Implement first draft of DNS-SD support in git-daemon
[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 satisfy 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 int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
347 {
348         char hex_pfx[40];
349         unsigned char bin_pfx[20];
350         struct disambiguate_state ds;
351         int len = strlen(prefix);
352
353         if (len < MINIMUM_ABBREV || len > 40)
354                 return -1;
355         if (prepare_prefixes(prefix, len, bin_pfx, hex_pfx) < 0)
356                 return -1;
357
358         prepare_alt_odb();
359
360         memset(&ds, 0, sizeof(ds));
361         ds.always_call_fn = 1;
362         ds.cb_data = cb_data;
363         ds.fn = fn;
364
365         find_short_object_filename(len, hex_pfx, &ds);
366         find_short_packed_object(len, bin_pfx, &ds);
367         return ds.ambiguous;
368 }
369
370 const char *find_unique_abbrev(const unsigned char *sha1, int len)
371 {
372         int status, exists;
373         static char hex[41];
374
375         exists = has_sha1_file(sha1);
376         memcpy(hex, sha1_to_hex(sha1), 40);
377         if (len == 40 || !len)
378                 return hex;
379         while (len < 40) {
380                 unsigned char sha1_ret[20];
381                 status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
382                 if (exists
383                     ? !status
384                     : status == SHORT_NAME_NOT_FOUND) {
385                         hex[len] = 0;
386                         return hex;
387                 }
388                 len++;
389         }
390         return hex;
391 }
392
393 static int ambiguous_path(const char *path, int len)
394 {
395         int slash = 1;
396         int cnt;
397
398         for (cnt = 0; cnt < len; cnt++) {
399                 switch (*path++) {
400                 case '\0':
401                         break;
402                 case '/':
403                         if (slash)
404                                 break;
405                         slash = 1;
406                         continue;
407                 case '.':
408                         continue;
409                 default:
410                         slash = 0;
411                         continue;
412                 }
413                 break;
414         }
415         return slash;
416 }
417
418 static inline int upstream_mark(const char *string, int len)
419 {
420         const char *suffix[] = { "@{upstream}", "@{u}" };
421         int i;
422
423         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
424                 int suffix_len = strlen(suffix[i]);
425                 if (suffix_len <= len
426                     && !memcmp(string, suffix[i], suffix_len))
427                         return suffix_len;
428         }
429         return 0;
430 }
431
432 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
433 static int interpret_nth_prior_checkout(const char *name, int namelen, struct strbuf *buf);
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.objectNameWarning false\"");
448         unsigned char tmp_sha1[20];
449         char *real_ref = NULL;
450         int refs_found = 0;
451         int at, reflog_len, nth_prior = 0;
452
453         if (len == 40 && !get_sha1_hex(str, sha1)) {
454                 if (warn_ambiguous_refs && warn_on_object_refname_ambiguity) {
455                         refs_found = dwim_ref(str, len, tmp_sha1, &real_ref);
456                         if (refs_found > 0) {
457                                 warning(warn_msg, len, str);
458                                 if (advice_object_name_warning)
459                                         fprintf(stderr, "%s\n", _(object_name_msg));
460                         }
461                         free(real_ref);
462                 }
463                 return 0;
464         }
465
466         /* basic@{time or number or -number} format to query ref-log */
467         reflog_len = at = 0;
468         if (len && str[len-1] == '}') {
469                 for (at = len-4; at >= 0; at--) {
470                         if (str[at] == '@' && str[at+1] == '{') {
471                                 if (str[at+2] == '-') {
472                                         if (at != 0)
473                                                 /* @{-N} not at start */
474                                                 return -1;
475                                         nth_prior = 1;
476                                         continue;
477                                 }
478                                 if (!upstream_mark(str + at, len - at)) {
479                                         reflog_len = (len-1) - (at+2);
480                                         len = at;
481                                 }
482                                 break;
483                         }
484                 }
485         }
486
487         /* Accept only unambiguous ref paths. */
488         if (len && ambiguous_path(str, len))
489                 return -1;
490
491         if (nth_prior) {
492                 struct strbuf buf = STRBUF_INIT;
493                 int detached;
494
495                 if (interpret_nth_prior_checkout(str, len, &buf) > 0) {
496                         detached = (buf.len == 40 && !get_sha1_hex(buf.buf, sha1));
497                         strbuf_release(&buf);
498                         if (detached)
499                                 return 0;
500                 }
501         }
502
503         if (!len && reflog_len)
504                 /* allow "@{...}" to mean the current branch reflog */
505                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
506         else if (reflog_len)
507                 refs_found = dwim_log(str, len, sha1, &real_ref);
508         else
509                 refs_found = dwim_ref(str, len, sha1, &real_ref);
510
511         if (!refs_found)
512                 return -1;
513
514         if (warn_ambiguous_refs &&
515             (refs_found > 1 ||
