]> git.scripts.mit.edu Git - git.git/blob - sha1_name.c
packed_object_info: hoist delta type resolution to helper
[git.git] / sha1_name.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "tree-walk.h"
7 #include "refs.h"
8 #include "remote.h"
9
10 static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
11
12 typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
13
14 struct disambiguate_state {
15         disambiguate_hint_fn fn;
16         void *cb_data;
17         unsigned char candidate[20];
18         unsigned candidate_exists:1;
19         unsigned candidate_checked:1;
20         unsigned candidate_ok:1;
21         unsigned disambiguate_fn_used:1;
22         unsigned ambiguous:1;
23         unsigned always_call_fn:1;
24 };
25
26 static void update_candidates(struct disambiguate_state *ds, const unsigned char *current)
27 {
28         if (ds->always_call_fn) {
29                 ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
30                 return;
31         }
32         if (!ds->candidate_exists) {
33                 /* this is the first candidate */
34                 hashcpy(ds->candidate, current);
35                 ds->candidate_exists = 1;
36                 return;
37         } else if (!hashcmp(ds->candidate, current)) {
38                 /* the same as what we already have seen */
39                 return;
40         }
41
42         if (!ds->fn) {
43                 /* cannot disambiguate between ds->candidate and current */
44                 ds->ambiguous = 1;
45                 return;
46         }
47
48         if (!ds->candidate_checked) {
49                 ds->candidate_ok = ds->fn(ds->candidate, ds->cb_data);
50                 ds->disambiguate_fn_used = 1;
51                 ds->candidate_checked = 1;
52         }
53
54         if (!ds->candidate_ok) {
55                 /* discard the candidate; we know it does not satisify fn */
56                 hashcpy(ds->candidate, current);
57                 ds->candidate_checked = 0;
58                 return;
59         }
60
61         /* if we reach this point, we know ds->candidate satisfies fn */
62         if (ds->fn(current, ds->cb_data)) {
63                 /*
64                  * if both current and candidate satisfy fn, we cannot
65                  * disambiguate.
66                  */
67                 ds->candidate_ok = 0;
68                 ds->ambiguous = 1;
69         }
70
71         /* otherwise, current can be discarded and candidate is still good */
72 }
73
74 static void find_short_object_filename(int len, const char *hex_pfx, struct disambiguate_state *ds)
75 {
76         struct alternate_object_database *alt;
77         char hex[40];
78         static struct alternate_object_database *fakeent;
79
80         if (!fakeent) {
81                 /*
82                  * Create a "fake" alternate object database that
83                  * points to our own object database, to make it
84                  * easier to get a temporary working space in
85                  * alt->name/alt->base while iterating over the
86                  * object databases including our own.
87                  */
88                 const char *objdir = get_object_directory();
89                 int objdir_len = strlen(objdir);
90                 int entlen = objdir_len + 43;
91                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
92                 memcpy(fakeent->base, objdir, objdir_len);
93                 fakeent->name = fakeent->base + objdir_len + 1;
94                 fakeent->name[-1] = '/';
95         }
96         fakeent->next = alt_odb_list;
97
98         sprintf(hex, "%.2s", hex_pfx);
99         for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
100                 struct dirent *de;
101                 DIR *dir;
102                 sprintf(alt->name, "%.2s/", hex_pfx);
103                 dir = opendir(alt->base);
104                 if (!dir)
105                         continue;
106
107                 while (!ds->ambiguous && (de = readdir(dir)) != NULL) {
108                         unsigned char sha1[20];
109
110                         if (strlen(de->d_name) != 38)
111                                 continue;
112                         if (memcmp(de->d_name, hex_pfx + 2, len - 2))
113                                 continue;
114                         memcpy(hex + 2, de->d_name, 38);
115                         if (!get_sha1_hex(hex, sha1))
116                                 update_candidates(ds, sha1);
117                 }
118                 closedir(dir);
119         }
120 }
121
122 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
123 {
124         do {
125                 if (*a != *b)
126                         return 0;
127                 a++;
128                 b++;
129                 len -= 2;
130         } while (len > 1);
131         if (len)
132                 if ((*a ^ *b) & 0xf0)
133                         return 0;
134         return 1;
135 }
136
137 static void unique_in_pack(int len,
138                           const unsigned char *bin_pfx,
139                            struct packed_git *p,
140                            struct disambiguate_state *ds)
141 {
142         uint32_t num, last, i, first = 0;
143         const unsigned char *current = NULL;
144
145         open_pack_index(p);
146         num = p->num_objects;
147         last = num;
148         while (first < last) {
149                 uint32_t mid = (first + last) / 2;
150                 const unsigned char *current;
151                 int cmp;
152
153                 current = nth_packed_object_sha1(p, mid);
154                 cmp = hashcmp(bin_pfx, current);
155                 if (!cmp) {
156                         first = mid;
157                         break;
158                 }
159                 if (cmp > 0) {
160                         first = mid+1;
161                         continue;
162                 }
163                 last = mid;
164         }
165
166         /*
167          * At this point, "first" is the location of the lowest object
168          * with an object name that could match "bin_pfx".  See if we have
169          * 0, 1 or more objects that actually match(es).
