]> git.scripts.mit.edu Git - git.git/blob - sha1_name.c
t5308: check that index-pack --strict detects duplicate objects
[git.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(parse_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         char *real_ref = NULL;
439         int refs_found = 0;
440         int at, reflog_len;
441
442         if (len == 40 && !get_sha1_hex(str, sha1))
443                 return 0;
444
445         /* basic@{time or number or -number} format to query ref-log */
446         reflog_len = at = 0;
447         if (len && str[len-1] == '}') {
448                 for (at = len-2; at >= 0; at--) {
449                         if (str[at] == '@' && str[at+1] == '{') {
450                                 if (!upstream_mark(str + at, len - at)) {
451                                         reflog_len = (len-1) - (at+2);
452                                         len = at;
453                                 }
454                                 break;
455                         }
456                 }
457         }
458
459         /* Accept only unambiguous ref paths. */
460         if (len && ambiguous_path(str, len))
461                 return -1;
462
463         if (!len && reflog_len) {
464                 struct strbuf buf = STRBUF_INIT;
465                 int ret;
466                 /* try the @{-N} syntax for n-th checkout */
467                 ret = interpret_branch_name(str+at, &buf);
468                 if (ret > 0) {
469                         /* substitute this branch name and restart */
470                         return get_sha1_1(buf.buf, buf.len, sha1, 0);
471                 } else if (ret == 0) {
472                         return -1;
473                 }
474                 /* allow "@{...}" to mean the current branch reflog */
475                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
476         } else if (reflog_len)
477                 refs_found = dwim_log(str, len, sha1, &real_ref);
478         else
479                 refs_found = dwim_ref(str, len, sha1, &real_ref);
480
481         if (!refs_found)
482                 return -1;
483
484         if (warn_ambiguous_refs && refs_found > 1)
485                 warning(warn_msg, len, str);
486
487         if (reflog_len) {
488                 int nth, i;
489                 unsigned long at_time;
490                 unsigned long co_time;
491                 int co_tz, co_cnt;
492
493                 /* a @{-N} placed anywhere except the start is an error */
494                 if (str[at+2] == '-')
495                         return -1;
496
497                 /* Is it asking for N-th entry, or approxidate? */
498                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
499                         char ch = str[at+2+i];
500                         if ('0' <= ch && ch <= '9')
501                                 nth = nth * 10 + ch - '0';
502                         else
503                                 nth = -1;
504                 }
505                 if (100000000 <= nth) {
506                         at_time = nth;
507                         nth = -1;
508                 } else if (0 <= nth)
509                         at_time = 0;
510                 else {
511                         int errors = 0;
512                         char *tmp = xstrndup(str + at + 2, reflog_len);
513                         at_time = approxidate_careful(tmp, &errors);
514                         free(tmp);
515                         if (errors)
516                                 return -1;
517                 }
518                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
519                                 &co_time, &co_tz, &co_cnt)) {
520                         if (!len) {
521                                 if (!prefixcmp(real_ref, "refs/heads/")) {
522                                         str = real_ref + 11;
523                                         len = strlen(real_ref + 11);
524                                 } else {
525                                         /* detached HEAD */
526                                         str = "HEAD";
527                                         len = 4;
528                                 }
529                         }
530                         if (at_time)
531                                 warning("Log for '%.*s' only goes "
532                                         "back to %s.", len, str,
533                                         show_date(co_time, co_tz, DATE_RFC2822));
534                         else {
535                                 die("Log for '%.*s' only has %d entries.",
536                                     len, str, co_cnt);
537                         }
538                 }
539         }
540
541         free(real_ref);
542         return 0;
543 }
544
545 static int get_parent(const char *name, int len,
546                       unsigned char *result, int idx)
547 {
548         unsigned char sha1[20];
549         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
550         struct commit *commit;
551         struct commit_list *p;
552
553         if (ret)
554                 return ret;
555         commit = lookup_commit_reference(sha1);
556         if (!commit)
557                 return -1;
558         if (parse_commit(commit))
559                 return -1;
560         if (!idx) {
561                 hashcpy(result, commit->object.sha1);
