Merge branch 'jk/shortlog-tolerate-broken-commit'
[git] / sha1_name.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "tree-walk.h"
7 #include "refs.h"
8 #include "remote.h"
9
10 static int get_sha1_oneline(const char *, unsigned char *, struct commit_list *);
11
12 typedef int (*disambiguate_hint_fn)(const unsigned char *, void *);
13
14 struct disambiguate_state {
15         disambiguate_hint_fn fn;
16         void *cb_data;
17         unsigned char candidate[20];
18         unsigned candidate_exists:1;
19         unsigned candidate_checked:1;
20         unsigned candidate_ok:1;
21         unsigned disambiguate_fn_used:1;
22         unsigned ambiguous:1;
23         unsigned always_call_fn:1;
24 };
25
26 static void update_candidates(struct disambiguate_state *ds, const unsigned char *current)
27 {
28         if (ds->always_call_fn) {
29                 ds->ambiguous = ds->fn(current, ds->cb_data) ? 1 : 0;
30                 return;
31         }
32         if (!ds->candidate_exists) {
33                 /* this is the first candidate */
34                 hashcpy(ds->candidate, current);
35                 ds->candidate_exists = 1;
36                 return;
37         } else if (!hashcmp(ds->candidate, current)) {
38                 /* the same as what we already have seen */
39                 return;
40         }
41
42         if (!ds->fn) {
43                 /* cannot disambiguate between ds->candidate and current */
44                 ds->ambiguous = 1;
45                 return;
46         }
47
48         if (!ds->candidate_checked) {
49                 ds->candidate_ok = ds->fn(ds->candidate, ds->cb_data);
50                 ds->disambiguate_fn_used = 1;
51                 ds->candidate_checked = 1;
52         }
53
54         if (!ds->candidate_ok) {
55                 /* discard the candidate; we know it does not satisfy fn */
56                 hashcpy(ds->candidate, current);
57                 ds->candidate_checked = 0;
58                 return;
59         }
60
61         /* if we reach this point, we know ds->candidate satisfies fn */
62         if (ds->fn(current, ds->cb_data)) {
63                 /*
64                  * if both current and candidate satisfy fn, we cannot
65                  * disambiguate.
66                  */
67                 ds->candidate_ok = 0;
68                 ds->ambiguous = 1;
69         }
70
71         /* otherwise, current can be discarded and candidate is still good */
72 }
73
74 static void find_short_object_filename(int len, const char *hex_pfx, struct disambiguate_state *ds)
75 {
76         struct alternate_object_database *alt;
77         char hex[40];
78         static struct alternate_object_database *fakeent;
79
80         if (!fakeent) {
81                 /*
82                  * Create a "fake" alternate object database that
83                  * points to our own object database, to make it
84                  * easier to get a temporary working space in
85                  * alt->name/alt->base while iterating over the
86                  * object databases including our own.
87                  */
88                 const char *objdir = get_object_directory();
89                 int objdir_len = strlen(objdir);
90                 int entlen = objdir_len + 43;
91                 fakeent = xmalloc(sizeof(*fakeent) + entlen);
92                 memcpy(fakeent->base, objdir, objdir_len);
93                 fakeent->name = fakeent->base + objdir_len + 1;
94                 fakeent->name[-1] = '/';
95         }
96         fakeent->next = alt_odb_list;
97
98         sprintf(hex, "%.2s", hex_pfx);
99         for (alt = fakeent; alt && !ds->ambiguous; alt = alt->next) {
100                 struct dirent *de;
101                 DIR *dir;
102                 sprintf(alt->name, "%.2s/", hex_pfx);
103                 dir = opendir(alt->base);
104                 if (!dir)
105                         continue;
106
107                 while (!ds->ambiguous && (de = readdir(dir)) != NULL) {
108                         unsigned char sha1[20];
109
110                         if (strlen(de->d_name) != 38)
111                                 continue;
112                         if (memcmp(de->d_name, hex_pfx + 2, len - 2))
113                                 continue;
114                         memcpy(hex + 2, de->d_name, 38);
115                         if (!get_sha1_hex(hex, sha1))
116                                 update_candidates(ds, sha1);
117                 }
118                 closedir(dir);
119         }
120 }
121
122 static int match_sha(unsigned len, const unsigned char *a, const unsigned char *b)
123 {
124         do {
125                 if (*a != *b)
126                         return 0;
127                 a++;
128                 b++;
129                 len -= 2;
130         } while (len > 1);
131         if (len)
132                 if ((*a ^ *b) & 0xf0)
133                         return 0;
134         return 1;
135 }
136
137 static void unique_in_pack(int len,
138                           const unsigned char *bin_pfx,
139                            struct packed_git *p,
140                            struct disambiguate_state *ds)
141 {
142         uint32_t num, last, i, first = 0;
143         const unsigned char *current = NULL;
144
145         open_pack_index(p);
146         num = p->num_objects;
147         last = num;
148         while (first < last) {
149                 uint32_t mid = (first + last) / 2;
150                 const unsigned char *current;
151                 int cmp;
152
153                 current = nth_packed_object_sha1(p, mid);
154                 cmp = hashcmp(bin_pfx, current);
155                 if (!cmp) {
156                         first = mid;
157                         break;
158                 }
159                 if (cmp > 0) {
160                         first = mid+1;
161                         continue;
162                 }
163                 last = mid;
164         }
165
166         /*
167          * At this point, "first" is the location of the lowest object
168          * with an object name that could match "bin_pfx".  See if we have
169          * 0, 1 or more objects that actually match(es).