170          */
171         for (i = first; i < num && !ds->ambiguous; i++) {
172                 current = nth_packed_object_sha1(p, i);
173                 if (!match_sha(len, bin_pfx, current))
174                         break;
175                 update_candidates(ds, current);
176         }
177 }
178
179 static void find_short_packed_object(int len, const unsigned char *bin_pfx,
180                                      struct disambiguate_state *ds)
181 {
182         struct packed_git *p;
183
184         prepare_packed_git();
185         for (p = packed_git; p && !ds->ambiguous; p = p->next)
186                 unique_in_pack(len, bin_pfx, p, ds);
187 }
188
189 #define SHORT_NAME_NOT_FOUND (-1)
190 #define SHORT_NAME_AMBIGUOUS (-2)
191
192 static int finish_object_disambiguation(struct disambiguate_state *ds,
193                                         unsigned char *sha1)
194 {
195         if (ds->ambiguous)
196                 return SHORT_NAME_AMBIGUOUS;
197
198         if (!ds->candidate_exists)
199                 return SHORT_NAME_NOT_FOUND;
200
201         if (!ds->candidate_checked)
202                 /*
203                  * If this is the only candidate, there is no point
204                  * calling the disambiguation hint callback.
205                  *
206                  * On the other hand, if the current candidate
207                  * replaced an earlier candidate that did _not_ pass
208                  * the disambiguation hint callback, then we do have
209                  * more than one objects that match the short name
210                  * given, so we should make sure this one matches;
211                  * otherwise, if we discovered this one and the one
212                  * that we previously discarded in the reverse order,
213                  * we would end up showing different results in the
214                  * same repository!
215                  */
216                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
217                                     ds->fn(ds->candidate, ds->cb_data));
218
219         if (!ds->candidate_ok)
220                 return SHORT_NAME_AMBIGUOUS;
221
222         hashcpy(sha1, ds->candidate);
223         return 0;
224 }
225
226 static int disambiguate_commit_only(const unsigned char *sha1, void *cb_data_unused)
227 {
228         int kind = sha1_object_info(sha1, NULL);
229         return kind == OBJ_COMMIT;
230 }
231
232 static int disambiguate_committish_only(const unsigned char *sha1, void *cb_data_unused)
233 {
234         struct object *obj;
235         int kind;
236
237         kind = sha1_object_info(sha1, NULL);
238         if (kind == OBJ_COMMIT)
239                 return 1;
240         if (kind != OBJ_TAG)
241                 return 0;
242
243         /* We need to do this the hard way... */
244         obj = deref_tag(lookup_object(sha1), NULL, 0);
245         if (obj && obj->type == OBJ_COMMIT)
246                 return 1;
247         return 0;
248 }
249
250 static int disambiguate_tree_only(const unsigned char *sha1, void *cb_data_unused)
251 {
252         int kind = sha1_object_info(sha1, NULL);
253         return kind == OBJ_TREE;
254 }
255
256 static int disambiguate_treeish_only(const unsigned char *sha1, void *cb_data_unused)
257 {
258         struct object *obj;
259         int kind;
260
261         kind = sha1_object_info(sha1, NULL);
262         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
263                 return 1;
264         if (kind != OBJ_TAG)
265                 return 0;
266
267         /* We need to do this the hard way... */
268         obj = deref_tag(lookup_object(sha1), NULL, 0);
269         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
270                 return 1;
271         return 0;
272 }
273
274 static int disambiguate_blob_only(const unsigned char *sha1, void *cb_data_unused)
275 {
276         int kind = sha1_object_info(sha1, NULL);
277         return kind == OBJ_BLOB;
278 }
279
280 static int prepare_prefixes(const char *name, int len,
281                             unsigned char *bin_pfx,
282                             char *hex_pfx)
283 {
284         int i;
285
286         hashclr(bin_pfx);
287         memset(hex_pfx, 'x', 40);
288         for (i = 0; i < len ;i++) {
289                 unsigned char c = name[i];
290                 unsigned char val;
291                 if (c >= '0' && c <= '9')
292                         val = c - '0';
293                 else if (c >= 'a' && c <= 'f')
294                         val = c - 'a' + 10;
295                 else if (c >= 'A' && c <='F') {
296                         val = c - 'A' + 10;
297                         c -= 'A' - 'a';
298                 }
299                 else
300                         return -1;
301                 hex_pfx[i] = c;
302                 if (!(i & 1))
303                         val <<= 4;
304                 bin_pfx[i >> 1] |= val;
305         }
306         return 0;
307 }
308
309 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
310                           unsigned flags)
311 {
312         int status;
313         char hex_pfx[40];
314         unsigned char bin_pfx[20];
315         struct disambiguate_state ds;
316         int quietly = !!(flags & GET_SHA1_QUIETLY);
317
318         if (len < MINIMUM_ABBREV || len > 40)
319                 return -1;
320         if (prepare_prefixes(name, len, bin_pfx, hex_pfx) < 0)
321                 return -1;
322
323         prepare_alt_odb();
324
325         memset(&ds, 0, sizeof(ds));
326         if (flags & GET_SHA1_COMMIT)
327                 ds.fn = disambiguate_commit_only;
328         else if (flags & GET_SHA1_COMMITTISH)
329                 ds.fn = disambiguate_committish_only;
330         else if (flags & GET_SHA1_TREE)
331                 ds.fn = disambiguate_tree_only;
332         else if (flags & GET_SHA1_TREEISH)
333                 ds.fn = disambiguate_treeish_only;
334         else if (flags & GET_SHA1_BLOB)
335                 ds.fn = disambiguate_blob_only;
336
337         find_short_object_filename(len, hex_pfx, &ds);
338         find_short_packed_object(len, bin_pfx, &ds);
339         status = finish_object_disambiguation(&ds, sha1);
340
341         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
342                 return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
343         return status;
344 }
345
346
347 int for_each_abbrev(const char *prefix, each_abbrev_fn fn, void *cb_data)
348 {
349         char hex_pfx[40];
350         unsigned char bin_pfx[20];