562                 return 0;
563         }
564         p = commit->parents;
565         while (p) {
566                 if (!--idx) {
567                         hashcpy(result, p->item->object.sha1);
568                         return 0;
569                 }
570                 p = p->next;
571         }
572         return -1;
573 }
574
575 static int get_nth_ancestor(const char *name, int len,
576                             unsigned char *result, int generation)
577 {
578         unsigned char sha1[20];
579         struct commit *commit;
580         int ret;
581
582         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
583         if (ret)
584                 return ret;
585         commit = lookup_commit_reference(sha1);
586         if (!commit)
587                 return -1;
588
589         while (generation--) {
590                 if (parse_commit(commit) || !commit->parents)
591                         return -1;
592                 commit = commit->parents->item;
593         }
594         hashcpy(result, commit->object.sha1);
595         return 0;
596 }
597
598 struct object *peel_to_type(const char *name, int namelen,
599                             struct object *o, enum object_type expected_type)
600 {
601         if (name && !namelen)
602                 namelen = strlen(name);
603         while (1) {
604                 if (!o || (!o->parsed && !parse_object(o->sha1)))
605                         return NULL;
606                 if (expected_type == OBJ_ANY || o->type == expected_type)
607                         return o;
608                 if (o->type == OBJ_TAG)
609                         o = ((struct tag*) o)->tagged;
610                 else if (o->type == OBJ_COMMIT)
611                         o = &(((struct commit *) o)->tree->object);
612                 else {
613                         if (name)
614                                 error("%.*s: expected %s type, but the object "
615                                       "dereferences to %s type",
616                                       namelen, name, typename(expected_type),
617                                       typename(o->type));
618                         return NULL;
619                 }
620         }
621 }
622
623 static int peel_onion(const char *name, int len, unsigned char *sha1)
624 {
625         unsigned char outer[20];
626         const char *sp;
627         unsigned int expected_type = 0;
628         unsigned lookup_flags = 0;
629         struct object *o;
630
631         /*
632          * "ref^{type}" dereferences ref repeatedly until you cannot
633          * dereference anymore, or you get an object of given type,
634          * whichever comes first.  "ref^{}" means just dereference
635          * tags until you get a non-tag.  "ref^0" is a shorthand for
636          * "ref^{commit}".  "commit^{tree}" could be used to find the
637          * top-level tree of the given commit.
638          */
639         if (len < 4 || name[len-1] != '}')
640                 return -1;
641
642         for (sp = name + len - 1; name <= sp; sp--) {
643                 int ch = *sp;
644                 if (ch == '{' && name < sp && sp[-1] == '^')
645                         break;
646         }
647         if (sp <= name)
648                 return -1;
649
650         sp++; /* beginning of type name, or closing brace for empty */
651         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
652                 expected_type = OBJ_COMMIT;
653         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
654                 expected_type = OBJ_TREE;
655         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
656                 expected_type = OBJ_BLOB;
657         else if (!prefixcmp(sp, "object}"))
658                 expected_type = OBJ_ANY;
659         else if (sp[0] == '}')
660                 expected_type = OBJ_NONE;
661         else if (sp[0] == '/')
662                 expected_type = OBJ_COMMIT;
663         else
664                 return -1;
665
666         if (expected_type == OBJ_COMMIT)
667                 lookup_flags = GET_SHA1_COMMITTISH;
668         else if (expected_type == OBJ_TREE)
669                 lookup_flags = GET_SHA1_TREEISH;
670
671         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
672                 return -1;
673
674         o = parse_object(outer);
675         if (!o)
676                 return -1;
677         if (!expected_type) {
678                 o = deref_tag(o, name, sp - name - 2);
679                 if (!o || (!o->parsed && !parse_object(o->sha1)))
680                         return -1;
681                 hashcpy(sha1, o->sha1);
682                 return 0;
683         }
684
685         /*
686          * At this point, the syntax look correct, so
687          * if we do not get the needed object, we should
688          * barf.
689          */
690         o = peel_to_type(name, len, o, expected_type);
691         if (!o)
692                 return -1;
693
694         hashcpy(sha1, o->sha1);
695         if (sp[0] == '/') {
696                 /* "$commit^{/foo}" */
697                 char *prefix;
698                 int ret;
699                 struct commit_list *list = NULL;
700
701                 /*
702                  * $commit^{/}. Some regex implementation may reject.
703                  * We don't need regex anyway. '' pattern always matches.