516              !get_short_sha1(str, len, tmp_sha1, GET_SHA1_QUIETLY)))
517                 warning(warn_msg, len, str);
518
519         if (reflog_len) {
520                 int nth, i;
521                 unsigned long at_time;
522                 unsigned long co_time;
523                 int co_tz, co_cnt;
524
525                 /* Is it asking for N-th entry, or approxidate? */
526                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
527                         char ch = str[at+2+i];
528                         if ('0' <= ch && ch <= '9')
529                                 nth = nth * 10 + ch - '0';
530                         else
531                                 nth = -1;
532                 }
533                 if (100000000 <= nth) {
534                         at_time = nth;
535                         nth = -1;
536                 } else if (0 <= nth)
537                         at_time = 0;
538                 else {
539                         int errors = 0;
540                         char *tmp = xstrndup(str + at + 2, reflog_len);
541                         at_time = approxidate_careful(tmp, &errors);
542                         free(tmp);
543                         if (errors)
544                                 return -1;
545                 }
546                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
547                                 &co_time, &co_tz, &co_cnt)) {
548                         if (!len) {
549                                 if (starts_with(real_ref, "refs/heads/")) {
550                                         str = real_ref + 11;
551                                         len = strlen(real_ref + 11);
552                                 } else {
553                                         /* detached HEAD */
554                                         str = "HEAD";
555                                         len = 4;
556                                 }
557                         }
558                         if (at_time)
559                                 warning("Log for '%.*s' only goes "
560                                         "back to %s.", len, str,
561                                         show_date(co_time, co_tz, DATE_RFC2822));
562                         else {
563                                 die("Log for '%.*s' only has %d entries.",
564                                     len, str, co_cnt);
565                         }
566                 }
567         }
568
569         free(real_ref);
570         return 0;
571 }
572
573 static int get_parent(const char *name, int len,
574                       unsigned char *result, int idx)
575 {
576         unsigned char sha1[20];
577         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
578         struct commit *commit;
579         struct commit_list *p;
580
581         if (ret)
582                 return ret;
583         commit = lookup_commit_reference(sha1);
584         if (parse_commit(commit))
585                 return -1;
586         if (!idx) {
587                 hashcpy(result, commit->object.sha1);
588                 return 0;
589         }
590         p = commit->parents;
591         while (p) {
592                 if (!--idx) {
593                         hashcpy(result, p->item->object.sha1);
594                         return 0;
595                 }
596                 p = p->next;
597         }
598         return -1;
599 }
600
601 static int get_nth_ancestor(const char *name, int len,
602                             unsigned char *result, int generation)
603 {
604         unsigned char sha1[20];
605         struct commit *commit;
606         int ret;
607
608         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
609         if (ret)
610                 return ret;
611         commit = lookup_commit_reference(sha1);
612         if (!commit)
613                 return -1;
614
615         while (generation--) {
616                 if (parse_commit(commit) || !commit->parents)
617                         return -1;
618                 commit = commit->parents->item;
619         }
620         hashcpy(result, commit->object.sha1);
621         return 0;
622 }
623
624 struct object *peel_to_type(const char *name, int namelen,
625                             struct object *o, enum object_type expected_type)
626 {
627         if (name && !namelen)
628                 namelen = strlen(name);
629         while (1) {
630                 if (!o || (!o->parsed && !parse_object(o->sha1)))
631                         return NULL;
632                 if (expected_type == OBJ_ANY || o->type == expected_type)
633                         return o;
634                 if (o->type == OBJ_TAG)
635                         o = ((struct tag*) o)->tagged;
636                 else if (o->type == OBJ_COMMIT)
637                         o = &(((struct commit *) o)->tree->object);
638                 else {
639                         if (name)
640                                 error("%.*s: expected %s type, but the object "
641                                       "dereferences to %s type",
642                                       namelen, name, typename(expected_type),
643                                       typename(o->type));
644                         return NULL;
645                 }
646         }
647 }
648
649 static int peel_onion(const char *name, int len, unsigned char *sha1)
650 {
651         unsigned char outer[20];
652         const char *sp;
653         unsigned int expected_type = 0;
654         unsigned lookup_flags = 0;
655         struct object *o;
656
657         /*
658          * "ref^{type}" dereferences ref repeatedly until you cannot
659          * dereference anymore, or you get an object of given type,
660          * whichever comes first.  "ref^{}" means just dereference
661          * tags until you get a non-tag.  "ref^0" is a shorthand for
662          * "ref^{commit}".  "commit^{tree}" could be used to find the
663          * top-level tree of the given commit.
