void *cb_data)
887 {
888         struct grab_nth_branch_switch_cbdata *cb = cb_data;
889         const char *match = NULL, *target = NULL;
890         size_t len;
891         int nth;
892
893         if (!prefixcmp(message, "checkout: moving from ")) {
894                 match = message + strlen("checkout: moving from ");
895                 target = strstr(match, " to ");
896         }
897
898         if (!match || !target)
899                 return 0;
900
901         len = target - match;
902         nth = cb->cnt++ % cb->alloc;
903         strbuf_reset(&cb->buf[nth]);
904         strbuf_add(&cb->buf[nth], match, len);
905         return 0;
906 }
907
908 /*
909  * Parse @{-N} syntax, return the number of characters parsed
910  * if successful; otherwise signal an error with negative value.
911  */
912 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
913 {
914         long nth;
915         int i, retval;
916         struct grab_nth_branch_switch_cbdata cb;
917         const char *brace;
918         char *num_end;
919
920         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
921                 return -1;
922         brace = strchr(name, '}');
923         if (!brace)
924                 return -1;
925         nth = strtol(name+3, &num_end, 10);
926         if (num_end != brace)
927                 return -1;
928         if (nth <= 0)
929                 return -1;
930         cb.alloc = nth;
931         cb.buf = xmalloc(nth * sizeof(struct strbuf));
932         for (i = 0; i < nth; i++)
933                 strbuf_init(&cb.buf[i], 20);
934         cb.cnt = 0;
935         retval = 0;
936         for_each_recent_reflog_ent("HEAD", grab_nth_branch_switch, 40960, &cb);
937         if (cb.cnt < nth) {
938                 cb.cnt = 0;
939                 for_each_reflog_ent("HEAD", grab_nth_branch_switch, &cb);
940         }
941         if (cb.cnt < nth)
942                 goto release_return;
943         i = cb.cnt % nth;
944         strbuf_reset(buf);
945         strbuf_add(buf, cb.buf[i].buf, cb.buf[i].len);
946         retval = brace-name+1;
947
948 release_return:
949         for (i = 0; i < nth; i++)
950                 strbuf_release(&cb.buf[i]);
951         free(cb.buf);
952
953         return retval;
954 }
955
956 int get_sha1_mb(const char *name, unsigned char *sha1)
957 {
958         struct commit *one, *two;
959         struct commit_list *mbs;
960         unsigned char sha1_tmp[20];
961         const char *dots;
962         int st;
963
964         dots = strstr(name, "...");
965         if (!dots)
966                 return get_sha1(name, sha1);
967         if (dots == name)
968                 st = get_sha1("HEAD", sha1_tmp);
969         else {
970                 struct strbuf sb;
971                 strbuf_init(&sb, dots - name);
972                 strbuf_add(&sb, name, dots - name);
973                 st = get_sha1_committish(sb.buf, sha1_tmp);
974                 strbuf_release(&sb);
975         }
976         if (st)
977                 return st;
978         one = lookup_commit_reference_gently(sha1_tmp, 0);
979         if (!one)
980                 return -1;
981
982         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
983                 return -1;
984         two = lookup_commit_reference_gently(sha1_tmp, 0);
985         if (!two)
986                 return -1;
987         mbs = get_merge_bases(one, two, 1);
988         if (!mbs || mbs->next)
989                 st = -1;
990         else {
991                 st = 0;
992                 hashcpy(sha1, mbs->item->object.sha1);
993         }
994         free_commit_list(mbs);
995         return st;
996 }
997
998 /*
999  * This reads short-hand syntax that not only evaluates to a commit
1000  * object name, but also can act as if the end user spelled the name
1001  * of the branch from the command line.
1002  *
1003  * - "@{-N}" finds the name of the Nth previous branch we were on, and
1004  *   places the name of the branch in the given buf and returns the
1005  *   number of characters parsed if successful.
1006  *
1007  * - "<branch>@{upstream}" finds the name of the other ref that
1008  *   <branch> is configured to merge with (missing <branch> defaults
1009  *   to the current branch), and places the name of the branch in the
1010  *   given buf and returns the number of characters parsed if
1011  *   successful.
1012  *
1013  * If the input is not of the accepted format, it returns a negative
1014  * number to signal an error.
1015  *
1016  * If the input was ok but there are not N branch switches in the
1017  * reflog, it returns 0.