704                  */
705                 if (sp[1] == '}')
706                         return 0;
707
708                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
709                 commit_list_insert((struct commit *)o, &list);
710                 ret = get_sha1_oneline(prefix, sha1, list);
711                 free(prefix);
712                 return ret;
713         }
714         return 0;
715 }
716
717 static int get_describe_name(const char *name, int len, unsigned char *sha1)
718 {
719         const char *cp;
720         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
721
722         for (cp = name + len - 1; name + 2 <= cp; cp--) {
723                 char ch = *cp;
724                 if (hexval(ch) & ~0377) {
725                         /* We must be looking at g in "SOMETHING-g"
726                          * for it to be describe output.
727                          */
728                         if (ch == 'g' && cp[-1] == '-') {
729                                 cp++;
730                                 len -= cp - name;
731                                 return get_short_sha1(cp, len, sha1, flags);
732                         }
733                 }
734         }
735         return -1;
736 }
737
738 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
739 {
740         int ret, has_suffix;
741         const char *cp;
742
743         /*
744          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
745          */
746         has_suffix = 0;
747         for (cp = name + len - 1; name <= cp; cp--) {
748                 int ch = *cp;
749                 if ('0' <= ch && ch <= '9')
750                         continue;
751                 if (ch == '~' || ch == '^')
752                         has_suffix = ch;
753                 break;
754         }
755
756         if (has_suffix) {
757                 int num = 0;
758                 int len1 = cp - name;
759                 cp++;
760                 while (cp < name + len)
761                         num = num * 10 + *cp++ - '0';
762                 if (!num && len1 == len - 1)
763                         num = 1;
764                 if (has_suffix == '^')
765                         return get_parent(name, len1, sha1, num);
766                 /* else if (has_suffix == '~') -- goes without saying */
767                 return get_nth_ancestor(name, len1, sha1, num);
768         }
769
770         ret = peel_onion(name, len, sha1);
771         if (!ret)
772                 return 0;
773
774         ret = get_sha1_basic(name, len, sha1);
775         if (!ret)
776                 return 0;
777
778         /* It could be describe output that is "SOMETHING-gXXXX" */
779         ret = get_describe_name(name, len, sha1);
780         if (!ret)
781                 return 0;
782
783         return get_short_sha1(name, len, sha1, lookup_flags);
784 }
785
786 /*
787  * This interprets names like ':/Initial revision of "git"' by searching
788  * through history and returning the first commit whose message starts
789  * the given regular expression.
790  *
791  * For future extension, ':/!' is reserved. If you want to match a message
792  * beginning with a '!', you have to repeat the exclamation mark.
793  */
794 #define ONELINE_SEEN (1u<<20)
795
796 static int handle_one_ref(const char *path,
797                 const unsigned char *sha1, int flag, void *cb_data)
798 {
799         struct commit_list **list = cb_data;
800         struct object *object = parse_object(sha1);
801         if (!object)
802                 return 0;
803         if (object->type == OBJ_TAG) {
804                 object = deref_tag(object, path, strlen(path));
805                 if (!object)
806                         return 0;
807         }
808         if (object->type != OBJ_COMMIT)
809                 return 0;
810         commit_list_insert_by_date((struct commit *)object, list);
811         return 0;
812 }
813
814 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
815                             struct commit_list *list)
816 {
817         struct commit_list *backup = NULL, *l;
818         int found = 0;
819         regex_t regex;
820
821         if (prefix[0] == '!') {
822                 if (prefix[1] != '!')