170          */
171         for (i = first; i < num && !ds->ambiguous; i++) {
172                 current = nth_packed_object_sha1(p, i);
173                 if (!match_sha(len, bin_pfx, current))
174                         break;
175                 update_candidates(ds, current);
176         }
177 }
178
179 static void find_short_packed_object(int len, const unsigned char *bin_pfx,
180                                      struct disambiguate_state *ds)
181 {
182         struct packed_git *p;
183
184         prepare_packed_git();
185         for (p = packed_git; p && !ds->ambiguous; p = p->next)
186                 unique_in_pack(len, bin_pfx, p, ds);
187 }
188
189 #define SHORT_NAME_NOT_FOUND (-1)
190 #define SHORT_NAME_AMBIGUOUS (-2)
191
192 static int finish_object_disambiguation(struct disambiguate_state *ds,
193                                         unsigned char *sha1)
194 {
195         if (ds->ambiguous)
196                 return SHORT_NAME_AMBIGUOUS;
197
198         if (!ds->candidate_exists)
199                 return SHORT_NAME_NOT_FOUND;
200
201         if (!ds->candidate_checked)
202                 /*
203                  * If this is the only candidate, there is no point
204                  * calling the disambiguation hint callback.
205                  *
206                  * On the other hand, if the current candidate
207                  * replaced an earlier candidate that did _not_ pass
208                  * the disambiguation hint callback, then we do have
209                  * more than one objects that match the short name
210                  * given, so we should make sure this one matches;
211                  * otherwise, if we discovered this one and the one
212                  * that we previously discarded in the reverse order,
213                  * we would end up showing different results in the
214                  * same repository!
215                  */
216                 ds->candidate_ok = (!ds->disambiguate_fn_used ||
217                                     ds->fn(ds->candidate, ds->cb_data));
218
219         if (!ds->candidate_ok)
220                 return SHORT_NAME_AMBIGUOUS;
221
222         hashcpy(sha1, ds->candidate);
223         return 0;
224 }
225
226 static int disambiguate_commit_only(const unsigned char *sha1, void *cb_data_unused)
227 {
228         int kind = sha1_object_info(sha1, NULL);
229         return kind == OBJ_COMMIT;
230 }
231
232 static int disambiguate_committish_only(const unsigned char *sha1, void *cb_data_unused)
233 {
234         struct object *obj;
235         int kind;
236
237         kind = sha1_object_info(sha1, NULL);
238         if (kind == OBJ_COMMIT)
239                 return 1;
240         if (kind != OBJ_TAG)
241                 return 0;
242
243         /* We need to do this the hard way... */
244         obj = deref_tag(parse_object(sha1), NULL, 0);
245         if (obj && obj->type == OBJ_COMMIT)
246                 return 1;
247         return 0;
248 }
249
250 static int disambiguate_tree_only(const unsigned char *sha1, void *cb_data_unused)
251 {
252         int kind = sha1_object_info(sha1, NULL);
253         return kind == OBJ_TREE;
254 }
255
256 static int disambiguate_treeish_only(const unsigned char *sha1, void *cb_data_unused)
257 {
258         struct object *obj;
259         int kind;
260
261         kind = sha1_object_info(sha1, NULL);
262         if (kind == OBJ_TREE || kind == OBJ_COMMIT)
263                 return 1;
264         if (kind != OBJ_TAG)
265                 return 0;
266
267         /* We need to do this the hard way... */
268         obj = deref_tag(lookup_object(sha1), NULL, 0);
269         if (obj && (obj->type == OBJ_TREE || obj->type == OBJ_COMMIT))
270                 return 1;
271         return 0;
272 }
273
274 static int disambiguate_blob_only(const unsigned char *sha1, void *cb_data_unused)
275 {
276         int kind = sha1_object_info(sha1, NULL);
277         return kind == OBJ_BLOB;
278 }
279
280 static int prepare_prefixes(const char *name, int len,
281                             unsigned char *bin_pfx,
282                             char *hex_pfx)
283 {
284         int i;
285
286         hashclr(bin_pfx);
287         memset(hex_pfx, 'x', 40);
288         for (i = 0; i < len ;i++) {
289                 unsigned char c = name[i];
290                 unsigned char val;
291                 if (c >= '0' && c <= '9')
292                         val = c - '0';
293                 else if (c >= 'a' && c <= 'f')
294                         val = c - 'a' + 10;
295                 else if (c >= 'A' && c <='F') {
296                         val = c - 'A' + 10;
297                         c -= 'A' - 'a';
298                 }
299                 else