351         struct disambiguate_state ds;
352         int len = strlen(prefix);
353
354         if (len < MINIMUM_ABBREV || len > 40)
355                 return -1;
356         if (prepare_prefixes(prefix, len, bin_pfx, hex_pfx) < 0)
357                 return -1;
358
359         prepare_alt_odb();
360
361         memset(&ds, 0, sizeof(ds));
362         ds.always_call_fn = 1;
363         ds.cb_data = cb_data;
364         ds.fn = fn;
365
366         find_short_object_filename(len, hex_pfx, &ds);
367         find_short_packed_object(len, bin_pfx, &ds);
368         return ds.ambiguous;
369 }
370
371 const char *find_unique_abbrev(const unsigned char *sha1, int len)
372 {
373         int status, exists;
374         static char hex[41];
375
376         exists = has_sha1_file(sha1);
377         memcpy(hex, sha1_to_hex(sha1), 40);
378         if (len == 40 || !len)
379                 return hex;
380         while (len < 40) {
381                 unsigned char sha1_ret[20];
382                 status = get_short_sha1(hex, len, sha1_ret, GET_SHA1_QUIETLY);
383                 if (exists
384                     ? !status
385                     : status == SHORT_NAME_NOT_FOUND) {
386                         hex[len] = 0;
387                         return hex;
388                 }
389                 len++;
390         }
391         return hex;
392 }
393
394 static int ambiguous_path(const char *path, int len)
395 {
396         int slash = 1;
397         int cnt;
398
399         for (cnt = 0; cnt < len; cnt++) {
400                 switch (*path++) {
401                 case '\0':
402                         break;
403                 case '/':
404                         if (slash)
405                                 break;
406                         slash = 1;
407                         continue;
408                 case '.':
409                         continue;
410                 default:
411                         slash = 0;
412                         continue;
413                 }
414                 break;
415         }
416         return slash;
417 }
418
419 static inline int upstream_mark(const char *string, int len)
420 {
421         const char *suffix[] = { "@{upstream}", "@{u}" };
422         int i;
423
424         for (i = 0; i < ARRAY_SIZE(suffix); i++) {
425                 int suffix_len = strlen(suffix[i]);
426                 if (suffix_len <= len
427                     && !memcmp(string, suffix[i], suffix_len))
428                         return suffix_len;
429         }
430         return 0;
431 }
432
433 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags);
434
435 static int get_sha1_basic(const char *str, int len, unsigned char *sha1)
436 {
437         static const char *warn_msg = "refname '%.*s' is ambiguous.";
438         static const char *object_name_msg = N_(
439         "Git normally never creates a ref that ends with 40 hex characters\n"
440         "because it will be ignored when you just specify 40-hex. These refs\n"
441         "may be created by mistake. For example,\n"
442         "\n"
443         "  git checkout -b $br $(git rev-parse ...)\n"
444         "\n"
445         "where \"$br\" is somehow empty and a 40-hex ref is created. Please\n"
446         "examine these refs and maybe delete them. Turn this message off by\n"
447         "running \"git config advice.object_name_warning false\"");
448         unsigned char tmp_sha1[20];
449         char *real_ref = NULL;
450         int refs_found = 0;
451         int at, reflog_len;
452
453         if (len == 40 && !get_sha1_hex(str, sha1)) {
454                 if (warn_on_object_refname_ambiguity) {
455                         refs_found = dwim_ref(str, len, tmp_sha1, &real_ref);
456                         if (refs_found > 0 && warn_ambiguous_refs) {
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-2; at >= 0; at--) {
470                         if (str[at] == '@' && str[at+1] == '{') {
471                                 if (!upstream_mark(str + at, len - at)) {
472                                         reflog_len = (len-1) - (at+2);
473                                         len = at;
474                                 }
475                                 break;
476                         }
477                 }
478         }
479
480         /* Accept only unambiguous ref paths. */
481         if (len && ambiguous_path(str, len))
482                 return -1;
483
484         if (!len && reflog_len) {
485                 struct strbuf buf = STRBUF_INIT;
486                 int ret;
487                 /* try the @{-N} syntax for n-th checkout */
488                 ret = interpret_branch_name(str+at, &buf);
489                 if (ret > 0) {
490                         /* substitute this branch name and restart */
491                         return get_sha1_1(buf.buf, buf.len, sha1, 0);
492                 } else if (ret == 0) {
493                         return -1;
494                 }
495                 /* allow "@{...}" to mean the current branch reflog */
496                 refs_found = dwim_ref("HEAD", 4, sha1, &real_ref);
497         } else if (reflog_len)
498                 refs_found = dwim_log(str, len, sha1, &real_ref);
499         else
500                 refs_found = dwim_ref(str, len, sha1, &real_ref);
501
502         if (!refs_found)
503                 return -1;
504
505         if (warn_ambiguous_refs &&
506             (refs_found > 1 ||
507              !get_short_sha1(str, len, tmp_sha1, GET_SHA1_QUIETLY)))
508                 warning(warn_msg, len, str);
509
510         if (reflog_len) {
511                 int nth, i;
512                 unsigned long at_time;
513                 unsigned long co_time;
514                 int co_tz, co_cnt;
515
516                 /* a @{-N} placed anywhere except the start is an error */
517                 if (str[at+2] == '-')
518                         return -1;
519
520                 /* Is it asking for N-th entry, or approxidate? */
521                 for (i = nth = 0; 0 <= nth && i < reflog_len; i++) {
522                         char ch = str[at+2+i];
523                         if ('0' <= ch && ch <= '9')
524                                 nth = nth * 10 + ch - '0';
525                         else
526                                 nth = -1;
527                 }
528                 if (100000000 <= nth) {
529                         at_time = nth;
530                         nth = -1;
531                 } else if (0 <= nth)
532                         at_time = 0;
533                 else {