823                         die ("Invalid search pattern: %s", prefix);
824                 prefix++;
825         }
826
827         if (regcomp(&regex, prefix, REG_EXTENDED))
828                 die("Invalid search pattern: %s", prefix);
829
830         for (l = list; l; l = l->next) {
831                 l->item->object.flags |= ONELINE_SEEN;
832                 commit_list_insert(l->item, &backup);
833         }
834         while (list) {
835                 char *p, *to_free = NULL;
836                 struct commit *commit;
837                 enum object_type type;
838                 unsigned long size;
839                 int matches;
840
841                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
842                 if (!parse_object(commit->object.sha1))
843                         continue;
844                 if (commit->buffer)
845                         p = commit->buffer;
846                 else {
847                         p = read_sha1_file(commit->object.sha1, &type, &size);
848                         if (!p)
849                                 continue;
850                         to_free = p;
851                 }
852
853                 p = strstr(p, "\n\n");
854                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
855                 free(to_free);
856
857                 if (matches) {
858                         hashcpy(sha1, commit->object.sha1);
859                         found = 1;
860                         break;
861                 }
862         }
863         regfree(&regex);
864         free_commit_list(list);
865         for (l = backup; l; l = l->next)
866                 clear_commit_marks(l->item, ONELINE_SEEN);
867         free_commit_list(backup);
868         return found ? 0 : -1;
869 }
870
871 struct grab_nth_branch_switch_cbdata {
872         int remaining;
873         struct strbuf buf;
874 };
875
876 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
877                                   const char *email, unsigned long timestamp, int tz,
878                                   const char *message, void *cb_data)
879 {
880         struct grab_nth_branch_switch_cbdata *cb = cb_data;
881         const char *match = NULL, *target = NULL;
882         size_t len;
883
884         if (!prefixcmp(message, "checkout: moving from ")) {
885                 match = message + strlen("checkout: moving from ");
886                 target = strstr(match, " to ");
887         }
888
889         if (!match || !target)
890                 return 0;
891         if (--(cb->remaining) == 0) {
892                 len = target - match;
893                 strbuf_reset(&cb->buf);
894                 strbuf_add(&cb->buf, match, len);
895                 return 1; /* we are done */
896         }
897         return 0;
898 }
899
900 /*
901  * Parse @{-N} syntax, return the number of characters parsed
902  * if successful; otherwise signal an error with negative value.
903  */
904 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
905 {
906         long nth;
907         int retval;
908         struct grab_nth_branch_switch_cbdata cb;
909         const char *brace;
910         char *num_end;
911
912         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
913                 return -1;
914         brace = strchr(name, '}');
915         if (!brace)
916                 return -1;
917         nth = strtol(name + 3, &num_end, 10);
918         if (num_end != brace)
919                 return -1;
920         if (nth <= 0)
921                 return -1;
922         cb.remaining = nth;
923         strbuf_init(&cb.buf, 20);
924
925         retval = 0;
926         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
927                 strbuf_reset(buf);
928                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
929                 retval = brace - name + 1;
930         }
931
932         strbuf_release(&cb.buf);
933         return retval;
934 }
935
936 int get_sha1_mb(const char *name, unsigned char *sha1)
937 {
938         struct commit *one, *two;
939         struct commit_list *mbs;
940         unsigned char sha1_tmp[20];
941         const char *dots;
942         int st;
943
944         dots = strstr(name, "...");
945         if (!dots)
946                 return get_sha1(name, sha1);
947         if (dots == name)
948                 st = get_sha1("HEAD", sha1_tmp);
949         else {
950                 struct strbuf sb;
951                 strbuf_init(&sb, dots - name);
952                 strbuf_add(&sb, name, dots - name);
953                 st = get_sha1_committish(sb.buf, sha1_tmp);
954                 strbuf_release(&sb);
955         }
956         if (st)
957                 return st;
958         one = lookup_commit_reference_gently(sha1_tmp, 0);
959         if (!one)
960                 return -1;
961
962         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
963                 return -1;
964         two = lookup_commit_reference_gently(sha1_tmp, 0);
965         if (!two)
966                 return -1;
967         mbs = get_merge_bases(one, two, 1);
968         if (!mbs || mbs->next)
969                 st = -1;
970         else {
971                 st = 0;
972                 hashcpy(sha1, mbs->item->object.sha1);
973         }
974         free_commit_list(mbs);
975         return st;
976 }
977
978 /*
979  * This reads short-hand syntax that not only evaluates to a commit
980  * object name, but also can act as if the end user spelled the name
981  * of the branch from the command line.
982  *
983  * - "@{-N}" finds the name of the Nth previous branch we were on, and
984  *   places the name of the branch in the given buf and returns the
985  *   number of characters parsed if successful.
986  *
987  * - "<branch>@{upstream}" finds the name of the other ref that
988  *   <branch> is configured to merge with (missing <branch> defaults
989  *   to the current branch), and places the name of the branch in the
990  *   given buf and returns the number of characters parsed if
991  *   successful.
992  *
993  * If the input is not of the accepted format, it returns a negative
994  * number to signal an error.
995  *
996  * If the input was ok but there are not N branch switches in the
997  * reflog, it returns 0.
