Merge branch 'ml/cygwin-updates'
[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 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf);
435
436 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
437 {
438         static const char *warn_msg = "refname '%.*s' is ambiguous.";
439         static const char *object_name_msg = N_(
440         "Git normally never creates a ref that ends with 40 hex characters\n"
441         "because it will be ignored when you just specify 40-hex. These refs\n"
442         "may be created by mistake. For example,\n"
443         "\n"
444         "  git checkout -b $br $(git rev-parse ...)\n"
445         "\n"
446         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
447         "examine these refs and maybe delete them. Turn this message off by\n"
448         "running \"git config advice.object_name_warning false\"");
449         unsigned char tmp_sha1[20];
450         char *real_ref = NULL;
451         int refs_found = 0;
452         int at, reflog_len, nth_prior = 0;
453
454         if (len == 40 && !get_sha1_hex(str, sha1)) {
455                 if (warn_on_object_refname_ambiguity) {
456                         refs_found = dwim_ref(str, len, tmp_sha1, &real_ref);
457                         if (refs_found > 0 && warn_ambiguous_refs) {
458                                 warning(warn_msg, len, str);
459                                 if (advice_object_name_warning)
460                                         fprintf(stderr, "%s\n", _(object_name_msg));
461                         }
462                         free(real_ref);
463                 }
464                 return 0;
465         }
466
467         /* basic@{time or number or -number} format to query ref-log */
468         reflog_len = at = 0;
469         if (len && str[len-1] == '}') {
470                 for (at = len-4; at >= 0; at--) {
471                         if (str[at] == '@' && str[at+1] == '{') {
472                                 if (str[at+2] == '-') {
473                                         if (at != 0)
474                                                 /* @{-N} not at start */
475                                                 return -1;
476                                         nth_prior = 1;
477                                         continue;
478                                 }
479                                 if (!upstream_mark(str + at, len - at)) {
480                                         reflog_len = (len-1) - (at+2);
481                                         len = at;
482                                 }
483                                 break;
484                         }
485                 }
486         }
487
488         /* Accept only unambiguous ref paths. */
489         if (len && ambiguous_path(str, len))
490                 return -1;
491
492         if (nth_prior) {
493                 struct strbuf buf = STRBUF_INIT;
494                 int detached;
495
496                 if (interpret_nth_prior_checkout(str, &buf) > 0) {
497                         detached = (buf.len == 40 && !get_sha1_hex(buf.buf, sha1));
498                         strbuf_release(&buf);
499                         if (detached)
500                                 return 0;
501                 }
502         }
503
504         if (!len && reflog_len)
505                 /* allow "@{...}" to mean the current branch reflog */
506                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
507         else if (reflog_len)
508                 refs_found = dwim_log(str, len, sha1, &real_ref);
509         else
510                 refs_found = dwim_ref(str, len, sha1, &real_ref);
511
512         if (!refs_found)
513                 return -1;
514
515         if (warn_ambiguous_refs &&
516             (refs_found > 1 ||
517              !get_short_sha1(str, len, tmp_sha1, GET_SHA1_QUIETLY)))
518                 warning(warn_msg, len, str);
519
520         if (reflog_len) {
521                 int nth, i;
522                 unsigned long at_time;
523                 unsigned long co_time;
524                 int co_tz, co_cnt;
525
526                 /* Is it asking for N-th entry, or approxidate? */
527                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
528                         char ch = str[at+2+i];
529                         if ('0' <= ch && ch <= '9')
530                                 nth = nth * 10 + ch - '0';
531                         else
532                                 nth = -1;
533                 }
534                 if (100000000 <= nth) {
535                         at_time = nth;
536                         nth = -1;
537                 } else if (0 <= nth)
538                         at_time = 0;
539                 else {
540                         int errors = 0;
541                         char *tmp = xstrndup(str + at + 2, reflog_len);
542                         at_time = approxidate_careful(tmp, &errors);