534                         int errors = 0;
535                         char *tmp = xstrndup(str + at + 2, reflog_len);
536                         at_time = approxidate_careful(tmp, &errors);
537                         free(tmp);
538                         if (errors)
539                                 return -1;
540                 }
541                 if (read_ref_at(real_ref, at_time, nth, sha1, NULL,
542                                 &co_time, &co_tz, &co_cnt)) {
543                         if (at_time)
544                                 warning("Log for '%.*s' only goes "
545                                         "back to %s.", len, str,
546                                         show_date(co_time, co_tz, DATE_RFC2822));
547                         else {
548                                 free(real_ref);
549                                 die("Log for '%.*s' only has %d entries.",
550                                     len, str, co_cnt);
551                         }
552                 }
553         }
554
555         free(real_ref);
556         return 0;
557 }
558
559 static int get_parent(const char *name, int len,
560                       unsigned char *result, int idx)
561 {
562         unsigned char sha1[20];
563         int ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
564         struct commit *commit;
565         struct commit_list *p;
566
567         if (ret)
568                 return ret;
569         commit = lookup_commit_reference(sha1);
570         if (!commit)
571                 return -1;
572         if (parse_commit(commit))
573                 return -1;
574         if (!idx) {
575                 hashcpy(result, commit->object.sha1);
576                 return 0;
577         }
578         p = commit->parents;
579         while (p) {
580                 if (!--idx) {
581                         hashcpy(result, p->item->object.sha1);
582                         return 0;
583                 }
584                 p = p->next;
585         }
586         return -1;
587 }
588
589 static int get_nth_ancestor(const char *name, int len,
590                             unsigned char *result, int generation)
591 {
592         unsigned char sha1[20];
593         struct commit *commit;
594         int ret;
595
596         ret = get_sha1_1(name, len, sha1, GET_SHA1_COMMITTISH);
597         if (ret)
598                 return ret;
599         commit = lookup_commit_reference(sha1);
600         if (!commit)
601                 return -1;
602
603         while (generation--) {
604                 if (parse_commit(commit) || !commit->parents)
605                         return -1;
606                 commit = commit->parents->item;
607         }
608         hashcpy(result, commit->object.sha1);
609         return 0;
610 }
611
612 struct object *peel_to_type(const char *name, int namelen,
613                             struct object *o, enum object_type expected_type)
614 {
615         if (name && !namelen)
616                 namelen = strlen(name);
617         while (1) {
618                 if (!o || (!o->parsed && !parse_object(o->sha1)))
619                         return NULL;
620                 if (expected_type == OBJ_ANY || o->type == expected_type)
621                         return o;
622                 if (o->type == OBJ_TAG)
623                         o = ((struct tag*) o)->tagged;
624                 else if (o->type == OBJ_COMMIT)
625                         o = &(((struct commit *) o)->tree->object);
626                 else {
627                         if (name)
628                                 error("%.*s: expected %s type, but the object "
629                                       "dereferences to %s type",
630                                       namelen, name, typename(expected_type),
631                                       typename(o->type));
632                         return NULL;
633                 }
634         }
635 }
636
637 static int peel_onion(const char *name, int len, unsigned char *sha1)
638 {
639         unsigned char outer[20];
640         const char *sp;
641         unsigned int expected_type = 0;
642         unsigned lookup_flags = 0;
643         struct object *o;
644
645         /*
646          * "ref^{type}" dereferences ref repeatedly until you cannot
647          * dereference anymore, or you get an object of given type,
648          * whichever comes first.  "ref^{}" means just dereference
649          * tags until you get a non-tag.  "ref^0" is a shorthand for
650          * "ref^{commit}".  "commit^{tree}" could be used to find the
651          * top-level tree of the given commit.
652          */
653         if (len < 4 || name[len-1] != '}')
654                 return -1;
655
656         for (sp = name + len - 1; name <= sp; sp--) {
657                 int ch = *sp;
658                 if (ch == '{' && name < sp && sp[-1] == '^')
659                         break;
660         }
661         if (sp <= name)
662                 return -1;
663
664         sp++; /* beginning of type name, or closing brace for empty */
665         if (!strncmp(commit_type, sp, 6) && sp[6] == '}')
666                 expected_type = OBJ_COMMIT;
667         else if (!strncmp(tree_type, sp, 4) && sp[4] == '}')
668                 expected_type = OBJ_TREE;
669         else if (!strncmp(blob_type, sp, 4) && sp[4] == '}')
670                 expected_type = OBJ_BLOB;
671         else if (!prefixcmp(sp, "object}"))
672                 expected_type = OBJ_ANY;
673         else if (sp[0] == '}')
674                 expected_type = OBJ_NONE;
675         else if (sp[0] == '/')
676                 expected_type = OBJ_COMMIT;
677         else
678                 return -1;
679
680         if (expected_type == OBJ_COMMIT)
681                 lookup_flags = GET_SHA1_COMMITTISH;
682         else if (expected_type == OBJ_TREE)
683                 lookup_flags = GET_SHA1_TREEISH;
684
685         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
686                 return -1;
687
688         o = parse_object(outer);
689         if (!o)
690                 return -1;
691         if (!expected_type) {
692                 o = deref_tag(o, name, sp - name - 2);
693                 if (!o || (!o->parsed && !parse_object(o->sha1)))
694                         return -1;
695                 hashcpy(sha1, o->sha1);
696                 return 0;
697         }
698
699         /*
700          * At this point, the syntax look correct, so
701          * if we do not get the needed object, we should
702          * barf.