664          */
665         if (len < 4 || name[len-1] != '}')
666                 return -1;
667
668         for (sp = name + len - 1; name <= sp; sp--) {
669                 int ch = *sp;
670                 if (ch == '{' && name < sp && sp[-1] == '^')
671                         break;
672         }
673         if (sp <= name)
674                 return -1;
675
676         sp++; /* beginning of type name, or closing brace for empty */
677         if (starts_with(sp, "commit}"))
678                 expected_type = OBJ_COMMIT;
679         else if (starts_with(sp, "tag}"))
680                 expected_type = OBJ_TAG;
681         else if (starts_with(sp, "tree}"))
682                 expected_type = OBJ_TREE;
683         else if (starts_with(sp, "blob}"))
684                 expected_type = OBJ_BLOB;
685         else if (starts_with(sp, "object}"))
686                 expected_type = OBJ_ANY;
687         else if (sp[0] == '}')
688                 expected_type = OBJ_NONE;
689         else if (sp[0] == '/')
690                 expected_type = OBJ_COMMIT;
691         else
692                 return -1;
693
694         if (expected_type == OBJ_COMMIT)
695                 lookup_flags = GET_SHA1_COMMITTISH;
696         else if (expected_type == OBJ_TREE)
697                 lookup_flags = GET_SHA1_TREEISH;
698
699         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
700                 return -1;
701
702         o = parse_object(outer);
703         if (!o)
704                 return -1;
705         if (!expected_type) {
706                 o = deref_tag(o, name, sp - name - 2);
707                 if (!o || (!o->parsed && !parse_object(o->sha1)))
708                         return -1;
709                 hashcpy(sha1, o->sha1);
710                 return 0;
711         }
712
713         /*
714          * At this point, the syntax look correct, so
715          * if we do not get the needed object, we should
716          * barf.
717          */
718         o = peel_to_type(name, len, o, expected_type);
719         if (!o)
720                 return -1;
721
722         hashcpy(sha1, o->sha1);
723         if (sp[0] == '/') {
724                 /* "$commit^{/foo}" */
725                 char *prefix;
726                 int ret;
727                 struct commit_list *list = NULL;
728
729                 /*
730                  * $commit^{/}. Some regex implementation may reject.
731                  * We don't need regex anyway. '' pattern always matches.
732                  */
733                 if (sp[1] == '}')
734                         return 0;
735
736                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
737                 commit_list_insert((struct commit *)o, &list);
738                 ret = get_sha1_oneline(prefix, sha1, list);
739                 free(prefix);
740                 return ret;
741         }
742         return 0;
743 }
744
745 static int get_describe_name(const char *name, int len, unsigned char *sha1)
746 {
747         const char *cp;
748         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
749
750         for (cp = name + len - 1; name + 2 <= cp; cp--) {
751                 char ch = *cp;
752                 if (hexval(ch) & ~0377) {
753                         /* We must be looking at g in "SOMETHING-g"
754                          * for it to be describe output.
755                          */
756                         if (ch == 'g' && cp[-1] == '-') {
757                                 cp++;
758                                 len -= cp - name;
759                                 return get_short_sha1(cp, len, sha1, flags);
760                         }
761                 }
762         }
763         return -1;
764 }
765
766 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
767 {
768         int ret, has_suffix;
769         const char *cp;
770
771         /*
772          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
773          */
774         has_suffix = 0;
775         for (cp = name + len - 1; name <= cp; cp--) {
776                 int ch = *cp;
777                 if ('0' <= ch && ch <= '9')
778                         continue;
779                 if (ch == '~' || ch == '^')
780                         has_suffix = ch;
781                 break;
782         }
783
784         if (has_suffix) {
785                 int num = 0;
786                 int len1 = cp - name;
787                 cp++;
788                 while (cp < name + len)
789                         num = num * 10 + *cp++ - '0';
790                 if (!num && len1 == len - 1)
791                         num = 1;
792                 if (has_suffix == '^')
793                         return get_parent(name, len1, sha1, num);
794                 /* else if (has_suffix == '~') -- goes without saying */
795                 return get_nth_ancestor(name, len1, sha1, num);
796         }
797
798         ret = peel_onion(name, len, sha1);
799         if (!ret)
800                 return 0;
801
802         ret = get_sha1_basic(name, len, sha1);
803         if (!ret)
804                 return 0;
805
806         /* It could be describe output that is "SOMETHING-gXXXX" */
807         ret = get_describe_name(name, len, sha1);
808         if (!ret)
809                 return 0;
810
811         return get_short_sha1(name, len, sha1, lookup_flags);
812 }
813
814 /*
815  * This interprets names like ':/Initial revision of "git"' by searching
816  * through history and returning the first commit whose message starts
817  * the given regular expression.