543                         free(tmp);
544                         if (errors)
545                                 return -1;
546                 }
547                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
548                                 &co_time, &co_tz, &co_cnt)) {
549                         if (!len) {
550                                 if (!prefixcmp(real_ref, "refs/heads/")) {
551                                         str = real_ref + 11;
552                                         len = strlen(real_ref + 11);
553                                 } else {
554                                         /* detached HEAD */
555                                         str = "HEAD";
556                                         len = 4;
557                                 }
558                         }
559                         if (at_time)
560                                 warning("Log for '%.*s' only goes "
561                                         "back to %s.", len, str,
562                                         show_date(co_time, co_tz, DATE_RFC2822));
563                         else {
564                                 die("Log for '%.*s' only has %d entries.",
565                                     len, str, co_cnt);
566                         }
567                 }
568         }
569
570         free(real_ref);
571         return 0;
572 }
573
574 static int get_parent(const char *name, int len,
575                       unsigned char *result, int idx)
576 {
577         unsigned char sha1[20];
578         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
579         struct commit *commit;
580         struct commit_list *p;
581
582         if (ret)
583                 return ret;
584         commit = lookup_commit_reference(sha1);
585         if (!commit)
586                 return -1;
587         if (parse_commit(commit))
588                 return -1;
589         if (!idx) {
590                 hashcpy(result, commit->object.sha1);
591                 return 0;
592         }
593         p = commit->parents;
594         while (p) {
595                 if (!--idx) {
596                         hashcpy(result, p->item->object.sha1);
597                         return 0;
598                 }
599                 p = p->next;
600         }
601         return -1;
602 }
603
604 static int get_nth_ancestor(const char *name, int len,
605                             unsigned char *result, int generation)
606 {
607         unsigned char sha1[20];
608         struct commit *commit;
609         int ret;
610
611         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
612         if (ret)
613                 return ret;
614         commit = lookup_commit_reference(sha1);
615         if (!commit)
616                 return -1;
617
618         while (generation--) {
619                 if (parse_commit(commit) || !commit->parents)
620                         return -1;
621                 commit = commit->parents->item;
622         }
623         hashcpy(result, commit->object.sha1);
624         return 0;
625 }
626
627 struct object *peel_to_type(const char *name, int namelen,
628                             struct object *o, enum object_type expected_type)
629 {
630         if (name && !namelen)
631                 namelen = strlen(name);
632         while (1) {
633                 if (!o || (!o->parsed && !parse_object(o->sha1)))
634                         return NULL;
635                 if (expected_type == OBJ_ANY || o->type == expected_type)
636                         return o;
637                 if (o->type == OBJ_TAG)
638                         o = ((struct tag*) o)->tagged;
639                 else if (o->type == OBJ_COMMIT)
640                         o = &(((struct commit *) o)->tree->object);
641                 else {
642                         if (name)
643                                 error("%.*s: expected %s type, but the object "
644                                       "dereferences to %s type",
645                                       namelen, name, typename(expected_type),
646                                       typename(o->type));
647                         return NULL;
648                 }
649         }
650 }
651
652 static int peel_onion(const char *name, int len, unsigned char *sha1)
653 {
654         unsigned char outer[20];
655         const char *sp;
656         unsigned int expected_type = 0;
657         unsigned lookup_flags = 0;
658         struct object *o;
659
660         /*
661          * "ref^{type}" dereferences ref repeatedly until you cannot
662          * dereference anymore, or you get an object of given type,
663          * whichever comes first.  "ref^{}" means just dereference
664          * tags until you get a non-tag.  "ref^0" is a shorthand for
665          * "ref^{commit}".  "commit^{tree}" could be used to find the
666          * top-level tree of the given commit.