998  */
999 int interpret_branch_name(const char *name, struct strbuf *buf)
1000 {
1001         char *cp;
1002         struct branch *upstream;
1003         int namelen = strlen(name);
1004         int len = interpret_nth_prior_checkout(name, buf);
1005         int tmp_len;
1006
1007         if (!len)
1008                 return len; /* syntax Ok, not enough switches */
1009         if (0 < len && len == namelen)
1010                 return len; /* consumed all */
1011         else if (0 < len) {
1012                 /* we have extra data, which might need further processing */
1013                 struct strbuf tmp = STRBUF_INIT;
1014                 int used = buf->len;
1015                 int ret;
1016
1017                 strbuf_add(buf, name + len, namelen - len);
1018                 ret = interpret_branch_name(buf->buf, &tmp);
1019                 /* that data was not interpreted, remove our cruft */
1020                 if (ret < 0) {
1021                         strbuf_setlen(buf, used);
1022                         return len;
1023                 }
1024                 strbuf_reset(buf);
1025                 strbuf_addbuf(buf, &tmp);
1026                 strbuf_release(&tmp);
1027                 /* tweak for size of {-N} versus expanded ref name */
1028                 return ret - used + len;
1029         }
1030
1031         cp = strchr(name, '@');
1032         if (!cp)
1033                 return -1;
1034         tmp_len = upstream_mark(cp, namelen - (cp - name));
1035         if (!tmp_len)
1036                 return -1;
1037         len = cp + tmp_len - name;
1038         cp = xstrndup(name, cp - name);
1039         upstream = branch_get(*cp ? cp : NULL);
1040         /*
1041          * Upstream can be NULL only if cp refers to HEAD and HEAD
1042          * points to something different than a branch.
1043          */
1044         if (!upstream)
1045                 die(_("HEAD does not point to a branch"));
1046         if (!upstream->merge || !upstream->merge[0]->dst) {
1047                 if (!ref_exists(upstream->refname))
1048                         die(_("No such branch: '%s'"), cp);
1049                 if (!upstream->merge) {
1050                         die(_("No upstream configured for branch '%s'"),
1051                                 upstream->name);
1052                 }
1053                 die(
1054                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1055                         upstream->merge[0]->src);
1056         }
1057         free(cp);
1058         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1059         strbuf_reset(buf);
1060         strbuf_addstr(buf, cp);
1061         free(cp);
1062         return len;
1063 }
1064
1065 int strbuf_branchname(struct strbuf *sb, const char *name)
1066 {
1067         int len = strlen(name);
1068         int used = interpret_branch_name(name, sb);
1069
1070         if (used == len)
1071                 return 0;
1072         if (used < 0)
1073                 used = 0;
1074         strbuf_add(sb, name + used, len - used);
1075         return len;
1076 }
1077
1078 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1079 {
1080         strbuf_branchname(sb, name);
1081         if (name[0] == '-')
1082                 return -1;
1083         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1084         return check_refname_format(sb->buf, 0);
1085 }
1086
1087 /*
1088  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1089  * notably "xyz^" for "parent of xyz"
1090  */
1091 int get_sha1(const char *name, unsigned char *sha1)
1092 {
1093         struct object_context unused;
1094         return get_sha1_with_context(name, 0, sha1, &unused);
1095 }
1096
1097 /*
1098  * Many callers know that the user meant to name a committish by
1099  * syntactical positions where the object name appears.  Calling this
1100  * function allows the machinery to disambiguate shorter-than-unique
1101  * abbreviated object names between committish and others.
1102  *
1103  * Note that this does NOT error out when the named object is not a
1104  * committish. It is merely to give a hint to the disambiguation
1105  * machinery.