1018  */
1019 int interpret_branch_name(const char *name, struct strbuf *buf)
1020 {
1021         char *cp;
1022         struct branch *upstream;
1023         int namelen = strlen(name);
1024         int len = interpret_nth_prior_checkout(name, buf);
1025         int tmp_len;
1026
1027         if (!len)
1028                 return len; /* syntax Ok, not enough switches */
1029         if (0 < len && len == namelen)
1030                 return len; /* consumed all */
1031         else if (0 < len) {
1032                 /* we have extra data, which might need further processing */
1033                 struct strbuf tmp = STRBUF_INIT;
1034                 int used = buf->len;
1035                 int ret;
1036
1037                 strbuf_add(buf, name + len, namelen - len);
1038                 ret = interpret_branch_name(buf->buf, &tmp);
1039                 /* that data was not interpreted, remove our cruft */
1040                 if (ret < 0) {
1041                         strbuf_setlen(buf, used);
1042                         return len;
1043                 }
1044                 strbuf_reset(buf);
1045                 strbuf_addbuf(buf, &tmp);
1046                 strbuf_release(&tmp);
1047                 /* tweak for size of {-N} versus expanded ref name */
1048                 return ret - used + len;
1049         }
1050
1051         cp = strchr(name, '@');
1052         if (!cp)
1053                 return -1;
1054         tmp_len = upstream_mark(cp, namelen - (cp - name));
1055         if (!tmp_len)
1056                 return -1;
1057         len = cp + tmp_len - name;
1058         cp = xstrndup(name, cp - name);
1059         upstream = branch_get(*cp ? cp : NULL);
1060         /*
1061          * Upstream can be NULL only if cp refers to HEAD and HEAD
1062          * points to something different than a branch.
1063          */
1064         if (!upstream)
1065                 return error(_("HEAD does not point to a branch"));
1066         if (!upstream->merge || !upstream->merge[0]->dst) {
1067                 if (!ref_exists(upstream->refname))
1068                         return error(_("No such branch: '%s'"), cp);
1069                 if (!upstream->merge)
1070                         return error(_("No upstream configured for branch '%s'"),
1071                                      upstream->name);
1072                 return error(
1073                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1074                         upstream->merge[0]->src);
1075         }
1076         free(cp);
1077         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1078         strbuf_reset(buf);
1079         strbuf_addstr(buf, cp);
1080         free(cp);
1081         return len;
1082 }
1083
1084 int strbuf_branchname(struct strbuf *sb, const char *name)
1085 {
1086         int len = strlen(name);
1087         if (interpret_branch_name(name, sb) == len)
1088                 return 0;
1089         strbuf_add(sb, name, len);
1090         return len;
1091 }
1092
1093 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1094 {
1095         strbuf_branchname(sb, name);
1096         if (name[0] == '-')
1097                 return -1;
1098         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1099         return check_refname_format(sb->buf, 0);
1100 }
1101
1102 /*
1103  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1104  * notably "xyz^" for "parent of xyz"
1105  */
1106 int get_sha1(const char *name, unsigned char *sha1)
1107 {
1108         struct object_context unused;
1109         return get_sha1_with_context(name, 0, sha1, &unused);
1110 }
1111
1112 /*
1113  * Many callers know that the user meant to name a committish by
1114  * syntactical positions where the object name appears.  Calling this
1115  * function allows the machinery to disambiguate shorter-than-unique
1116  * abbreviated object names between committish and others.
1117  *
1118  * Note that this does NOT error out when the named object is not a
1119  * committish. It is merely to give a hint to the disambiguation
1120  * machinery.