667          */
668         if (len < 4 || name[len-1] != '}')
669                 return -1;
670
671         for (sp = name + len - 1; name <= sp; sp--) {
672                 int ch = *sp;
673                 if (ch == '{' && name < sp && sp[-1] == '^')
674                         break;
675         }
676         if (sp <= name)
677                 return -1;
678
679         sp++; /* beginning of type name, or closing brace for empty */
680         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
681                 expected_type = OBJ_COMMIT;
682         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
683                 expected_type = OBJ_TREE;
684         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
685                 expected_type = OBJ_BLOB;
686         else if (!prefixcmp(sp, "object}"))
687                 expected_type = OBJ_ANY;
688         else if (sp[0] == '}')
689                 expected_type = OBJ_NONE;
690         else if (sp[0] == '/')
691                 expected_type = OBJ_COMMIT;
692         else
693                 return -1;
694
695         if (expected_type == OBJ_COMMIT)
696                 lookup_flags = GET_SHA1_COMMITTISH;
697         else if (expected_type == OBJ_TREE)
698                 lookup_flags = GET_SHA1_TREEISH;
699
700         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
701                 return -1;
702
703         o = parse_object(outer);
704         if (!o)
705                 return -1;
706         if (!expected_type) {
707                 o = deref_tag(o, name, sp - name - 2);
708                 if (!o || (!o->parsed && !parse_object(o->sha1)))
709                         return -1;
710                 hashcpy(sha1, o->sha1);
711                 return 0;
712         }
713
714         /*
715          * At this point, the syntax look correct, so
716          * if we do not get the needed object, we should
717          * barf.
718          */
719         o = peel_to_type(name, len, o, expected_type);
720         if (!o)
721                 return -1;
722
723         hashcpy(sha1, o->sha1);
724         if (sp[0] == '/') {
725                 /* "$commit^{/foo}" */
726                 char *prefix;
727                 int ret;
728                 struct commit_list *list = NULL;
729
730                 /*
731                  * $commit^{/}. Some regex implementation may reject.
732                  * We don't need regex anyway. '' pattern always matches.
733                  */
734                 if (sp[1] == '}')
735                         return 0;
736
737                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
738                 commit_list_insert((struct commit *)o, &list);
739                 ret = get_sha1_oneline(prefix, sha1, list);
740                 free(prefix);
741                 return ret;
742         }
743         return 0;
744 }
745
746 static int get_describe_name(const char *name, int len, unsigned char *sha1)
747 {
748         const char *cp;
749         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
750
751         for (cp = name + len - 1; name + 2 <= cp; cp--) {
752                 char ch = *cp;
753                 if (hexval(ch) & ~0377) {
754                         /* We must be looking at g in "SOMETHING-g"
755                          * for it to be describe output.
756                          */
757                         if (ch == 'g' && cp[-1] == '-') {
758                                 cp++;
759                                 len -= cp - name;
760                                 return get_short_sha1(cp, len, sha1, flags);
761                         }
762                 }
763         }
764         return -1;
765 }
766
767 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
768 {
769         int ret, has_suffix;
770         const char *cp;
771
772         /*
773          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
774          */
775         has_suffix = 0;
776         for (cp = name + len - 1; name <= cp; cp--) {
777                 int ch = *cp;
778                 if ('0' <= ch && ch <= '9')
779                         continue;
780                 if (ch == '~' || ch == '^')
781                         has_suffix = ch;
782                 break;
783         }
784
785         if (has_suffix) {
786                 int num = 0;
787                 int len1 = cp - name;
788                 cp++;
789                 while (cp < name + len)
790                         num = num * 10 + *cp++ - '0';
791                 if (!num && len1 == len - 1)
792                         num = 1;
793                 if (has_suffix == '^')
794                         return get_parent(name, len1, sha1, num);
795                 /* else if (has_suffix == '~') -- goes without saying */
796                 return get_nth_ancestor(name, len1, sha1, num);
797         }
798
799         ret = peel_onion(name, len, sha1);
800         if (!ret)
801                 return 0;
802
803         ret = get_sha1_basic(name, len, sha1);
804         if (!ret)
805                 return 0;
806
807         /* It could be describe output that is "SOMETHING-gXXXX" */
808         ret = get_describe_name(name, len, sha1);
809         if (!ret)
810                 return 0;
811
812         return get_short_sha1(name, len, sha1, lookup_flags);
813 }
814
815 /*
816  * This interprets names like ':/Initial revision of "git"' by searching
817  * through history and returning the first commit whose message starts
818  * the given regular expression.