1106  */
1107 int get_sha1_committish(const char *name, unsigned char *sha1)
1108 {
1109         struct object_context unused;
1110         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1111                                      sha1, &unused);
1112 }
1113
1114 int get_sha1_treeish(const char *name, unsigned char *sha1)
1115 {
1116         struct object_context unused;
1117         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1118                                      sha1, &unused);
1119 }
1120
1121 int get_sha1_commit(const char *name, unsigned char *sha1)
1122 {
1123         struct object_context unused;
1124         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1125                                      sha1, &unused);
1126 }
1127
1128 int get_sha1_tree(const char *name, unsigned char *sha1)
1129 {
1130         struct object_context unused;
1131         return get_sha1_with_context(name, GET_SHA1_TREE,
1132                                      sha1, &unused);
1133 }
1134
1135 int get_sha1_blob(const char *name, unsigned char *sha1)
1136 {
1137         struct object_context unused;
1138         return get_sha1_with_context(name, GET_SHA1_BLOB,
1139                                      sha1, &unused);
1140 }
1141
1142 /* Must be called only when object_name:filename doesn't exist. */
1143 static void diagnose_invalid_sha1_path(const char *prefix,
1144                                        const char *filename,
1145                                        const unsigned char *tree_sha1,
1146                                        const char *object_name,
1147                                        int object_name_len)
1148 {
1149         struct stat st;
1150         unsigned char sha1[20];
1151         unsigned mode;
1152
1153         if (!prefix)
1154                 prefix = "";
1155
1156         if (!lstat(filename, &st))
1157                 die("Path '%s' exists on disk, but not in '%.*s'.",
1158                     filename, object_name_len, object_name);
1159         if (errno == ENOENT || errno == ENOTDIR) {
1160                 char *fullname = xmalloc(strlen(filename)
1161                                              + strlen(prefix) + 1);
1162                 strcpy(fullname, prefix);
1163                 strcat(fullname, filename);
1164
1165                 if (!get_tree_entry(tree_sha1, fullname,
1166                                     sha1, &mode)) {
1167                         die("Path '%s' exists, but not '%s'.\n"
1168                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1169                             fullname,
1170                             filename,
1171                             object_name_len, object_name,
1172                             fullname,
1173                             object_name_len, object_name,
1174                             filename);
1175                 }
1176                 die("Path '%s' does not exist in '%.*s'",
1177                     filename, object_name_len, object_name);
1178         }
1179 }
1180
1181 /* Must be called only when :stage:filename doesn't exist. */
1182 static void diagnose_invalid_index_path(int stage,
1183                                         const char *prefix,
1184                                         const char *filename)
1185 {
1186         struct stat st;
1187         struct cache_entry *ce;
1188         int pos;
1189         unsigned namelen = strlen(filename);
1190         unsigned fullnamelen;
1191         char *fullname;
1192
1193         if (!prefix)
1194                 prefix = "";
1195
1196         /* Wrong stage number? */
1197         pos = cache_name_pos(filename, namelen);
1198         if (pos < 0)
1199                 pos = -pos - 1;
1200         if (pos < active_nr) {
1201                 ce = active_cache[pos];
1202                 if (ce_namelen(ce) == namelen &&
1203                     !memcmp(ce->name, filename, namelen))
1204                         die("Path '%s' is in the index, but not at stage %d.\n"
1205                             "Did you mean ':%d:%s'?",
1206                             filename, stage,
1207                             ce_stage(ce), filename);
1208         }
1209
1210         /* Confusion between relative and absolute filenames? */
1211         fullnamelen = namelen + strlen(prefix);
1212         fullname = xmalloc(fullnamelen + 1);
1213         strcpy(fullname, prefix);
1214         strcat(fullname, filename);
1215         pos = cache_name_pos(fullname, fullnamelen);
1216         if (pos < 0)
1217                 pos = -pos - 1;
1218         if (pos < active_nr) {
1219                 ce = active_cache[pos];
1220                 if (ce_namelen(ce) == fullnamelen &&
1221                     !memcmp(ce->name, fullname, fullnamelen))
1222                         die("Path '%s' is in the index, but not '%s'.\n"
1223                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1224                             fullname, filename,
1225                             ce_stage(ce), fullname,
1226                             ce_stage(ce), filename);
1227         }
1228
1229         if (!lstat(filename, &st))
1230                 die("Path '%s' exists on disk, but not in the index.", filename);
1231         if (errno == ENOENT || errno == ENOTDIR)
1232                 die("Path '%s' does not exist (neither on disk nor in the index).",
1233                     filename);
1234
1235         free(fullname);
1236 }
1237
1238
1239 static char *resolve_relative_path(const char *rel)
1240 {
1241         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1242                 return NULL;
1243
1244         if (!startup_info)
1245                 die("BUG: startup_info struct is not initialized.");
1246
1247         if (!is_inside_work_tree())
1248                 die("relative path syntax can't be used outside working tree.");
1249
1250         /* die() inside prefix_path() if resolved path is outside worktree */
1251         return prefix_path(startup_info->prefix,