300                         return -1;
301                 hex_pfx[i] = c;
302                 if (!(i & 1))
303                         val <<= 4;
304                 bin_pfx[i >> 1] |= val;
305         }
306         return 0;
307 }
308
309 static int get_short_sha1(const char *name, int len, unsigned char *sha1,
310                           unsigned flags)
311 {
312         int status;
313         char hex_pfx[40];
314         unsigned char bin_pfx[20];
315         struct disambiguate_state ds;
316         int quietly = !!(flags & GET_SHA1_QUIETLY);
317
318         if (len < MINIMUM_ABBREV || len > 40)
319                 return -1;
320         if (prepare_prefixes(name, len, bin_pfx, hex_pfx) < 0)
321                 return -1;
322
323         prepare_alt_odb();
324
325         memset(&ds, 0, sizeof(ds));
326         if (flags & GET_SHA1_COMMIT)
327                 ds.fn = disambiguate_commit_only;
328         else if (flags & GET_SHA1_COMMITTISH)
329                 ds.fn = disambiguate_committish_only;
330         else if (flags & GET_SHA1_TREE)
331                 ds.fn = disambiguate_tree_only;
332         else if (flags & GET_SHA1_TREEISH)
333                 ds.fn = disambiguate_treeish_only;
334         else if (flags & GET_SHA1_BLOB)
335                 ds.fn = disambiguate_blob_only;
336
337         find_short_object_filename(len, hex_pfx, &ds);
338         find_short_packed_object(len, bin_pfx, &ds);
339         status = finish_object_disambiguation(&ds, sha1);
340
341         if (!quietly && (status == SHORT_NAME_AMBIGUOUS))
342                 return error("short SHA1 %.*s is ambiguous.", len, hex_pfx);
343         return status;
344 }
345
346
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.objectNameWarning 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 (!prefixcmp(sp, "commit}"))
681                 expected_type = OBJ_COMMIT;
682         else if (!prefixcmp(sp, "tag}"))
683                 expected_type = OBJ_TAG;
684         else if (!prefixcmp(sp, "tree}"))
685                 expected_type = OBJ_TREE;
686         else if (!prefixcmp(sp, "blob}"))
687                 expected_type = OBJ_BLOB;
688         else if (!prefixcmp(sp, "object}"))
689                 expected_type = OBJ_ANY;
690         else if (sp[0] == '}')
691                 expected_type = OBJ_NONE;
692         else if (sp[0] == '/')
693                 expected_type = OBJ_COMMIT;
694         else
695                 return -1;
696
697         if (expected_type == OBJ_COMMIT)
698                 lookup_flags = GET_SHA1_COMMITTISH;
699         else if (expected_type == OBJ_TREE)
700                 lookup_flags = GET_SHA1_TREEISH;
701
702         if (get_sha1_1(name, sp - name - 2, outer, lookup_flags))
703                 return -1;
704
705         o = parse_object(outer);
706         if (!o)
707                 return -1;
708         if (!expected_type) {
709                 o = deref_tag(o, name, sp - name - 2);
710                 if (!o || (!o->parsed && !parse_object(o->sha1)))
711                         return -1;
712                 hashcpy(sha1, o->sha1);
713                 return 0;
714         }
715
716         /*
717          * At this point, the syntax look correct, so
718          * if we do not get the needed object, we should
719          * barf.
720          */
721         o = peel_to_type(name, len, o, expected_type);
722         if (!o)
723                 return -1;
724
725         hashcpy(sha1, o->sha1);
726         if (sp[0] == '/') {
727                 /* "$commit^{/foo}" */
728                 char *prefix;
729                 int ret;
730                 struct commit_list *list = NULL;
731
732                 /*
733                  * $commit^{/}. Some regex implementation may reject.
734                  * We don't need regex anyway. '' pattern always matches.
735                  */
736                 if (sp[1] == '}')
737                         return 0;
738
739                 prefix = xstrndup(sp + 1, name + len - 1 - (sp + 1));
740                 commit_list_insert((struct commit *)o, &list);
741                 ret = get_sha1_oneline(prefix, sha1, list);
742                 free(prefix);
743                 return ret;
744         }
745         return 0;
746 }
747
748 static int get_describe_name(const char *name, int len, unsigned char *sha1)
749 {
750         const char *cp;
751         unsigned flags = GET_SHA1_QUIETLY | GET_SHA1_COMMIT;
752
753         for (cp = name + len - 1; name + 2 <= cp; cp--) {
754                 char ch = *cp;
755                 if (hexval(ch) & ~0377) {
756                         /* We must be looking at g in "SOMETHING-g"
757                          * for it to be describe output.