703          */
704         o = peel_to_type(name, len, o, expected_type);
705         if (!o)
706                 return -1;
707
708         hashcpy(sha1, o->sha1);
709         if (sp[0] == '/') {
710                 /* "$commit^{/foo}" */
711                 char *prefix;
712                 int ret;
713                 struct commit_list *list = NULL;
714
715                 /*
716                  * $commit^{/}. Some regex implementation may reject.
717                  * We don't need regex anyway. '' pattern always matches.
718                  */
719                 if (sp[1] == '}')
720                         return 0;
721
722                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
723                 commit_list_insert((struct commit *)o, &list);
724                 ret = get_sha1_oneline(prefix, sha1, list);
725                 free(prefix);
726                 return ret;
727         }
728         return 0;
729 }
730
731 static int get_describe_name(const char *name, int len, unsigned char *sha1)
732 {
733         const char *cp;
734         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
735
736         for (cp = name + len - 1; name + 2 <= cp; cp--) {
737                 char ch = *cp;
738                 if (hexval(ch) & ~0377) {
739                         /* We must be looking at g in "SOMETHING-g"
740                          * for it to be describe output.
741                          */
742                         if (ch == 'g' && cp[-1] == '-') {
743                                 cp++;
744                                 len -= cp - name;
745                                 return get_short_sha1(cp, len, sha1, flags);
746                         }
747                 }
748         }
749         return -1;
750 }
751
752 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
753 {
754         int ret, has_suffix;
755         const char *cp;
756
757         /*
758          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
759          */
760         has_suffix = 0;
761         for (cp = name + len - 1; name <= cp; cp--) {
762                 int ch = *cp;
763                 if ('0' <= ch && ch <= '9')
764                         continue;
765                 if (ch == '~' || ch == '^')
766                         has_suffix = ch;
767                 break;
768         }
769
770         if (has_suffix) {
771                 int num = 0;
772                 int len1 = cp - name;
773                 cp++;
774                 while (cp < name + len)
775                         num = num * 10 + *cp++ - '0';
776                 if (!num && len1 == len - 1)
777                         num = 1;
778                 if (has_suffix == '^')
779                         return get_parent(name, len1, sha1, num);
780                 /* else if (has_suffix == '~') -- goes without saying */
781                 return get_nth_ancestor(name, len1, sha1, num);
782         }
783
784         ret = peel_onion(name, len, sha1);
785         if (!ret)
786                 return 0;
787
788         ret = get_sha1_basic(name, len, sha1);
789         if (!ret)
790                 return 0;
791
792         /* It could be describe output that is "SOMETHING-gXXXX" */
793         ret = get_describe_name(name, len, sha1);
794         if (!ret)
795                 return 0;
796
797         return get_short_sha1(name, len, sha1, lookup_flags);
798 }
799
800 /*
801  * This interprets names like ':/Initial revision of "git"' by searching
802  * through history and returning the first commit whose message starts
803  * the given regular expression.
804  *
805  * For future extension, ':/!' is reserved. If you want to match a message
806  * beginning with a '!', you have to repeat the exclamation mark.
807  */
808 #define ONELINE_SEEN (1u<<20)
809
810 static int handle_one_ref(const char *path,
811                 const unsigned char *sha1, int flag, void *cb_data)
812 {
813         struct commit_list **list = cb_data;
814         struct object *object = parse_object(sha1);
815         if (!object)
816                 return 0;
817         if (object->type == OBJ_TAG) {
818                 object = deref_tag(object, path, strlen(path));
819                 if (!object)
820                         return 0;
821         }
822         if (object->type != OBJ_COMMIT)
823                 return 0;
824         commit_list_insert_by_date((struct commit *)object, list);
825         return 0;
826 }
827
828 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
829                             struct commit_list *list)
830 {
831         struct commit_list *backup = NULL, *l;
832         int found = 0;
833         regex_t regex;
834
835         if (prefix[0] == '!') {
836                 if (prefix[1] != '!')
837                         die ("Invalid search pattern: %s", prefix);
838                 prefix++;
839         }
840
841         if (regcomp(&regex, prefix, REG_EXTENDED))
842                 die("Invalid search pattern: %s", prefix);
843
844         for (l = list; l; l = l->next) {
845                 l->item->object.flags |= ONELINE_SEEN;
846                 commit_list_insert(l->item, &backup);
847         }
848         while (list) {
849                 char *p, *to_free = NULL;
850                 struct commit *commit;
851                 enum object_type type;
852                 unsigned long size;
853                 int matches;
854
855                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
856                 if (!parse_object(commit->object.sha1))
857                         continue;
858                 if (commit->buffer)
859                         p = commit->buffer;
860                 else {
861                         p = read_sha1_file(commit->object.sha1, &type, &size);
862                         if (!p)
863                                 continue;
864                         to_free = p;
865                 }
866
867                 p = strstr(p, "\n\n");
868                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
869                 free(to_free);
870
871                 if (matches) {
872                         hashcpy(sha1, commit->object.sha1);
873                         found = 1;
874                         break;
875                 }
876         }
877         regfree(&regex);
878         free_commit_list(list);
879         for (l = backup; l; l = l->next)
880                 clear_commit_marks(l->item, ONELINE_SEEN);
881         free_commit_list(backup);
882         return found ? 0 : -1;
883 }
884
885 struct grab_nth_branch_switch_cbdata {
886         int remaining;
887         struct strbuf buf;
888 };
889
890 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
891                                   const char *email, unsigned long timestamp, int tz,
892                                   const char *message, void *cb_data)
893 {
894         struct grab_nth_branch_switch_cbdata *cb = cb_data;
895         const char *match = NULL, *target = NULL;
896         size_t len;
897
898         if (!prefixcmp(message, "checkout: moving from ")) {
899                 match = message + strlen("checkout: moving from ");
900                 target = strstr(match, " to ");
901         }
902
903         if (!match || !target)
904                 return 0;
905         if (--(cb->remaining) == 0) {
906                 len = target - match;
907                 strbuf_reset(&cb->buf);
908                 strbuf_add(&cb->buf, match, len);
909                 return 1; /* we are done */
910         }
911         return 0;
912 }
913
914 /*
915  * Parse @{-N} syntax, return the number of characters parsed
916  * if successful; otherwise signal an error with negative value.