818  *
819  * For future extension, ':/!' is reserved. If you want to match a message
820  * beginning with a '!', you have to repeat the exclamation mark.
821  */
822
823 /* Remember to update object flag allocation in object.h */
824 #define ONELINE_SEEN (1u<<20)
825
826 static int handle_one_ref(const char *path,
827                 const unsigned char *sha1, int flag, void *cb_data)
828 {
829         struct commit_list **list = cb_data;
830         struct object *object = parse_object(sha1);
831         if (!object)
832                 return 0;
833         if (object->type == OBJ_TAG) {
834                 object = deref_tag(object, path, strlen(path));
835                 if (!object)
836                         return 0;
837         }
838         if (object->type != OBJ_COMMIT)
839                 return 0;
840         commit_list_insert_by_date((struct commit *)object, list);
841         return 0;
842 }
843
844 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
845                             struct commit_list *list)
846 {
847         struct commit_list *backup = NULL, *l;
848         int found = 0;
849         regex_t regex;
850
851         if (prefix[0] == '!') {
852                 if (prefix[1] != '!')
853                         die ("Invalid search pattern: %s", prefix);
854                 prefix++;
855         }
856
857         if (regcomp(&regex, prefix, REG_EXTENDED))
858                 die("Invalid search pattern: %s", prefix);
859
860         for (l = list; l; l = l->next) {
861                 l->item->object.flags |= ONELINE_SEEN;
862                 commit_list_insert(l->item, &backup);
863         }
864         while (list) {
865                 const char *p, *buf;
866                 struct commit *commit;
867                 int matches;
868
869                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
870                 if (!parse_object(commit->object.sha1))
871                         continue;
872                 buf = get_commit_buffer(commit, NULL);
873                 p = strstr(buf, "\n\n");
874                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
875                 unuse_commit_buffer(commit, buf);
876
877                 if (matches) {
878                         hashcpy(sha1, commit->object.sha1);
879                         found = 1;
880                         break;
881                 }
882         }
883         regfree(&regex);
884         free_commit_list(list);
885         for (l = backup; l; l = l->next)
886                 clear_commit_marks(l->item, ONELINE_SEEN);
887         free_commit_list(backup);
888         return found ? 0 : -1;
889 }
890
891 struct grab_nth_branch_switch_cbdata {
892         int remaining;
893         struct strbuf buf;
894 };
895
896 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
897                                   const char *email, unsigned long timestamp, int tz,
898                                   const char *message, void *cb_data)
899 {
900         struct grab_nth_branch_switch_cbdata *cb = cb_data;
901         const char *match = NULL, *target = NULL;
902         size_t len;
903
904         if (skip_prefix(message, "checkout: moving from ", &match))
905                 target = strstr(match, " to ");
906
907         if (!match || !target)
908                 return 0;
909         if (--(cb->remaining) == 0) {
910                 len = target - match;
911                 strbuf_reset(&cb->buf);
912                 strbuf_add(&cb->buf, match, len);
913                 return 1; /* we are done */
914         }
915         return 0;
916 }
917
918 /*
919  * Parse @{-N} syntax, return the number of characters parsed
920  * if successful; otherwise signal an error with negative value.