1121  */
1122 int get_sha1_committish(const char *name, unsigned char *sha1)
1123 {
1124         struct object_context unused;
1125         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1126                                      sha1, &unused);
1127 }
1128
1129 int get_sha1_treeish(const char *name, unsigned char *sha1)
1130 {
1131         struct object_context unused;
1132         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1133                                      sha1, &unused);
1134 }
1135
1136 int get_sha1_commit(const char *name, unsigned char *sha1)
1137 {
1138         struct object_context unused;
1139         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1140                                      sha1, &unused);
1141 }
1142
1143 int get_sha1_tree(const char *name, unsigned char *sha1)
1144 {
1145         struct object_context unused;
1146         return get_sha1_with_context(name, GET_SHA1_TREE,
1147                                      sha1, &unused);
1148 }
1149
1150 int get_sha1_blob(const char *name, unsigned char *sha1)
1151 {
1152         struct object_context unused;
1153         return get_sha1_with_context(name, GET_SHA1_BLOB,
1154                                      sha1, &unused);
1155 }
1156
1157 /* Must be called only when object_name:filename doesn't exist. */
1158 static void diagnose_invalid_sha1_path(const char *prefix,
1159                                        const char *filename,
1160                                        const unsigned char *tree_sha1,
1161                                        const char *object_name)
1162 {
1163         struct stat st;
1164         unsigned char sha1[20];
1165         unsigned mode;
1166
1167         if (!prefix)
1168                 prefix = "";
1169
1170         if (!lstat(filename, &st))
1171                 die("Path '%s' exists on disk, but not in '%s'.",
1172                     filename, object_name);
1173         if (errno == ENOENT || errno == ENOTDIR) {
1174                 char *fullname = xmalloc(strlen(filename)
1175                                              + strlen(prefix) + 1);
1176                 strcpy(fullname, prefix);
1177                 strcat(fullname, filename);
1178
1179                 if (!get_tree_entry(tree_sha1, fullname,
1180                                     sha1, &mode)) {
1181                         die("Path '%s' exists, but not '%s'.\n"
1182                             "Did you mean '%s:%s' aka '%s:./%s'?",
1183                             fullname,
1184                             filename,
1185                             object_name,
1186                             fullname,
1187                             object_name,
1188                             filename);
1189                 }
1190                 die("Path '%s' does not exist in '%s'",
1191                     filename, object_name);
1192         }
1193 }
1194
1195 /* Must be called only when :stage:filename doesn't exist. */
1196 static void diagnose_invalid_index_path(int stage,
1197                                         const char *prefix,
1198                                         const char *filename)
1199 {
1200         struct stat st;
1201         struct cache_entry *ce;
1202         int pos;
1203         unsigned namelen = strlen(filename);
1204         unsigned fullnamelen;
1205         char *fullname;
1206
1207         if (!prefix)
1208                 prefix = "";
1209
1210         /* Wrong stage number? */
1211         pos = cache_name_pos(filename, namelen);
1212         if (pos < 0)
1213                 pos = -pos - 1;
1214         if (pos < active_nr) {
1215                 ce = active_cache[pos];
1216                 if (ce_namelen(ce) == namelen &&
1217                     !memcmp(ce->name, filename, namelen))
1218                         die("Path '%s' is in the index, but not at stage %d.\n"
1219                             "Did you mean ':%d:%s'?",
1220                             filename, stage,
1221                             ce_stage(ce), filename);
1222         }
1223
1224         /* Confusion between relative and absolute filenames? */
1225         fullnamelen = namelen + strlen(prefix);
1226         fullname = xmalloc(fullnamelen + 1);
1227         strcpy(fullname, prefix);
1228         strcat(fullname, filename);
1229         pos = cache_name_pos(fullname, fullnamelen);
1230         if (pos < 0)
1231                 pos = -pos - 1;
1232         if (pos < active_nr) {
1233                 ce = active_cache[pos];
1234                 if (ce_namelen(ce) == fullnamelen &&
1235                     !memcmp(ce->name, fullname, fullnamelen))
1236                         die("Path '%s' is in the index, but not '%s'.\n"
1237                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1238                             fullname, filename,
1239                             ce_stage(ce), fullname,
1240                             ce_stage(ce), filename);
1241         }
1242
1243         if (!lstat(filename, &st))
1244                 die("Path '%s' exists on disk, but not in the index.", filename);
1245         if (errno == ENOENT || errno == ENOTDIR)
1246                 die("Path '%s' does not exist (neither on disk nor in the index).",
1247                     filename);
1248
1249         free(fullname);
1250 }
1251
1252
1253 static char *resolve_relative_path(const char *rel)
1254 {
1255         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1256                 return NULL;
1257
1258         if (!startup_info)
1259                 die("BUG: startup_info struct is not initialized.");
1260
1261         if (!is_inside_work_tree())
1262                 die("relative path syntax can't be used outside working tree.");
1263
1264         /* die() inside prefix_path() if resolved path is outside worktree */
1265         return prefix_path(startup_info->prefix,
1266                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1267                            rel);
1268 }
1269
1270 static int get_sha1_with_context_1(const char *name,