917  */
918 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
919 {
920         long nth;
921         int retval;
922         struct grab_nth_branch_switch_cbdata cb;
923         const char *brace;
924         char *num_end;
925
926         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
927                 return -1;
928         brace = strchr(name, '}');
929         if (!brace)
930                 return -1;
931         nth = strtol(name + 3, &num_end, 10);
932         if (num_end != brace)
933                 return -1;
934         if (nth <= 0)
935                 return -1;
936         cb.remaining = nth;
937         strbuf_init(&cb.buf, 20);
938
939         retval = 0;
940         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
941                 strbuf_reset(buf);
942                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
943                 retval = brace - name + 1;
944         }
945
946         strbuf_release(&cb.buf);
947         return retval;
948 }
949
950 int get_sha1_mb(const char *name, unsigned char *sha1)
951 {
952         struct commit *one, *two;
953         struct commit_list *mbs;
954         unsigned char sha1_tmp[20];
955         const char *dots;
956         int st;
957
958         dots = strstr(name, "...");
959         if (!dots)
960                 return get_sha1(name, sha1);
961         if (dots == name)
962                 st = get_sha1("HEAD", sha1_tmp);
963         else {
964                 struct strbuf sb;
965                 strbuf_init(&sb, dots - name);
966                 strbuf_add(&sb, name, dots - name);
967                 st = get_sha1_committish(sb.buf, sha1_tmp);
968                 strbuf_release(&sb);
969         }
970         if (st)
971                 return st;
972         one = lookup_commit_reference_gently(sha1_tmp, 0);
973         if (!one)
974                 return -1;
975
976         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
977                 return -1;
978         two = lookup_commit_reference_gently(sha1_tmp, 0);
979         if (!two)
980                 return -1;
981         mbs = get_merge_bases(one, two, 1);
982         if (!mbs || mbs->next)
983                 st = -1;
984         else {
985                 st = 0;
986                 hashcpy(sha1, mbs->item->object.sha1);
987         }
988         free_commit_list(mbs);
989         return st;
990 }
991
992 /*
993  * This reads short-hand syntax that not only evaluates to a commit
994  * object name, but also can act as if the end user spelled the name
995  * of the branch from the command line.
996  *
997  * - "@{-N}" finds the name of the Nth previous branch we were on, and
998  *   places the name of the branch in the given buf and returns the
999  *   number of characters parsed if successful.
1000  *
1001  * - "<branch>@{upstream}" finds the name of the other ref that
1002  *   <branch> is configured to merge with (missing <branch> defaults
1003  *   to the current branch), and places the name of the branch in the
1004  *   given buf and returns the number of characters parsed if
1005  *   successful.
1006  *
1007  * If the input is not of the accepted format, it returns a negative
1008  * number to signal an error.
1009  *
1010  * If the input was ok but there are not N branch switches in the
1011  * reflog, it returns 0.
1012  */
1013 int interpret_branch_name(const char *name, struct strbuf *buf)
1014 {
1015         char *cp;
1016         struct branch *upstream;
1017         int namelen = strlen(name);
1018         int len = interpret_nth_prior_checkout(name, buf);
1019         int tmp_len;
1020
1021         if (!len)
1022                 return len; /* syntax Ok, not enough switches */
1023         if (0 < len && len == namelen)
1024                 return len; /* consumed all */
1025         else if (0 < len) {
1026                 /* we have extra data, which might need further processing */
1027                 struct strbuf tmp = STRBUF_INIT;
1028                 int used = buf->len;
1029                 int ret;
1030
1031                 strbuf_add(buf, name + len, namelen - len);
1032                 ret = interpret_branch_name(buf->buf, &tmp);
1033                 /* that data was not interpreted, remove our cruft */
1034                 if (ret < 0) {
1035                         strbuf_setlen(buf, used);
1036                         return len;
1037                 }
1038                 strbuf_reset(buf);
1039                 strbuf_addbuf(buf, &tmp);
1040                 strbuf_release(&tmp);
1041                 /* tweak for size of {-N} versus expanded ref name */
1042                 return ret - used + len;
1043         }
1044
1045         cp = strchr(name, '@');
1046         if (!cp)
1047                 return -1;
1048         tmp_len = upstream_mark(cp, namelen - (cp - name));
1049         if (!tmp_len)
1050                 return -1;
1051         len = cp + tmp_len - name;
1052         cp = xstrndup(name, cp - name);
1053         upstream = branch_get(*cp ? cp : NULL);
1054         /*
1055          * Upstream can be NULL only if cp refers to HEAD and HEAD
1056          * points to something different than a branch.