921  */
922 static int interpret_nth_prior_checkout(const char *name, int namelen,
923                                         struct strbuf *buf)
924 {
925         long nth;
926         int retval;
927         struct grab_nth_branch_switch_cbdata cb;
928         const char *brace;
929         char *num_end;
930
931         if (namelen < 4)
932                 return -1;
933         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
934                 return -1;
935         brace = memchr(name, '}', namelen);
936         if (!brace)
937                 return -1;
938         nth = strtol(name + 3, &num_end, 10);
939         if (num_end != brace)
940                 return -1;
941         if (nth <= 0)
942                 return -1;
943         cb.remaining = nth;
944         strbuf_init(&cb.buf, 20);
945
946         retval = 0;
947         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
948                 strbuf_reset(buf);
949                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
950                 retval = brace - name + 1;
951         }
952
953         strbuf_release(&cb.buf);
954         return retval;
955 }
956
957 int get_sha1_mb(const char *name, unsigned char *sha1)
958 {
959         struct commit *one, *two;
960         struct commit_list *mbs;
961         unsigned char sha1_tmp[20];
962         const char *dots;
963         int st;
964
965         dots = strstr(name, "...");
966         if (!dots)
967                 return get_sha1(name, sha1);
968         if (dots == name)
969                 st = get_sha1("HEAD", sha1_tmp);
970         else {
971                 struct strbuf sb;
972                 strbuf_init(&sb, dots - name);
973                 strbuf_add(&sb, name, dots - name);
974                 st = get_sha1_committish(sb.buf, sha1_tmp);
975                 strbuf_release(&sb);
976         }
977         if (st)
978                 return st;
979         one = lookup_commit_reference_gently(sha1_tmp, 0);
980         if (!one)
981                 return -1;
982
983         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
984                 return -1;
985         two = lookup_commit_reference_gently(sha1_tmp, 0);
986         if (!two)
987                 return -1;
988         mbs = get_merge_bases(one, two, 1);
989         if (!mbs || mbs->next)
990                 st = -1;
991         else {
992                 st = 0;
993                 hashcpy(sha1, mbs->item->object.sha1);
994         }
995         free_commit_list(mbs);
996         return st;
997 }
998
999 /* parse @something syntax, when 'something' is not {.*} */
1000 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1001 {
1002         const char *next;
1003
1004         if (len || name[1] == '{')
1005                 return -1;
1006
1007         /* make sure it's a single @, or @@{.*}, not @foo */
1008         next = memchr(name + len + 1, '@', namelen - len - 1);
1009         if (next && next[1] != '{')
1010                 return -1;
1011         if (!next)
1012                 next = name + namelen;
1013         if (next != name + 1)
1014                 return -1;
1015
1016         strbuf_reset(buf);
1017         strbuf_add(buf, "HEAD", 4);
1018         return 1;
1019 }
1020
1021 static int reinterpret(const char *name, int namelen, int len, struct strbuf *buf)
1022 {
1023         /* we have extra data, which might need further processing */
1024         struct strbuf tmp = STRBUF_INIT;
1025         int used = buf->len;
1026         int ret;
1027
1028         strbuf_add(buf, name + len, namelen - len);
1029         ret = interpret_branch_name(buf->buf, buf->len, &tmp);
1030         /* that data was not interpreted, remove our cruft */
1031         if (ret < 0) {
1032                 strbuf_setlen(buf, used);
1033                 return len;
1034         }
1035         strbuf_reset(buf);
1036         strbuf_addbuf(buf, &tmp);
1037         strbuf_release(&tmp);
1038         /* tweak for size of {-N} versus expanded ref name */
1039         return ret - used + len;
1040 }
1041
1042 static void set_shortened_ref(struct strbuf *buf, const char *ref)
1043 {
1044         char *s = shorten_unambiguous_ref(ref, 0);
1045         strbuf_reset(buf);
1046         strbuf_addstr(buf, s);
1047         free(s);
1048 }
1049
1050 static const char *get_upstream_branch(const char *branch_buf, int len)
1051 {
1052         char *branch = xstrndup(branch_buf, len);
1053         struct branch *upstream = branch_get(*branch ? branch : NULL);
1054
1055         /*
1056          * Upstream can be NULL only if branch refers to HEAD and HEAD
1057          * points to something different than a branch.