819  *
820  * For future extension, ':/!' is reserved. If you want to match a message
821  * beginning with a '!', you have to repeat the exclamation mark.
822  */
823 #define ONELINE_SEEN (1u<<20)
824
825 static int handle_one_ref(const char *path,
826                 const unsigned char *sha1, int flag, void *cb_data)
827 {
828         struct commit_list **list = cb_data;
829         struct object *object = parse_object(sha1);
830         if (!object)
831                 return 0;
832         if (object->type == OBJ_TAG) {
833                 object = deref_tag(object, path, strlen(path));
834                 if (!object)
835                         return 0;
836         }
837         if (object->type != OBJ_COMMIT)
838                 return 0;
839         commit_list_insert_by_date((struct commit *)object, list);
840         return 0;
841 }
842
843 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
844                             struct commit_list *list)
845 {
846         struct commit_list *backup = NULL, *l;
847         int found = 0;
848         regex_t regex;
849
850         if (prefix[0] == '!') {
851                 if (prefix[1] != '!')
852                         die ("Invalid search pattern: %s", prefix);
853                 prefix++;
854         }
855
856         if (regcomp(&regex, prefix, REG_EXTENDED))
857                 die("Invalid search pattern: %s", prefix);
858
859         for (l = list; l; l = l->next) {
860                 l->item->object.flags |= ONELINE_SEEN;
861                 commit_list_insert(l->item, &backup);
862         }
863         while (list) {
864                 char *p, *to_free = NULL;
865                 struct commit *commit;
866                 enum object_type type;
867                 unsigned long size;
868                 int matches;
869
870                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
871                 if (!parse_object(commit->object.sha1))
872                         continue;
873                 if (commit->buffer)
874                         p = commit->buffer;
875                 else {
876                         p = read_sha1_file(commit->object.sha1, &type, &size);
877                         if (!p)
878                                 continue;
879                         to_free = p;
880                 }
881
882                 p = strstr(p, "\n\n");
883                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
884                 free(to_free);
885
886                 if (matches) {
887                         hashcpy(sha1, commit->object.sha1);
888                         found = 1;
889                         break;
890                 }
891         }
892         regfree(&regex);
893         free_commit_list(list);
894         for (l = backup; l; l = l->next)
895                 clear_commit_marks(l->item, ONELINE_SEEN);
896         free_commit_list(backup);
897         return found ? 0 : -1;
898 }
899
900 struct grab_nth_branch_switch_cbdata {
901         int remaining;
902         struct strbuf buf;
903 };
904
905 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
906                                   const char *email, unsigned long timestamp, int tz,
907                                   const char *message, void *cb_data)
908 {
909         struct grab_nth_branch_switch_cbdata *cb = cb_data;
910         const char *match = NULL, *target = NULL;
911         size_t len;
912
913         if (!prefixcmp(message, "checkout: moving from ")) {
914                 match = message + strlen("checkout: moving from ");
915                 target = strstr(match, " to ");
916         }
917
918         if (!match || !target)
919                 return 0;
920         if (--(cb->remaining) == 0) {
921                 len = target - match;
922                 strbuf_reset(&cb->buf);
923                 strbuf_add(&cb->buf, match, len);
924                 return 1; /* we are done */
925         }
926         return 0;
927 }
928
929 /*
930  * Parse @{-N} syntax, return the number of characters parsed
931  * if successful; otherwise signal an error with negative value.