758                          */
759                         if (ch == 'g' && cp[-1] == '-') {
760                                 cp++;
761                                 len -= cp - name;
762                                 return get_short_sha1(cp, len, sha1, flags);
763                         }
764                 }
765         }
766         return -1;
767 }
768
769 static int get_sha1_1(const char *name, int len, unsigned char *sha1, unsigned lookup_flags)
770 {
771         int ret, has_suffix;
772         const char *cp;
773
774         /*
775          * "name~3" is "name^^^", "name~" is "name~1", and "name^" is "name^1".
776          */
777         has_suffix = 0;
778         for (cp = name + len - 1; name <= cp; cp--) {
779                 int ch = *cp;
780                 if ('0' <= ch && ch <= '9')
781                         continue;
782                 if (ch == '~' || ch == '^')
783                         has_suffix = ch;
784                 break;
785         }
786
787         if (has_suffix) {
788                 int num = 0;
789                 int len1 = cp - name;
790                 cp++;
791                 while (cp < name + len)
792                         num = num * 10 + *cp++ - '0';
793                 if (!num && len1 == len - 1)
794                         num = 1;
795                 if (has_suffix == '^')
796                         return get_parent(name, len1, sha1, num);
797                 /* else if (has_suffix == '~') -- goes without saying */
798                 return get_nth_ancestor(name, len1, sha1, num);
799         }
800
801         ret = peel_onion(name, len, sha1);
802         if (!ret)
803                 return 0;
804
805         ret = get_sha1_basic(name, len, sha1);
806         if (!ret)
807                 return 0;
808
809         /* It could be describe output that is "SOMETHING-gXXXX" */
810         ret = get_describe_name(name, len, sha1);
811         if (!ret)
812                 return 0;
813
814         return get_short_sha1(name, len, sha1, lookup_flags);
815 }
816
817 /*
818  * This interprets names like ':/Initial revision of "git"' by searching
819  * through history and returning the first commit whose message starts
820  * the given regular expression.
821  *
822  * For future extension, ':/!' is reserved. If you want to match a message
823  * beginning with a '!', you have to repeat the exclamation mark.
824  */
825 #define ONELINE_SEEN (1u<<20)
826
827 static int handle_one_ref(const char *path,
828                 const unsigned char *sha1, int flag, void *cb_data)
829 {
830         struct commit_list **list = cb_data;
831         struct object *object = parse_object(sha1);
832         if (!object)
833                 return 0;
834         if (object->type == OBJ_TAG) {
835                 object = deref_tag(object, path, strlen(path));
836                 if (!object)
837                         return 0;
838         }
839         if (object->type != OBJ_COMMIT)
840                 return 0;
841         commit_list_insert_by_date((struct commit *)object, list);
842         return 0;
843 }
844
845 static int get_sha1_oneline(const char *prefix, unsigned char *sha1,
846                             struct commit_list *list)
847 {
848         struct commit_list *backup = NULL, *l;
849         int found = 0;
850         regex_t regex;
851
852         if (prefix[0] == '!') {
853                 if (prefix[1] != '!')
854                         die ("Invalid search pattern: %s", prefix);
855                 prefix++;
856         }
857
858         if (regcomp(&regex, prefix, REG_EXTENDED))
859                 die("Invalid search pattern: %s", prefix);
860
861         for (l = list; l; l = l->next) {
862                 l->item->object.flags |= ONELINE_SEEN;
863                 commit_list_insert(l->item, &backup);
864         }
865         while (list) {
866                 char *p, *to_free = NULL;
867                 struct commit *commit;
868                 enum object_type type;
869                 unsigned long size;
870                 int matches;
871
872                 commit = pop_most_recent_commit(&list, ONELINE_SEEN);
873                 if (!parse_object(commit->object.sha1))
874                         continue;
875                 if (commit->buffer)
876                         p = commit->buffer;
877                 else {
878                         p = read_sha1_file(commit->object.sha1, &type, &size);
879                         if (!p)
880                                 continue;
881                         to_free = p;
882                 }
883
884                 p = strstr(p, "\n\n");
885                 matches = p && !regexec(&regex, p + 2, 0, NULL, 0);
886                 free(to_free);
887
888                 if (matches) {
889                         hashcpy(sha1, commit->object.sha1);
890                         found = 1;
891                         break;
892                 }
893         }
894         regfree(&regex);
895         free_commit_list(list);
896         for (l = backup; l; l = l->next)
897                 clear_commit_marks(l->item, ONELINE_SEEN);
898         free_commit_list(backup);
899         return found ? 0 : -1;
900 }
901
902 struct grab_nth_branch_switch_cbdata {
903         int remaining;
904         struct strbuf buf;
905 };
906
907 static int grab_nth_branch_switch(unsigned char *osha1, unsigned char *nsha1,
908                                   const char *email, unsigned long timestamp, int tz,
909                                   const char *message, void *cb_data)
910 {
911         struct grab_nth_branch_switch_cbdata *cb = cb_data;
912         const char *match = NULL, *target = NULL;
913         size_t len;
914
915         if (!prefixcmp(message, "checkout: moving from ")) {
916                 match = message + strlen("checkout: moving from ");
917                 target = strstr(match, " to ");
918         }
919
920         if (!match || !target)
921                 return 0;
922         if (--(cb->remaining) == 0) {
923                 len = target - match;
924                 strbuf_reset(&cb->buf);
925                 strbuf_add(&cb->buf, match, len);
926                 return 1; /* we are done */
927         }
928         return 0;
929 }
930
931 /*
932  * Parse @{-N} syntax, return the number of characters parsed
933  * if successful; otherwise signal an error with negative value.