1271                                    unsigned flags,
1272                                    const char *prefix,
1273                                    unsigned char *sha1,
1274                                    struct object_context *oc)
1275 {
1276         int ret, bracket_depth;
1277         int namelen = strlen(name);
1278         const char *cp;
1279         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1280
1281         memset(oc, 0, sizeof(*oc));
1282         oc->mode = S_IFINVALID;
1283         ret = get_sha1_1(name, namelen, sha1, flags);
1284         if (!ret)
1285                 return ret;
1286         /*
1287          * sha1:path --> object name of path in ent sha1
1288          * :path -> object name of absolute path in index
1289          * :./path -> object name of path relative to cwd in index
1290          * :[0-3]:path -> object name of path in index at stage
1291          * :/foo -> recent commit matching foo
1292          */
1293         if (name[0] == ':') {
1294                 int stage = 0;
1295                 struct cache_entry *ce;
1296                 char *new_path = NULL;
1297                 int pos;
1298                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1299                         struct commit_list *list = NULL;
1300                         for_each_ref(handle_one_ref, &list);
1301                         return get_sha1_oneline(name + 2, sha1, list);
1302                 }
1303                 if (namelen < 3 ||
1304                     name[2] != ':' ||
1305                     name[1] < '0' || '3' < name[1])
1306                         cp = name + 1;
1307                 else {
1308                         stage = name[1] - '0';
1309                         cp = name + 3;
1310                 }
1311                 new_path = resolve_relative_path(cp);
1312                 if (!new_path) {
1313                         namelen = namelen - (cp - name);
1314                 } else {
1315                         cp = new_path;
1316                         namelen = strlen(cp);
1317                 }
1318
1319                 strncpy(oc->path, cp,
1320                         sizeof(oc->path));
1321                 oc->path[sizeof(oc->path)-1] = '\0';
1322
1323                 if (!active_cache)
1324                         read_cache();
1325                 pos = cache_name_pos(cp, namelen);
1326                 if (pos < 0)
1327                         pos = -pos - 1;
1328                 while (pos < active_nr) {
1329                         ce = active_cache[pos];
1330                         if (ce_namelen(ce) != namelen ||
1331                             memcmp(ce->name, cp, namelen))
1332                                 break;
1333                         if (ce_stage(ce) == stage) {
1334                                 hashcpy(sha1, ce->sha1);
1335                                 oc->mode = ce->ce_mode;
1336                                 free(new_path);
1337                                 return 0;
1338                         }
1339                         pos++;
1340                 }
1341                 if (only_to_die && name[1] && name[1] != '/')
1342                         diagnose_invalid_index_path(stage, prefix, cp);
1343                 free(new_path);
1344                 return -1;
1345         }
1346         for (cp = name, bracket_depth = 0; *cp; cp++) {
1347                 if (*cp == '{')
1348                         bracket_depth++;
1349                 else if (bracket_depth && *cp == '}')
1350                         bracket_depth--;
1351                 else if (!bracket_depth && *cp == ':')
1352                         break;
1353         }
1354         if (*cp == ':') {
1355                 unsigned char tree_sha1[20];
1356                 char *object_name = NULL;
1357                 if (only_to_die) {
1358                         object_name = xmalloc(cp-name+1);
1359                         strncpy(object_name, name, cp-name);
1360                         object_name[cp-name] = '\0';
1361                 }
1362                 if (!get_sha1_1(name, cp-name, tree_sha1, GET_SHA1_TREEISH)) {
1363                         const char *filename = cp+1;
1364                         char *new_filename = NULL;
1365
1366                         new_filename = resolve_relative_path(filename);
1367                         if (new_filename)
1368                                 filename = new_filename;
1369                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1370                         if (ret && only_to_die) {
1371                                 diagnose_invalid_sha1_path(prefix, filename,
1372                                                            tree_sha1, object_name);
1373                                 free(object_name);
1374                         }
1375                         hashcpy(oc->tree, tree_sha1);
1376                         strncpy(oc->path, filename,
1377                                 sizeof(oc->path));
1378                         oc->path[sizeof(oc->path)-1] = '\0';
1379
1380                         free(new_filename);
1381                         return ret;
1382                 } else {
1383                         if (only_to_die)
1384                                 die("Invalid object name '%s'.", object_name);
1385                 }
1386         }
1387         return ret;
1388 }
1389
1390 /*
1391  * Call this function when you know "name" given by the end user must
1392  * name an object but it doesn't; the function _may_ die with a better
1393  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1394  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1395  * you have a chance to diagnose the error further.
1396  */
1397 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1398 {
1399         struct object_context oc;
1400         unsigned char sha1[20];
1401         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1402 }
1403
1404 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1405 {
1406         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1407 }