932  */
933 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
934 {
935         long nth;
936         int retval;
937         struct grab_nth_branch_switch_cbdata cb;
938         const char *brace;
939         char *num_end;
940
941         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
942                 return -1;
943         brace = strchr(name, '}');
944         if (!brace)
945                 return -1;
946         nth = strtol(name + 3, &num_end, 10);
947         if (num_end != brace)
948                 return -1;
949         if (nth <= 0)
950                 return -1;
951         cb.remaining = nth;
952         strbuf_init(&cb.buf, 20);
953
954         retval = 0;
955         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
956                 strbuf_reset(buf);
957                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
958                 retval = brace - name + 1;
959         }
960
961         strbuf_release(&cb.buf);
962         return retval;
963 }
964
965 int get_sha1_mb(const char *name, unsigned char *sha1)
966 {
967         struct commit *one, *two;
968         struct commit_list *mbs;
969         unsigned char sha1_tmp[20];
970         const char *dots;
971         int st;
972
973         dots = strstr(name, "...");
974         if (!dots)
975                 return get_sha1(name, sha1);
976         if (dots == name)
977                 st = get_sha1("HEAD", sha1_tmp);
978         else {
979                 struct strbuf sb;
980                 strbuf_init(&sb, dots - name);
981                 strbuf_add(&sb, name, dots - name);
982                 st = get_sha1_committish(sb.buf, sha1_tmp);
983                 strbuf_release(&sb);
984         }
985         if (st)
986                 return st;
987         one = lookup_commit_reference_gently(sha1_tmp, 0);
988         if (!one)
989                 return -1;
990
991         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
992                 return -1;
993         two = lookup_commit_reference_gently(sha1_tmp, 0);
994         if (!two)
995                 return -1;
996         mbs = get_merge_bases(one, two, 1);
997         if (!mbs || mbs->next)
998                 st = -1;
999         else {
1000                 st = 0;
1001                 hashcpy(sha1, mbs->item->object.sha1);
1002         }
1003         free_commit_list(mbs);
1004         return st;
1005 }
1006
1007 /* parse @something syntax, when 'something' is not {.*} */
1008 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1009 {
1010         if (len || name[1] == '{')
1011                 return -1;
1012
1013         strbuf_reset(buf);
1014         strbuf_add(buf, "HEAD", 4);
1015         return 1;
1016 }
1017
1018 static int reinterpret(const char *name, int namelen, int len, struct strbuf *buf)
1019 {
1020         /* we have extra data, which might need further processing */
1021         struct strbuf tmp = STRBUF_INIT;
1022         int used = buf->len;
1023         int ret;
1024
1025         strbuf_add(buf, name + len, namelen - len);
1026         ret = interpret_branch_name(buf->buf, &tmp);
1027         /* that data was not interpreted, remove our cruft */
1028         if (ret < 0) {
1029                 strbuf_setlen(buf, used);
1030                 return len;
1031         }
1032         strbuf_reset(buf);
1033         strbuf_addbuf(buf, &tmp);
1034         strbuf_release(&tmp);
1035         /* tweak for size of {-N} versus expanded ref name */
1036         return ret - used + len;
1037 }
1038
1039 /*
1040  * This reads short-hand syntax that not only evaluates to a commit
1041  * object name, but also can act as if the end user spelled the name
1042  * of the branch from the command line.
1043  *
1044  * - "@{-N}" finds the name of the Nth previous branch we were on, and
1045  *   places the name of the branch in the given buf and returns the
1046  *   number of characters parsed if successful.
1047  *
1048  * - "<branch>@{upstream}" finds the name of the other ref that
1049  *   <branch> is configured to merge with (missing <branch> defaults
1050  *   to the current branch), and places the name of the branch in the
1051  *   given buf and returns the number of characters parsed if
1052  *   successful.
1053  *
1054  * If the input is not of the accepted format, it returns a negative
1055  * number to signal an error.
1056  *
1057  * If the input was ok but there are not N branch switches in the
1058  * reflog, it returns 0.
1059  */
1060 int interpret_branch_name(const char *name, struct strbuf *buf)
1061 {
1062         char *cp;
1063         struct branch *upstream;
1064         int namelen = strlen(name);
1065         int len = interpret_nth_prior_checkout(name, buf);
1066         int tmp_len;
1067
1068         if (!len) {
1069                 return len; /* syntax Ok, not enough switches */
1070         } else if (len > 0) {
1071                 if (len == namelen)
1072                         return len; /* consumed all */
1073                 else
1074                         return reinterpret(name, namelen, len, buf);
1075         }
1076
1077         cp = strchr(name, '@');
1078         if (!cp)
1079                 return -1;
1080
1081         len = interpret_empty_at(name, namelen, cp - name, buf);
1082         if (len > 0)
1083                 return reinterpret(name, namelen, len, buf);
1084
1085         tmp_len = upstream_mark(cp, namelen - (cp - name));
1086         if (!tmp_len)
1087                 return -1;
1088
1089         len = cp + tmp_len - name;
1090         cp = xstrndup(name, cp - name);
1091         upstream = branch_get(*cp ? cp : NULL);
1092         /*
1093          * Upstream can be NULL only if cp refers to HEAD and HEAD
1094          * points to something different than a branch.