934  */
935 static int interpret_nth_prior_checkout(const char *name, struct strbuf *buf)
936 {
937         long nth;
938         int retval;
939         struct grab_nth_branch_switch_cbdata cb;
940         const char *brace;
941         char *num_end;
942
943         if (name[0] != '@' || name[1] != '{' || name[2] != '-')
944                 return -1;
945         brace = strchr(name, '}');
946         if (!brace)
947                 return -1;
948         nth = strtol(name + 3, &num_end, 10);
949         if (num_end != brace)
950                 return -1;
951         if (nth <= 0)
952                 return -1;
953         cb.remaining = nth;
954         strbuf_init(&cb.buf, 20);
955
956         retval = 0;
957         if (0 < for_each_reflog_ent_reverse("HEAD", grab_nth_branch_switch, &cb)) {
958                 strbuf_reset(buf);
959                 strbuf_add(buf, cb.buf.buf, cb.buf.len);
960                 retval = brace - name + 1;
961         }
962
963         strbuf_release(&cb.buf);
964         return retval;
965 }
966
967 int get_sha1_mb(const char *name, unsigned char *sha1)
968 {
969         struct commit *one, *two;
970         struct commit_list *mbs;
971         unsigned char sha1_tmp[20];
972         const char *dots;
973         int st;
974
975         dots = strstr(name, "...");
976         if (!dots)
977                 return get_sha1(name, sha1);
978         if (dots == name)
979                 st = get_sha1("HEAD", sha1_tmp);
980         else {
981                 struct strbuf sb;
982                 strbuf_init(&sb, dots - name);
983                 strbuf_add(&sb, name, dots - name);
984                 st = get_sha1_committish(sb.buf, sha1_tmp);
985                 strbuf_release(&sb);
986         }
987         if (st)
988                 return st;
989         one = lookup_commit_reference_gently(sha1_tmp, 0);
990         if (!one)
991                 return -1;
992
993         if (get_sha1_committish(dots[3] ? (dots + 3) : "HEAD", sha1_tmp))
994                 return -1;
995         two = lookup_commit_reference_gently(sha1_tmp, 0);
996         if (!two)
997                 return -1;
998         mbs = get_merge_bases(one, two, 1);
999         if (!mbs || mbs->next)
1000                 st = -1;
1001         else {
1002                 st = 0;
1003                 hashcpy(sha1, mbs->item->object.sha1);
1004         }
1005         free_commit_list(mbs);
1006         return st;
1007 }
1008
1009 /* parse @something syntax, when 'something' is not {.*} */
1010 static int interpret_empty_at(const char *name, int namelen, int len, struct strbuf *buf)
1011 {
1012         const char *next;
1013
1014         if (len || name[1] == '{')
1015                 return -1;
1016
1017         /* make sure it's a single @, or @@{.*}, not @foo */
1018         next = strchr(name + len + 1, '@');
1019         if (next && next[1] != '{')
1020                 return -1;
1021         if (!next)
1022                 next = name + namelen;
1023         if (next != name + 1)
1024                 return -1;
1025
1026         strbuf_reset(buf);
1027         strbuf_add(buf, "HEAD", 4);
1028         return 1;
1029 }
1030
1031 static int reinterpret(const char *name, int namelen, int len, struct strbuf *buf)
1032 {
1033         /* we have extra data, which might need further processing */
1034         struct strbuf tmp = STRBUF_INIT;
1035         int used = buf->len;
1036         int ret;
1037
1038         strbuf_add(buf, name + len, namelen - len);
1039         ret = interpret_branch_name(buf->buf, buf->len, &tmp);
1040         /* that data was not interpreted, remove our cruft */
1041         if (ret < 0) {
1042                 strbuf_setlen(buf, used);
1043                 return len;
1044         }
1045         strbuf_reset(buf);
1046         strbuf_addbuf(buf, &tmp);
1047         strbuf_release(&tmp);
1048         /* tweak for size of {-N} versus expanded ref name */
1049         return ret - used + len;
1050 }
1051
1052 /*
1053  * This reads short-hand syntax that not only evaluates to a commit
1054  * object name, but also can act as if the end user spelled the name
1055  * of the branch from the command line.
1056  *
1057  * - "@{-N}" finds the name of the Nth previous branch we were on, and
1058  *   places the name of the branch in the given buf and returns the
1059  *   number of characters parsed if successful.