1058          */
1059         if (!upstream)
1060                 die(_("HEAD does not point to a branch"));
1061         if (!upstream->merge || !upstream->merge[0]->dst) {
1062                 if (!ref_exists(upstream->refname))
1063                         die(_("No such branch: '%s'"), branch);
1064                 if (!upstream->merge) {
1065                         die(_("No upstream configured for branch '%s'"),
1066                                 upstream->name);
1067                 }
1068                 die(
1069                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1070                         upstream->merge[0]->src);
1071         }
1072         free(branch);
1073
1074         return upstream->merge[0]->dst;
1075 }
1076
1077 static int interpret_upstream_mark(const char *name, int namelen,
1078                                    int at, struct strbuf *buf)
1079 {
1080         int len;
1081
1082         len = upstream_mark(name + at, namelen - at);
1083         if (!len)
1084                 return -1;
1085
1086         if (memchr(name, ':', at))
1087                 return -1;
1088
1089         set_shortened_ref(buf, get_upstream_branch(name, at));
1090         return len + at;
1091 }
1092
1093 /*
1094  * This reads short-hand syntax that not only evaluates to a commit
1095  * object name, but also can act as if the end user spelled the name
1096  * of the branch from the command line.
1097  *
1098  * - "@{-N}" finds the name of the Nth previous branch we were on, and
1099  *   places the name of the branch in the given buf and returns the
1100  *   number of characters parsed if successful.
1101  *
1102  * - "<branch>@{upstream}" finds the name of the other ref that
1103  *   <branch> is configured to merge with (missing <branch> defaults
1104  *   to the current branch), and places the name of the branch in the
1105  *   given buf and returns the number of characters parsed if
1106  *   successful.
1107  *
1108  * If the input is not of the accepted format, it returns a negative
1109  * number to signal an error.
1110  *
1111  * If the input was ok but there are not N branch switches in the
1112  * reflog, it returns 0.
1113  */
1114 int interpret_branch_name(const char *name, int namelen, struct strbuf *buf)
1115 {
1116         char *at;
1117         const char *start;
1118         int len = interpret_nth_prior_checkout(name, namelen, buf);
1119
1120         if (!namelen)
1121                 namelen = strlen(name);
1122
1123         if (!len) {
1124                 return len; /* syntax Ok, not enough switches */
1125         } else if (len > 0) {
1126                 if (len == namelen)
1127                         return len; /* consumed all */
1128                 else
1129                         return reinterpret(name, namelen, len, buf);
1130         }
1131
1132         for (start = name;
1133              (at = memchr(start, '@', namelen - (start - name)));
1134              start = at + 1) {
1135
1136                 len = interpret_empty_at(name, namelen, at - name, buf);
1137                 if (len > 0)
1138                         return reinterpret(name, namelen, len, buf);
1139
1140                 len = interpret_upstream_mark(name, namelen, at - name, buf);
1141                 if (len > 0)
1142                         return len;
1143         }
1144
1145         return -1;
1146 }
1147
1148 int strbuf_branchname(struct strbuf *sb, const char *name)
1149 {
1150         int len = strlen(name);
1151         int used = interpret_branch_name(name, len, sb);
1152
1153         if (used == len)
1154                 return 0;
1155         if (used < 0)
1156                 used = 0;
1157         strbuf_add(sb, name + used, len - used);
1158         return len;
1159 }
1160
1161 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1162 {
1163         strbuf_branchname(sb, name);
1164         if (name[0] == '-')
1165                 return -1;
1166         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1167         return check_refname_format(sb->buf, 0);
1168 }
1169
1170 /*
1171  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1172  * notably "xyz^" for "parent of xyz"
1173  */
1174 int get_sha1(const char *name, unsigned char *sha1)
1175 {
1176         struct object_context unused;
1177         return get_sha1_with_context(name, 0, sha1, &unused);
1178 }
1179
1180 /*
1181  * Many callers know that the user meant to name a commit-ish by
1182  * syntactical positions where the object name appears.  Calling this
1183  * function allows the machinery to disambiguate shorter-than-unique
1184  * abbreviated object names between commit-ish and others.
1185  *
1186  * Note that this does NOT error out when the named object is not a
1187  * commit-ish. It is merely to give a hint to the disambiguation
1188  * machinery.