1095          */
1096         if (!upstream)
1097                 die(_("HEAD does not point to a branch"));
1098         if (!upstream->merge || !upstream->merge[0]->dst) {
1099                 if (!ref_exists(upstream->refname))
1100                         die(_("No such branch: '%s'"), cp);
1101                 if (!upstream->merge) {
1102                         die(_("No upstream configured for branch '%s'"),
1103                                 upstream->name);
1104                 }
1105                 die(
1106                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1107                         upstream->merge[0]->src);
1108         }
1109         free(cp);
1110         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1111         strbuf_reset(buf);
1112         strbuf_addstr(buf, cp);
1113         free(cp);
1114         return len;
1115 }
1116
1117 int strbuf_branchname(struct strbuf *sb, const char *name)
1118 {
1119         int len = strlen(name);
1120         int used = interpret_branch_name(name, sb);
1121
1122         if (used == len)
1123                 return 0;
1124         if (used < 0)
1125                 used = 0;
1126         strbuf_add(sb, name + used, len - used);
1127         return len;
1128 }
1129
1130 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1131 {
1132         strbuf_branchname(sb, name);
1133         if (name[0] == '-')
1134                 return -1;
1135         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1136         return check_refname_format(sb->buf, 0);
1137 }
1138
1139 /*
1140  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1141  * notably "xyz^" for "parent of xyz"
1142  */
1143 int get_sha1(const char *name, unsigned char *sha1)
1144 {
1145         struct object_context unused;
1146         return get_sha1_with_context(name, 0, sha1, &unused);
1147 }
1148
1149 /*
1150  * Many callers know that the user meant to name a committish by
1151  * syntactical positions where the object name appears.  Calling this
1152  * function allows the machinery to disambiguate shorter-than-unique
1153  * abbreviated object names between committish and others.
1154  *
1155  * Note that this does NOT error out when the named object is not a
1156  * committish. It is merely to give a hint to the disambiguation
1157  * machinery.
1158  */
1159 int get_sha1_committish(const char *name, unsigned char *sha1)
1160 {
1161         struct object_context unused;
1162         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1163                                      sha1, &unused);
1164 }
1165
1166 int get_sha1_treeish(const char *name, unsigned char *sha1)
1167 {
1168         struct object_context unused;
1169         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1170                                      sha1, &unused);
1171 }
1172
1173 int get_sha1_commit(const char *name, unsigned char *sha1)
1174 {
1175         struct object_context unused;
1176         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1177                                      sha1, &unused);
1178 }
1179
1180 int get_sha1_tree(const char *name, unsigned char *sha1)
1181 {
1182         struct object_context unused;
1183         return get_sha1_with_context(name, GET_SHA1_TREE,
1184                                      sha1, &unused);
1185 }
1186
1187 int get_sha1_blob(const char *name, unsigned char *sha1)
1188 {
1189         struct object_context unused;
1190         return get_sha1_with_context(name, GET_SHA1_BLOB,
1191                                      sha1, &unused);
1192 }
1193
1194 /* Must be called only when object_name:filename doesn't exist. */
1195 static void diagnose_invalid_sha1_path(const char *prefix,
1196                                        const char *filename,
1197                                        const unsigned char *tree_sha1,
1198                                        const char *object_name,
1199                                        int object_name_len)
1200 {
1201         struct stat st;
1202         unsigned char sha1[20];
1203         unsigned mode;
1204
1205         if (!prefix)
1206                 prefix = "";
1207
1208         if (!lstat(filename, &st))
1209                 die("Path '%s' exists on disk, but not in '%.*s'.",
1210                     filename, object_name_len, object_name);
1211         if (errno == ENOENT || errno == ENOTDIR) {
1212                 char *fullname = xmalloc(strlen(filename)
1213                                              + strlen(prefix) + 1);
1214                 strcpy(fullname, prefix);
1215                 strcat(fullname, filename);
1216
1217                 if (!get_tree_entry(tree_sha1, fullname,
1218                                     sha1, &mode)) {
1219                         die("Path '%s' exists, but not '%s'.\n"