1060  *
1061  * - "<branch>@{upstream}" finds the name of the other ref that
1062  *   <branch> is configured to merge with (missing <branch> defaults
1063  *   to the current branch), and places the name of the branch in the
1064  *   given buf and returns the number of characters parsed if
1065  *   successful.
1066  *
1067  * If the input is not of the accepted format, it returns a negative
1068  * number to signal an error.
1069  *
1070  * If the input was ok but there are not N branch switches in the
1071  * reflog, it returns 0.
1072  */
1073 int interpret_branch_name(const char *name, int namelen, struct strbuf *buf)
1074 {
1075         char *cp;
1076         struct branch *upstream;
1077         int len = interpret_nth_prior_checkout(name, buf);
1078         int tmp_len;
1079
1080         if (!namelen)
1081                 namelen = strlen(name);
1082
1083         if (!len) {
1084                 return len; /* syntax Ok, not enough switches */
1085         } else if (len > 0) {
1086                 if (len == namelen)
1087                         return len; /* consumed all */
1088                 else
1089                         return reinterpret(name, namelen, len, buf);
1090         }
1091
1092         cp = strchr(name, '@');
1093         if (!cp)
1094                 return -1;
1095
1096         len = interpret_empty_at(name, namelen, cp - name, buf);
1097         if (len > 0)
1098                 return reinterpret(name, namelen, len, buf);
1099
1100         tmp_len = upstream_mark(cp, namelen - (cp - name));
1101         if (!tmp_len)
1102                 return -1;
1103
1104         len = cp + tmp_len - name;
1105         cp = xstrndup(name, cp - name);
1106         upstream = branch_get(*cp ? cp : NULL);
1107         /*
1108          * Upstream can be NULL only if cp refers to HEAD and HEAD
1109          * points to something different than a branch.
1110          */
1111         if (!upstream)
1112                 die(_("HEAD does not point to a branch"));
1113         if (!upstream->merge || !upstream->merge[0]->dst) {
1114                 if (!ref_exists(upstream->refname))
1115                         die(_("No such branch: '%s'"), cp);
1116                 if (!upstream->merge) {
1117                         die(_("No upstream configured for branch '%s'"),
1118                                 upstream->name);
1119                 }
1120                 die(
1121                         _("Upstream branch '%s' not stored as a remote-tracking branch"),
1122                         upstream->merge[0]->src);
1123         }
1124         free(cp);
1125         cp = shorten_unambiguous_ref(upstream->merge[0]->dst, 0);
1126         strbuf_reset(buf);
1127         strbuf_addstr(buf, cp);
1128         free(cp);
1129         return len;
1130 }
1131
1132 int strbuf_branchname(struct strbuf *sb, const char *name)
1133 {
1134         int len = strlen(name);
1135         int used = interpret_branch_name(name, len, sb);
1136
1137         if (used == len)
1138                 return 0;
1139         if (used < 0)
1140                 used = 0;
1141         strbuf_add(sb, name + used, len - used);
1142         return len;
1143 }
1144
1145 int strbuf_check_branch_ref(struct strbuf *sb, const char *name)
1146 {
1147         strbuf_branchname(sb, name);
1148         if (name[0] == '-')
1149                 return -1;
1150         strbuf_splice(sb, 0, 0, "refs/heads/", 11);
1151         return check_refname_format(sb->buf, 0);
1152 }
1153
1154 /*
1155  * This is like "get_sha1_basic()", except it allows "sha1 expressions",
1156  * notably "xyz^" for "parent of xyz"
1157  */
1158 int get_sha1(const char *name, unsigned char *sha1)
1159 {
1160         struct object_context unused;
1161         return get_sha1_with_context(name, 0, sha1, &unused);
1162 }
1163
1164 /*
1165  * Many callers know that the user meant to name a commit-ish by
1166  * syntactical positions where the object name appears.  Calling this
1167  * function allows the machinery to disambiguate shorter-than-unique
1168  * abbreviated object names between commit-ish and others.
1169  *
1170  * Note that this does NOT error out when the named object is not a
1171  * commit-ish. It is merely to give a hint to the disambiguation
1172  * machinery.