1189  */
1190 int get_sha1_committish(const char *name, unsigned char *sha1)
1191 {
1192         struct object_context unused;
1193         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1194                                      sha1, &unused);
1195 }
1196
1197 int get_sha1_treeish(const char *name, unsigned char *sha1)
1198 {
1199         struct object_context unused;
1200         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1201                                      sha1, &unused);
1202 }
1203
1204 int get_sha1_commit(const char *name, unsigned char *sha1)
1205 {
1206         struct object_context unused;
1207         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1208                                      sha1, &unused);
1209 }
1210
1211 int get_sha1_tree(const char *name, unsigned char *sha1)
1212 {
1213         struct object_context unused;
1214         return get_sha1_with_context(name, GET_SHA1_TREE,
1215                                      sha1, &unused);
1216 }
1217
1218 int get_sha1_blob(const char *name, unsigned char *sha1)
1219 {
1220         struct object_context unused;
1221         return get_sha1_with_context(name, GET_SHA1_BLOB,
1222                                      sha1, &unused);
1223 }
1224
1225 /* Must be called only when object_name:filename doesn't exist. */
1226 static void diagnose_invalid_sha1_path(const char *prefix,
1227                                        const char *filename,
1228                                        const unsigned char *tree_sha1,
1229                                        const char *object_name,
1230                                        int object_name_len)
1231 {
1232         struct stat st;
1233         unsigned char sha1[20];
1234         unsigned mode;
1235
1236         if (!prefix)
1237                 prefix = "";
1238
1239         if (!lstat(filename, &st))
1240                 die("Path '%s' exists on disk, but not in '%.*s'.",
1241                     filename, object_name_len, object_name);
1242         if (errno == ENOENT || errno == ENOTDIR) {
1243                 char *fullname = xstrfmt("%s%s", prefix, filename);
1244
1245                 if (!get_tree_entry(tree_sha1, fullname,
1246                                     sha1, &mode)) {
1247                         die("Path '%s' exists, but not '%s'.\n"
1248                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1249                             fullname,
1250                             filename,
1251                             object_name_len, object_name,
1252                             fullname,
1253                             object_name_len, object_name,
1254                             filename);
1255                 }
1256                 die("Path '%s' does not exist in '%.*s'",
1257                     filename, object_name_len, object_name);
1258         }
1259 }
1260
1261 /* Must be called only when :stage:filename doesn't exist. */
1262 static void diagnose_invalid_index_path(int stage,
1263                                         const char *prefix,
1264                                         const char *filename)
1265 {
1266         struct stat st;
1267         const struct cache_entry *ce;
1268         int pos;
1269         unsigned namelen = strlen(filename);
1270         unsigned fullnamelen;
1271         char *fullname;
1272
1273         if (!prefix)
1274                 prefix = "";
1275
1276         /* Wrong stage number? */
1277         pos = cache_name_pos(filename, namelen);
1278         if (pos < 0)
1279                 pos = -pos - 1;
1280         if (pos < active_nr) {
1281                 ce = active_cache[pos];
1282                 if (ce_namelen(ce) == namelen &&
1283                     !memcmp(ce->name, filename, namelen))
1284                         die("Path '%s' is in the index, but not at stage %d.\n"
1285                             "Did you mean ':%d:%s'?",
1286                             filename, stage,
1287                             ce_stage(ce), filename);
1288         }
1289
1290         /* Confusion between relative and absolute filenames? */
1291         fullnamelen = namelen + strlen(prefix);
1292         fullname = xmalloc(fullnamelen + 1);
1293         strcpy(fullname, prefix);
1294         strcat(fullname, filename);
1295         pos = cache_name_pos(fullname, fullnamelen);
1296         if (pos < 0)
1297                 pos = -pos - 1;
1298         if (pos < active_nr) {
1299                 ce = active_cache[pos];
1300                 if (ce_namelen(ce) == fullnamelen &&
1301                     !memcmp(ce->name, fullname, fullnamelen))
1302                         die("Path '%s' is in the index, but not '%s'.\n"
1303                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1304                             fullname, filename,
1305                             ce_stage(ce), fullname,
1306                             ce_stage(ce), filename);
1307         }
1308
1309         if (!lstat(filename, &st))
1310                 die("Path '%s' exists on disk, but not in the index.", filename);
1311         if (errno == ENOENT || errno == ENOTDIR)
1312                 die("Path '%s' does not exist (neither on disk nor in the index).",
1313                     filename);
1314
1315         free(fullname);
1316 }
1317
1318
1319 static char *resolve_relative_path(const char *rel)
1320 {
1321         if (!starts_with(rel, "./") && !starts_with(rel, "../"))
1322                 return NULL;
1323
1324         if (!startup_info)
1325                 die("BUG: startup_info struct is not initialized.");
1326
1327         if (!is_inside_work_tree())
1328                 die("relative path syntax can't be used outside working tree.");
1329
1330         /* die() inside prefix_path() if resolved path is outside worktree */
1331         return prefix_path(startup_info->prefix,
1332                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1333                            rel);