1220                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1221                             fullname,
1222                             filename,
1223                             object_name_len, object_name,
1224                             fullname,
1225                             object_name_len, object_name,
1226                             filename);
1227                 }
1228                 die("Path '%s' does not exist in '%.*s'",
1229                     filename, object_name_len, object_name);
1230         }
1231 }
1232
1233 /* Must be called only when :stage:filename doesn't exist. */
1234 static void diagnose_invalid_index_path(int stage,
1235                                         const char *prefix,
1236                                         const char *filename)
1237 {
1238         struct stat st;
1239         const struct cache_entry *ce;
1240         int pos;
1241         unsigned namelen = strlen(filename);
1242         unsigned fullnamelen;
1243         char *fullname;
1244
1245         if (!prefix)
1246                 prefix = "";
1247
1248         /* Wrong stage number? */
1249         pos = cache_name_pos(filename, namelen);
1250         if (pos < 0)
1251                 pos = -pos - 1;
1252         if (pos < active_nr) {
1253                 ce = active_cache[pos];
1254                 if (ce_namelen(ce) == namelen &&
1255                     !memcmp(ce->name, filename, namelen))
1256                         die("Path '%s' is in the index, but not at stage %d.\n"
1257                             "Did you mean ':%d:%s'?",
1258                             filename, stage,
1259                             ce_stage(ce), filename);
1260         }
1261
1262         /* Confusion between relative and absolute filenames? */
1263         fullnamelen = namelen + strlen(prefix);
1264         fullname = xmalloc(fullnamelen + 1);
1265         strcpy(fullname, prefix);
1266         strcat(fullname, filename);
1267         pos = cache_name_pos(fullname, fullnamelen);
1268         if (pos < 0)
1269                 pos = -pos - 1;
1270         if (pos < active_nr) {
1271                 ce = active_cache[pos];
1272                 if (ce_namelen(ce) == fullnamelen &&
1273                     !memcmp(ce->name, fullname, fullnamelen))
1274                         die("Path '%s' is in the index, but not '%s'.\n"
1275                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1276                             fullname, filename,
1277                             ce_stage(ce), fullname,
1278                             ce_stage(ce), filename);
1279         }
1280
1281         if (!lstat(filename, &st))
1282                 die("Path '%s' exists on disk, but not in the index.", filename);
1283         if (errno == ENOENT || errno == ENOTDIR)
1284                 die("Path '%s' does not exist (neither on disk nor in the index).",
1285                     filename);
1286
1287         free(fullname);
1288 }
1289
1290
1291 static char *resolve_relative_path(const char *rel)
1292 {
1293         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1294                 return NULL;
1295
1296         if (!startup_info)
1297                 die("BUG: startup_info struct is not initialized.");
1298
1299         if (!is_inside_work_tree())
1300                 die("relative path syntax can't be used outside working tree.");
1301
1302         /* die() inside prefix_path() if resolved path is outside worktree */
1303         return prefix_path(startup_info->prefix,
1304                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1305                            rel);
1306 }
1307
1308 static int get_sha1_with_context_1(const char *name,
1309                                    unsigned flags,
1310                                    const char *prefix,
1311                                    unsigned char *sha1,
1312                                    struct object_context *oc)
1313 {
1314         int ret, bracket_depth;
1315         int namelen = strlen(name);
1316         const char *cp;
1317         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1318
1319         memset(oc, 0, sizeof(*oc));
1320         oc->mode = S_IFINVALID;
1321         ret = get_sha1_1(name, namelen, sha1, flags);
1322         if (!ret)
1323                 return ret;
1324         /*
1325          * sha1:path --> object name of path in ent sha1
1326          * :path -> object name of absolute path in index
1327          * :./path -> object name of path relative to cwd in index
1328          * :[0-3]:path -> object name of path in index at stage
1329          * :/foo -> recent commit matching foo
1330          */
1331         if (name[0] == ':') {
1332                 int stage = 0;