1173  */
1174 int get_sha1_committish(const char *name, unsigned char *sha1)
1175 {
1176         struct object_context unused;
1177         return get_sha1_with_context(name, GET_SHA1_COMMITTISH,
1178                                      sha1, &unused);
1179 }
1180
1181 int get_sha1_treeish(const char *name, unsigned char *sha1)
1182 {
1183         struct object_context unused;
1184         return get_sha1_with_context(name, GET_SHA1_TREEISH,
1185                                      sha1, &unused);
1186 }
1187
1188 int get_sha1_commit(const char *name, unsigned char *sha1)
1189 {
1190         struct object_context unused;
1191         return get_sha1_with_context(name, GET_SHA1_COMMIT,
1192                                      sha1, &unused);
1193 }
1194
1195 int get_sha1_tree(const char *name, unsigned char *sha1)
1196 {
1197         struct object_context unused;
1198         return get_sha1_with_context(name, GET_SHA1_TREE,
1199                                      sha1, &unused);
1200 }
1201
1202 int get_sha1_blob(const char *name, unsigned char *sha1)
1203 {
1204         struct object_context unused;
1205         return get_sha1_with_context(name, GET_SHA1_BLOB,
1206                                      sha1, &unused);
1207 }
1208
1209 /* Must be called only when object_name:filename doesn't exist. */
1210 static void diagnose_invalid_sha1_path(const char *prefix,
1211                                        const char *filename,
1212                                        const unsigned char *tree_sha1,
1213                                        const char *object_name,
1214                                        int object_name_len)
1215 {
1216         struct stat st;
1217         unsigned char sha1[20];
1218         unsigned mode;
1219
1220         if (!prefix)
1221                 prefix = "";
1222
1223         if (!lstat(filename, &st))
1224                 die("Path '%s' exists on disk, but not in '%.*s'.",
1225                     filename, object_name_len, object_name);
1226         if (errno == ENOENT || errno == ENOTDIR) {
1227                 char *fullname = xmalloc(strlen(filename)
1228                                              + strlen(prefix) + 1);
1229                 strcpy(fullname, prefix);
1230                 strcat(fullname, filename);
1231
1232                 if (!get_tree_entry(tree_sha1, fullname,
1233                                     sha1, &mode)) {
1234                         die("Path '%s' exists, but not '%s'.\n"
1235                             "Did you mean '%.*s:%s' aka '%.*s:./%s'?",
1236                             fullname,
1237                             filename,
1238                             object_name_len, object_name,
1239                             fullname,
1240                             object_name_len, object_name,
1241                             filename);
1242                 }
1243                 die("Path '%s' does not exist in '%.*s'",
1244                     filename, object_name_len, object_name);
1245         }
1246 }
1247
1248 /* Must be called only when :stage:filename doesn't exist. */
1249 static void diagnose_invalid_index_path(int stage,
1250                                         const char *prefix,
1251                                         const char *filename)
1252 {
1253         struct stat st;
1254         const struct cache_entry *ce;
1255         int pos;
1256         unsigned namelen = strlen(filename);
1257         unsigned fullnamelen;
1258         char *fullname;
1259
1260         if (!prefix)
1261                 prefix = "";
1262
1263         /* Wrong stage number? */
1264         pos = cache_name_pos(filename, namelen);
1265         if (pos < 0)
1266                 pos = -pos - 1;
1267         if (pos < active_nr) {
1268                 ce = active_cache[pos];
1269                 if (ce_namelen(ce) == namelen &&
1270                     !memcmp(ce->name, filename, namelen))
1271                         die("Path '%s' is in the index, but not at stage %d.\n"
1272                             "Did you mean ':%d:%s'?",
1273                             filename, stage,
1274                             ce_stage(ce), filename);
1275         }
1276
1277         /* Confusion between relative and absolute filenames? */
1278         fullnamelen = namelen + strlen(prefix);
1279         fullname = xmalloc(fullnamelen + 1);
1280         strcpy(fullname, prefix);
1281         strcat(fullname, filename);
1282         pos = cache_name_pos(fullname, fullnamelen);
1283         if (pos < 0)
1284                 pos = -pos - 1;
1285         if (pos < active_nr) {
1286                 ce = active_cache[pos];
1287                 if (ce_namelen(ce) == fullnamelen &&
1288                     !memcmp(ce->name, fullname, fullnamelen))
1289                         die("Path '%s' is in the index, but not '%s'.\n"
1290                             "Did you mean ':%d:%s' aka ':%d:./%s'?",
1291                             fullname, filename,
1292                             ce_stage(ce), fullname,
1293                             ce_stage(ce), filename);
1294         }
1295
1296         if (!lstat(filename, &st))
1297                 die("Path '%s' exists on disk, but not in the index.", filename);
1298         if (errno == ENOENT || errno == ENOTDIR)
1299                 die("Path '%s' does not exist (neither on disk nor in the index).",
1300                     filename);
1301
1302         free(fullname);
1303 }
1304
1305
1306 static char *resolve_relative_path(const char *rel)
1307 {
1308         if (prefixcmp(rel, "./") && prefixcmp(rel, "../"))
1309                 return NULL;
1310
1311         if (!startup_info)
1312                 die("BUG: startup_info struct is not initialized.");
1313
1314         if (!is_inside_work_tree())
1315                 die("relative path syntax can't be used outside working tree.");
1316
1317         /* die() inside prefix_path() if resolved path is outside worktree */
1318         return prefix_path(startup_info->prefix,
1319                            startup_info->prefix ? strlen(startup_info->prefix) : 0,