1057          */
1058         if (!upstream)
1059                 return error(_("HEAD does not point to a branch"));
1060         if (!upstream->merge || !upstream->merge[0]->dst) {
1061                 if (!ref_exists(upstream->refname))
1062                         return error(_("No such branch: '%s'"), cp);
1063                 if (!upstream->merge)
1064                         return error(_("No upstream configured for branch '%s'"),
1065                                      upstream->name);
1066                 return error(
1067                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1068                         upstream->merge[0]->src);
1069         }
1070         free(cp);
1071         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1072         strbuf_reset(buf);
1073         strbuf_addstr(buf, cp);
1074         free(cp);
1075         return len;
1076 }
1077
1078 int strbuf_branchname(struct strbuf *sb, const char *name)
1079 {
1080         int len = strlen(name);
1081         if (interpret_branch_name(name, sb) == len)
1082                 return 0;
1083         strbuf_add(sb, name, len);
1084         return len;
1085 }
1086
1087 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1088 {
1089         strbuf_branchname(sb, name);
1090         if (name[0] == '-')
1091                 return -1;
1092         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1093         return check_refname_format(sb->buf, 0);
1094 }
1095
1096 /*
1097  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1098  * notably "xyz^" for "parent of xyz"
1099  */
1100 int get_sha1(const char *name, unsigned char *sha1)
1101 {
1102         struct object_context unused;
1103         return get_sha1_with_context(name, 0, sha1, &unused);
1104 }
1105
1106 /*
1107  * Many callers know that the user meant to name a committish by
1108  * syntactical positions where the object name appears.  Calling this
1109  * function allows the machinery to disambiguate shorter-than-unique
1110  * abbreviated object names between committish and others.
1111  *
1112  * Note that this does NOT error out when the named object is not a
1113  * committish. It is merely to give a hint to the disambiguation
1114  * machinery.
1115  */
1116 int get_sha1_committish(const char *name, unsigned char *sha1)
1117 {
1118         struct object_context unused;
1119         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1120                                      sha1, &unused);
1121 }
1122
1123 int get_sha1_treeish(const char *name, unsigned char *sha1)
1124 {
1125         struct object_context unused;
1126         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1127                                      sha1, &unused);
1128 }
1129
1130 int get_sha1_commit(const char *name, unsigned char *sha1)
1131 {
1132         struct object_context unused;
1133         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1134                                      sha1, &unused);
1135 }
1136
1137 int get_sha1_tree(const char *name, unsigned char *sha1)
1138 {
1139         struct object_context unused;
1140         return get_sha1_with_context(name, GET_SHA1_TREE,
1141                                      sha1, &unused);
1142 }
1143
1144 int get_sha1_blob(const char *name, unsigned char *sha1)
1145 {
1146         struct object_context unused;
1147         return get_sha1_with_context(name, GET_SHA1_BLOB,
1148                                      sha1, &unused);
1149 }
1150
1151 /* Must be called only when object_name:filename doesn't exist. */
1152 static void diagnose_invalid_sha1_path(const char *prefix,
1153                                        const char *filename,
1154                                        const unsigned char *tree_sha1,
1155                                        const char *object_name,
1156                                        int object_name_len)
1157 {
1158         struct stat st;
1159         unsigned char sha1[20];
1160         unsigned mode;
1161
1162         if (!prefix)
1163                 prefix = "";
1164
1165         if (!lstat(filename, &st))
1166                 die("Path '%s' exists on disk, but not in '%.*s'.",
1167                     filename, object_name_len, object_name);
1168         if (errno == ENOENT || errno == ENOTDIR) {
1169                 char *fullname = xmalloc(strlen(filename)
1170                                              + strlen(prefix) + 1);
1171                 strcpy(fullname, prefix);
1172                 strcat(fullname, filename);
1173
1174                 if (!get_tree_entry(tree_sha1, fullname,
1175                                     sha1, &mode)) {
1176                         die("Path '%s' exists, but not '%s'.\n"
1177                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1178                             fullname,
1179                             filename,
1180                             object_name_len, object_name,
1181                             fullname,
1182                             object_name_len, object_name,
1183                             filename);
1184                 }
1185                 die("Path '%s' does not exist in '%.*s'",
1186                     filename, object_name_len, object_name);
1187         }
1188 }
1189
1190 /* Must be called only when :stage:filename doesn't exist. */
1191 static void diagnose_invalid_index_path(int stage,
1192                                         const char *prefix,
1193                                         const char *filename)
1194 {
1195         struct stat st;
1196         struct cache_entry *ce;
1197         int pos;
1198         unsigned namelen = strlen(filename);
1199         unsigned fullnamelen;
1200         char *fullname;
1201
1202         if (!prefix)
1203                 prefix = "";
1204
1205         /* Wrong stage number? */
1206         pos = cache_name_pos(filename, namelen);
1207         if (pos < 0)
1208                 pos = -pos - 1;
1209         if (pos < active_nr) {
1210                 ce = active_cache[pos];
1211                 if (ce_namelen(ce) == namelen &&
1212                     !memcmp(ce->name, filename, namelen))
1213                         die("Path '%s' is in the index, but not at stage %d.\n"
1214                             "Did you mean ':%d:%s'?",
1215                             filename, stage,
1216                             ce_stage(ce), filename);
1217         }
1218
1219         /* Confusion between relative and absolute filenames? */
1220         fullnamelen = namelen + strlen(prefix);
1221         fullname = xmalloc(fullnamelen + 1);
1222         strcpy(fullname, prefix);
1223         strcat(fullname, filename);
1224         pos = cache_name_pos(fullname, fullnamelen);
1225         if (pos < 0)
1226                 pos = -pos - 1;
1227         if (pos < active_nr) {
1228                 ce = active_cache[pos];
1229                 if (ce_namelen(ce) == fullnamelen &&
1230                     !memcmp(ce->name, fullname, fullnamelen))
1231                         die("Path '%s' is in the index, but not '%s'.\n"