1333                 const struct cache_entry *ce;
1334                 char *new_path = NULL;
1335                 int pos;
1336                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1337                         struct commit_list *list = NULL;
1338                         for_each_ref(handle_one_ref, &list);
1339                         return get_sha1_oneline(name + 2, sha1, list);
1340                 }
1341                 if (namelen < 3 ||
1342                     name[2] != ':' ||
1343                     name[1] < '0' || '3' < name[1])
1344                         cp = name + 1;
1345                 else {
1346                         stage = name[1] - '0';
1347                         cp = name + 3;
1348                 }
1349                 new_path = resolve_relative_path(cp);
1350                 if (!new_path) {
1351                         namelen = namelen - (cp - name);
1352                 } else {
1353                         cp = new_path;
1354                         namelen = strlen(cp);
1355                 }
1356
1357                 strncpy(oc->path, cp,
1358                         sizeof(oc->path));
1359                 oc->path[sizeof(oc->path)-1] = '\0';
1360
1361                 if (!active_cache)
1362                         read_cache();
1363                 pos = cache_name_pos(cp, namelen);
1364                 if (pos < 0)
1365                         pos = -pos - 1;
1366                 while (pos < active_nr) {
1367                         ce = active_cache[pos];
1368                         if (ce_namelen(ce) != namelen ||
1369                             memcmp(ce->name, cp, namelen))
1370                                 break;
1371                         if (ce_stage(ce) == stage) {
1372                                 hashcpy(sha1, ce->sha1);
1373                                 oc->mode = ce->ce_mode;
1374                                 free(new_path);
1375                                 return 0;
1376                         }
1377                         pos++;
1378                 }
1379                 if (only_to_die && name[1] && name[1] != '/')
1380                         diagnose_invalid_index_path(stage, prefix, cp);
1381                 free(new_path);
1382                 return -1;
1383         }
1384         for (cp = name, bracket_depth = 0; *cp; cp++) {
1385                 if (*cp == '{')
1386                         bracket_depth++;
1387                 else if (bracket_depth && *cp == '}')
1388                         bracket_depth--;
1389                 else if (!bracket_depth && *cp == ':')
1390                         break;
1391         }
1392         if (*cp == ':') {
1393                 unsigned char tree_sha1[20];
1394                 int len = cp - name;
1395                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1396                         const char *filename = cp+1;
1397                         char *new_filename = NULL;
1398
1399                         new_filename = resolve_relative_path(filename);
1400                         if (new_filename)
1401                                 filename = new_filename;
1402                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1403                         if (ret && only_to_die) {
1404                                 diagnose_invalid_sha1_path(prefix, filename,
1405                                                            tree_sha1,
1406                                                            name, len);
1407                         }
1408                         hashcpy(oc->tree, tree_sha1);
1409                         strncpy(oc->path, filename,
1410                                 sizeof(oc->path));
1411                         oc->path[sizeof(oc->path)-1] = '\0';
1412
1413                         free(new_filename);
1414                         return ret;
1415                 } else {
1416                         if (only_to_die)
1417                                 die("Invalid object name '%.*s'.", len, name);
1418                 }
1419         }
1420         return ret;
1421 }
1422
1423 /*
1424  * Call this function when you know "name" given by the end user must
1425  * name an object but it doesn't; the function _may_ die with a better
1426  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1427  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1428  * you have a chance to diagnose the error further.
1429  */
1430 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1431 {
1432         struct object_context oc;
1433         unsigned char sha1[20];
1434         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1435 }
1436
1437 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1438 {
1439         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1440 }