1334 }
1335
1336 static int get_sha1_with_context_1(const char *name,
1337                                    unsigned flags,
1338                                    const char *prefix,
1339                                    unsigned char *sha1,
1340                                    struct object_context *oc)
1341 {
1342         int ret, bracket_depth;
1343         int namelen = strlen(name);
1344         const char *cp;
1345         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1346
1347         memset(oc, 0, sizeof(*oc));
1348         oc->mode = S_IFINVALID;
1349         ret = get_sha1_1(name, namelen, sha1, flags);
1350         if (!ret)
1351                 return ret;
1352         /*
1353          * sha1:path --> object name of path in ent sha1
1354          * :path -> object name of absolute path in index
1355          * :./path -> object name of path relative to cwd in index
1356          * :[0-3]:path -> object name of path in index at stage
1357          * :/foo -> recent commit matching foo
1358          */
1359         if (name[0] == ':') {
1360                 int stage = 0;
1361                 const struct cache_entry *ce;
1362                 char *new_path = NULL;
1363                 int pos;
1364                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1365                         struct commit_list *list = NULL;
1366                         for_each_ref(handle_one_ref, &list);
1367                         return get_sha1_oneline(name + 2, sha1, list);
1368                 }
1369                 if (namelen < 3 ||
1370                     name[2] != ':' ||
1371                     name[1] < '0' || '3' < name[1])
1372                         cp = name + 1;
1373                 else {
1374                         stage = name[1] - '0';
1375                         cp = name + 3;
1376                 }
1377                 new_path = resolve_relative_path(cp);
1378                 if (!new_path) {
1379                         namelen = namelen - (cp - name);
1380                 } else {
1381                         cp = new_path;
1382                         namelen = strlen(cp);
1383                 }
1384
1385                 strncpy(oc->path, cp,
1386                         sizeof(oc->path));
1387                 oc->path[sizeof(oc->path)-1] = '\0';
1388
1389                 if (!active_cache)
1390                         read_cache();
1391                 pos = cache_name_pos(cp, namelen);
1392                 if (pos < 0)
1393                         pos = -pos - 1;
1394                 while (pos < active_nr) {
1395                         ce = active_cache[pos];
1396                         if (ce_namelen(ce) != namelen ||
1397                             memcmp(ce->name, cp, namelen))
1398                                 break;
1399                         if (ce_stage(ce) == stage) {
1400                                 hashcpy(sha1, ce->sha1);
1401                                 oc->mode = ce->ce_mode;
1402                                 free(new_path);
1403                                 return 0;
1404                         }
1405                         pos++;
1406                 }
1407                 if (only_to_die && name[1] && name[1] != '/')
1408                         diagnose_invalid_index_path(stage, prefix, cp);
1409                 free(new_path);
1410                 return -1;
1411         }
1412         for (cp = name, bracket_depth = 0; *cp; cp++) {
1413                 if (*cp == '{')
1414                         bracket_depth++;
1415                 else if (bracket_depth && *cp == '}')
1416                         bracket_depth--;
1417                 else if (!bracket_depth && *cp == ':')
1418                         break;
1419         }
1420         if (*cp == ':') {
1421                 unsigned char tree_sha1[20];
1422                 int len = cp - name;
1423                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1424                         const char *filename = cp+1;
1425                         char *new_filename = NULL;
1426
1427                         new_filename = resolve_relative_path(filename);
1428                         if (new_filename)
1429                                 filename = new_filename;
1430                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1431                         if (ret && only_to_die) {
1432                                 diagnose_invalid_sha1_path(prefix, filename,
1433                                                            tree_sha1,
1434                                                            name, len);
1435                         }
1436                         hashcpy(oc->tree, tree_sha1);
1437                         strncpy(oc->path, filename,
1438                                 sizeof(oc->path));
1439                         oc->path[sizeof(oc->path)-1] = '\0';
1440
1441                         free(new_filename);
1442                         return ret;
1443                 } else {
1444                         if (only_to_die)
1445                                 die("Invalid object name '%.*s'.", len, name);
1446                 }
1447         }
1448         return ret;
1449 }
1450
1451 /*
1452  * Call this function when you know "name" given by the end user must
1453  * name an object but it doesn't; the function _may_ die with a better
1454  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1455  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1456  * you have a chance to diagnose the error further.
1457  */
1458 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1459 {
1460         struct object_context oc;
1461         unsigned char sha1[20];
1462         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1463 }
1464
1465 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1466 {
1467         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1468 }