1252                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1253                            rel);
1254 }
1255
1256 static int get_sha1_with_context_1(const char *name,
1257                                    unsigned flags,
1258                                    const char *prefix,
1259                                    unsigned char *sha1,
1260                                    struct object_context *oc)
1261 {
1262         int ret, bracket_depth;
1263         int namelen = strlen(name);
1264         const char *cp;
1265         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1266
1267         memset(oc, 0, sizeof(*oc));
1268         oc->mode = S_IFINVALID;
1269         ret = get_sha1_1(name, namelen, sha1, flags);
1270         if (!ret)
1271                 return ret;
1272         /*
1273          * sha1:path --> object name of path in ent sha1
1274          * :path -> object name of absolute path in index
1275          * :./path -> object name of path relative to cwd in index
1276          * :[0-3]:path -> object name of path in index at stage
1277          * :/foo -> recent commit matching foo
1278          */
1279         if (name[0] == ':') {
1280                 int stage = 0;
1281                 struct cache_entry *ce;
1282                 char *new_path = NULL;
1283                 int pos;
1284                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1285                         struct commit_list *list = NULL;
1286                         for_each_ref(handle_one_ref, &list);
1287                         return get_sha1_oneline(name + 2, sha1, list);
1288                 }
1289                 if (namelen < 3 ||
1290                     name[2] != ':' ||
1291                     name[1] < '0' || '3' < name[1])
1292                         cp = name + 1;
1293                 else {
1294                         stage = name[1] - '0';
1295                         cp = name + 3;
1296                 }
1297                 new_path = resolve_relative_path(cp);
1298                 if (!new_path) {
1299                         namelen = namelen - (cp - name);
1300                 } else {
1301                         cp = new_path;
1302                         namelen = strlen(cp);
1303                 }
1304
1305                 strncpy(oc->path, cp,
1306                         sizeof(oc->path));
1307                 oc->path[sizeof(oc->path)-1] = '\0';
1308
1309                 if (!active_cache)
1310                         read_cache();
1311                 pos = cache_name_pos(cp, namelen);
1312                 if (pos < 0)
1313                         pos = -pos - 1;
1314                 while (pos < active_nr) {
1315                         ce = active_cache[pos];
1316                         if (ce_namelen(ce) != namelen ||
1317                             memcmp(ce->name, cp, namelen))
1318                                 break;
1319                         if (ce_stage(ce) == stage) {
1320                                 hashcpy(sha1, ce->sha1);
1321                                 oc->mode = ce->ce_mode;
1322                                 free(new_path);
1323                                 return 0;
1324                         }
1325                         pos++;
1326                 }
1327                 if (only_to_die && name[1] && name[1] != '/')
1328                         diagnose_invalid_index_path(stage, prefix, cp);
1329                 free(new_path);
1330                 return -1;
1331         }
1332         for (cp = name, bracket_depth = 0; *cp; cp++) {
1333                 if (*cp == '{')
1334                         bracket_depth++;
1335                 else if (bracket_depth && *cp == '}')
1336                         bracket_depth--;
1337                 else if (!bracket_depth && *cp == ':')
1338                         break;
1339         }
1340         if (*cp == ':') {
1341                 unsigned char tree_sha1[20];
1342                 int len = cp - name;
1343                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1344                         const char *filename = cp+1;
1345                         char *new_filename = NULL;
1346
1347                         new_filename = resolve_relative_path(filename);
1348                         if (new_filename)
1349                                 filename = new_filename;
1350                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1351                         if (ret && only_to_die) {
1352                                 diagnose_invalid_sha1_path(prefix, filename,
1353                                                            tree_sha1,
1354                                                            name, len);
1355                         }
1356                         hashcpy(oc->tree, tree_sha1);
1357                         strncpy(oc->path, filename,
1358                                 sizeof(oc->path));
1359                         oc->path[sizeof(oc->path)-1] = '\0';
1360
1361                         free(new_filename);
1362                         return ret;
1363                 } else {
1364                         if (only_to_die)
1365                                 die("Invalid object name '%.*s'.", len, name);
1366                 }
1367         }
1368         return ret;
1369 }
1370
1371 /*
1372  * Call this function when you know "name" given by the end user must
1373  * name an object but it doesn't; the function _may_ die with a better
1374  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1375  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1376  * you have a chance to diagnose the error further.
1377  */
1378 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1379 {
1380         struct object_context oc;
1381         unsigned char sha1[20];
1382         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1383 }
1384
1385 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1386 {
1387         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1388 }