1232                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1233                             fullname, filename,
1234                             ce_stage(ce), fullname,
1235                             ce_stage(ce), filename);
1236         }
1237
1238         if (!lstat(filename, &st))
1239                 die("Path '%s' exists on disk, but not in the index.", filename);
1240         if (errno == ENOENT || errno == ENOTDIR)
1241                 die("Path '%s' does not exist (neither on disk nor in the index).",
1242                     filename);
1243
1244         free(fullname);
1245 }
1246
1247
1248 static char *resolve_relative_path(const char *rel)
1249 {
1250         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1251                 return NULL;
1252
1253         if (!startup_info)
1254                 die("BUG: startup_info struct is not initialized.");
1255
1256         if (!is_inside_work_tree())
1257                 die("relative path syntax can't be used outside working tree.");
1258
1259         /* die() inside prefix_path() if resolved path is outside worktree */
1260         return prefix_path(startup_info->prefix,
1261                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1262                            rel);
1263 }
1264
1265 static int get_sha1_with_context_1(const char *name,
1266                                    unsigned flags,
1267                                    const char *prefix,
1268                                    unsigned char *sha1,
1269                                    struct object_context *oc)
1270 {
1271         int ret, bracket_depth;
1272         int namelen = strlen(name);
1273         const char *cp;
1274         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1275
1276         memset(oc, 0, sizeof(*oc));
1277         oc->mode = S_IFINVALID;
1278         ret = get_sha1_1(name, namelen, sha1, flags);
1279         if (!ret)
1280                 return ret;
1281         /*
1282          * sha1:path --> object name of path in ent sha1
1283          * :path -> object name of absolute path in index
1284          * :./path -> object name of path relative to cwd in index
1285          * :[0-3]:path -> object name of path in index at stage
1286          * :/foo -> recent commit matching foo
1287          */
1288         if (name[0] == ':') {
1289                 int stage = 0;
1290                 struct cache_entry *ce;
1291                 char *new_path = NULL;
1292                 int pos;
1293                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1294                         struct commit_list *list = NULL;
1295                         for_each_ref(handle_one_ref, &list);
1296                         return get_sha1_oneline(name + 2, sha1, list);
1297                 }
1298                 if (namelen < 3 ||
1299                     name[2] != ':' ||
1300                     name[1] < '0' || '3' < name[1])
1301                         cp = name + 1;
1302                 else {
1303                         stage = name[1] - '0';
1304                         cp = name + 3;
1305                 }
1306                 new_path = resolve_relative_path(cp);
1307                 if (!new_path) {
1308                         namelen = namelen - (cp - name);
1309                 } else {
1310                         cp = new_path;
1311                         namelen = strlen(cp);
1312                 }
1313
1314                 strncpy(oc->path, cp,
1315                         sizeof(oc->path));
1316                 oc->path[sizeof(oc->path)-1] = '\0';
1317
1318                 if (!active_cache)
1319                         read_cache();
1320                 pos = cache_name_pos(cp, namelen);
1321                 if (pos < 0)
1322                         pos = -pos - 1;
1323                 while (pos < active_nr) {
1324                         ce = active_cache[pos];
1325                         if (ce_namelen(ce) != namelen ||
1326                             memcmp(ce->name, cp, namelen))
1327                                 break;
1328                         if (ce_stage(ce) == stage) {
1329                                 hashcpy(sha1, ce->sha1);
1330                                 oc->mode = ce->ce_mode;
1331                                 free(new_path);
1332                                 return 0;
1333                         }
1334                         pos++;
1335                 }
1336                 if (only_to_die && name[1] && name[1] != '/')
1337                         diagnose_invalid_index_path(stage, prefix, cp);
1338                 free(new_path);
1339                 return -1;
1340         }
1341         for (cp = name, bracket_depth = 0; *cp; cp++) {
1342                 if (*cp == '{')
1343                         bracket_depth++;
1344                 else if (bracket_depth && *cp == '}')
1345                         bracket_depth--;
1346                 else if (!bracket_depth && *cp == ':')
1347                         break;
1348         }
1349         if (*cp == ':') {
1350                 unsigned char tree_sha1[20];
1351                 int len = cp - name;
1352                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1353                         const char *filename = cp+1;
1354                         char *new_filename = NULL;
1355
1356                         new_filename = resolve_relative_path(filename);
1357                         if (new_filename)
1358                                 filename = new_filename;
1359                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1360                         if (ret && only_to_die) {
1361                                 diagnose_invalid_sha1_path(prefix, filename,
1362                                                            tree_sha1,
1363                                                            name, len);
1364                         }
1365                         hashcpy(oc->tree, tree_sha1);
1366                         strncpy(oc->path, filename,
1367                                 sizeof(oc->path));
1368                         oc->path[sizeof(oc->path)-1] = '\0';
1369
1370                         free(new_filename);
1371                         return ret;
1372                 } else {
1373                         if (only_to_die)
1374                                 die("Invalid object name '%.*s'.", len, name);
1375                 }
1376         }
1377         return ret;
1378 }
1379
1380 /*
1381  * Call this function when you know "name" given by the end user must
1382  * name an object but it doesn't; the function _may_ die with a better
1383  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1384  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1385  * you have a chance to diagnose the error further.
1386  */
1387 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1388 {
1389         struct object_context oc;
1390         unsigned char sha1[20];
1391         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1392 }
1393
1394 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1395 {
1396         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1397 }