1320                            rel);
1321 }
1322
1323 static int get_sha1_with_context_1(const char *name,
1324                                    unsigned flags,
1325                                    const char *prefix,
1326                                    unsigned char *sha1,
1327                                    struct object_context *oc)
1328 {
1329         int ret, bracket_depth;
1330         int namelen = strlen(name);
1331         const char *cp;
1332         int only_to_die = flags & GET_SHA1_ONLY_TO_DIE;
1333
1334         memset(oc, 0, sizeof(*oc));
1335         oc->mode = S_IFINVALID;
1336         ret = get_sha1_1(name, namelen, sha1, flags);
1337         if (!ret)
1338                 return ret;
1339         /*
1340          * sha1:path --> object name of path in ent sha1
1341          * :path -> object name of absolute path in index
1342          * :./path -> object name of path relative to cwd in index
1343          * :[0-3]:path -> object name of path in index at stage
1344          * :/foo -> recent commit matching foo
1345          */
1346         if (name[0] == ':') {
1347                 int stage = 0;
1348                 const struct cache_entry *ce;
1349                 char *new_path = NULL;
1350                 int pos;
1351                 if (!only_to_die && namelen > 2 && name[1] == '/') {
1352                         struct commit_list *list = NULL;
1353                         for_each_ref(handle_one_ref, &list);
1354                         return get_sha1_oneline(name + 2, sha1, list);
1355                 }
1356                 if (namelen < 3 ||
1357                     name[2] != ':' ||
1358                     name[1] < '0' || '3' < name[1])
1359                         cp = name + 1;
1360                 else {
1361                         stage = name[1] - '0';
1362                         cp = name + 3;
1363                 }
1364                 new_path = resolve_relative_path(cp);
1365                 if (!new_path) {
1366                         namelen = namelen - (cp - name);
1367                 } else {
1368                         cp = new_path;
1369                         namelen = strlen(cp);
1370                 }
1371
1372                 strncpy(oc->path, cp,
1373                         sizeof(oc->path));
1374                 oc->path[sizeof(oc->path)-1] = '\0';
1375
1376                 if (!active_cache)
1377                         read_cache();
1378                 pos = cache_name_pos(cp, namelen);
1379                 if (pos < 0)
1380                         pos = -pos - 1;
1381                 while (pos < active_nr) {
1382                         ce = active_cache[pos];
1383                         if (ce_namelen(ce) != namelen ||
1384                             memcmp(ce->name, cp, namelen))
1385                                 break;
1386                         if (ce_stage(ce) == stage) {
1387                                 hashcpy(sha1, ce->sha1);
1388                                 oc->mode = ce->ce_mode;
1389                                 free(new_path);
1390                                 return 0;
1391                         }
1392                         pos++;
1393                 }
1394                 if (only_to_die && name[1] && name[1] != '/')
1395                         diagnose_invalid_index_path(stage, prefix, cp);
1396                 free(new_path);
1397                 return -1;
1398         }
1399         for (cp = name, bracket_depth = 0; *cp; cp++) {
1400                 if (*cp == '{')
1401                         bracket_depth++;
1402                 else if (bracket_depth && *cp == '}')
1403                         bracket_depth--;
1404                 else if (!bracket_depth && *cp == ':')
1405                         break;
1406         }
1407         if (*cp == ':') {
1408                 unsigned char tree_sha1[20];
1409                 int len = cp - name;
1410                 if (!get_sha1_1(name, len, tree_sha1, GET_SHA1_TREEISH)) {
1411                         const char *filename = cp+1;
1412                         char *new_filename = NULL;
1413
1414                         new_filename = resolve_relative_path(filename);
1415                         if (new_filename)
1416                                 filename = new_filename;
1417                         ret = get_tree_entry(tree_sha1, filename, sha1, &oc->mode);
1418                         if (ret && only_to_die) {
1419                                 diagnose_invalid_sha1_path(prefix, filename,
1420                                                            tree_sha1,
1421                                                            name, len);
1422                         }
1423                         hashcpy(oc->tree, tree_sha1);
1424                         strncpy(oc->path, filename,
1425                                 sizeof(oc->path));
1426                         oc->path[sizeof(oc->path)-1] = '\0';
1427
1428                         free(new_filename);
1429                         return ret;
1430                 } else {
1431                         if (only_to_die)
1432                                 die("Invalid object name '%.*s'.", len, name);
1433                 }
1434         }
1435         return ret;
1436 }
1437
1438 /*
1439  * Call this function when you know "name" given by the end user must
1440  * name an object but it doesn't; the function _may_ die with a better
1441  * diagnostic message than "no such object 'name'", e.g. "Path 'doc' does not
1442  * exist in 'HEAD'" when given "HEAD:doc", or it may return in which case
1443  * you have a chance to diagnose the error further.
1444  */
1445 void maybe_die_on_misspelt_object_name(const char *name, const char *prefix)
1446 {
1447         struct object_context oc;
1448         unsigned char sha1[20];
1449         get_sha1_with_context_1(name, GET_SHA1_ONLY_TO_DIE, prefix, sha1, &oc);
1450 }
1451
1452 int get_sha1_with_context(const char *str, unsigned flags, unsigned char *sha1, struct object_context *orc)
1453 {
1454         return get_sha1_with_context_1(str, flags, NULL, sha1, orc);
1455 }