Merge branch 'nd/diff-parseopt-4'
[git] / unpack-trees.c
1 #include "cache.h"
2 #include "argv-array.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "dir.h"
6 #include "tree.h"
7 #include "tree-walk.h"
8 #include "cache-tree.h"
9 #include "unpack-trees.h"
10 #include "progress.h"
11 #include "refs.h"
12 #include "attr.h"
13 #include "split-index.h"
14 #include "dir.h"
15 #include "submodule.h"
16 #include "submodule-config.h"
17 #include "fsmonitor.h"
18 #include "object-store.h"
19 #include "fetch-object.h"
20
21 /*
22  * Error messages expected by scripts out of plumbing commands such as
23  * read-tree.  Non-scripted Porcelain is not required to use these messages
24  * and in fact are encouraged to reword them to better suit their particular
25  * situation better.  See how "git checkout" and "git merge" replaces
26  * them using setup_unpack_trees_porcelain(), for example.
27  */
28 static const char *unpack_plumbing_errors[NB_UNPACK_TREES_ERROR_TYPES] = {
29         /* ERROR_WOULD_OVERWRITE */
30         "Entry '%s' would be overwritten by merge. Cannot merge.",
31
32         /* ERROR_NOT_UPTODATE_FILE */
33         "Entry '%s' not uptodate. Cannot merge.",
34
35         /* ERROR_NOT_UPTODATE_DIR */
36         "Updating '%s' would lose untracked files in it",
37
38         /* ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN */
39         "Untracked working tree file '%s' would be overwritten by merge.",
40
41         /* ERROR_WOULD_LOSE_UNTRACKED_REMOVED */
42         "Untracked working tree file '%s' would be removed by merge.",
43
44         /* ERROR_BIND_OVERLAP */
45         "Entry '%s' overlaps with '%s'.  Cannot bind.",
46
47         /* ERROR_SPARSE_NOT_UPTODATE_FILE */
48         "Entry '%s' not uptodate. Cannot update sparse checkout.",
49
50         /* ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN */
51         "Working tree file '%s' would be overwritten by sparse checkout update.",
52
53         /* ERROR_WOULD_LOSE_ORPHANED_REMOVED */
54         "Working tree file '%s' would be removed by sparse checkout update.",
55
56         /* ERROR_WOULD_LOSE_SUBMODULE */
57         "Submodule '%s' cannot checkout new HEAD.",
58 };
59
60 #define ERRORMSG(o,type) \
61         ( ((o) && (o)->msgs[(type)]) \
62           ? ((o)->msgs[(type)])      \
63           : (unpack_plumbing_errors[(type)]) )
64
65 static const char *super_prefixed(const char *path)
66 {
67         /*
68          * It is necessary and sufficient to have two static buffers
69          * here, as the return value of this function is fed to
70          * error() using the unpack_*_errors[] templates we see above.
71          */
72         static struct strbuf buf[2] = {STRBUF_INIT, STRBUF_INIT};
73         static int super_prefix_len = -1;
74         static unsigned idx = ARRAY_SIZE(buf) - 1;
75
76         if (super_prefix_len < 0) {
77                 const char *super_prefix = get_super_prefix();
78                 if (!super_prefix) {
79                         super_prefix_len = 0;
80                 } else {
81                         int i;
82                         for (i = 0; i < ARRAY_SIZE(buf); i++)
83                                 strbuf_addstr(&buf[i], super_prefix);
84                         super_prefix_len = buf[0].len;
85                 }
86         }
87
88         if (!super_prefix_len)
89                 return path;
90
91         if (++idx >= ARRAY_SIZE(buf))
92                 idx = 0;
93
94         strbuf_setlen(&buf[idx], super_prefix_len);
95         strbuf_addstr(&buf[idx], path);
96
97         return buf[idx].buf;
98 }
99
100 void setup_unpack_trees_porcelain(struct unpack_trees_options *opts,
101                                   const char *cmd)
102 {
103         int i;
104         const char **msgs = opts->msgs;
105         const char *msg;
106
107         argv_array_init(&opts->msgs_to_free);
108
109         if (!strcmp(cmd, "checkout"))
110                 msg = advice_commit_before_merge
111                       ? _("Your local changes to the following files would be overwritten by checkout:\n%%s"
112                           "Please commit your changes or stash them before you switch branches.")
113                       : _("Your local changes to the following files would be overwritten by checkout:\n%%s");
114         else if (!strcmp(cmd, "merge"))
115                 msg = advice_commit_before_merge
116                       ? _("Your local changes to the following files would be overwritten by merge:\n%%s"
117                           "Please commit your changes or stash them before you merge.")
118                       : _("Your local changes to the following files would be overwritten by merge:\n%%s");
119         else
120                 msg = advice_commit_before_merge
121                       ? _("Your local changes to the following files would be overwritten by %s:\n%%s"
122                           "Please commit your changes or stash them before you %s.")
123                       : _("Your local changes to the following files would be overwritten by %s:\n%%s");
124         msgs[ERROR_WOULD_OVERWRITE] = msgs[ERROR_NOT_UPTODATE_FILE] =
125                 argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
126
127         msgs[ERROR_NOT_UPTODATE_DIR] =
128                 _("Updating the following directories would lose untracked files in them:\n%s");
129
130         if (!strcmp(cmd, "checkout"))
131                 msg = advice_commit_before_merge
132                       ? _("The following untracked working tree files would be removed by checkout:\n%%s"
133                           "Please move or remove them before you switch branches.")
134                       : _("The following untracked working tree files would be removed by checkout:\n%%s");
135         else if (!strcmp(cmd, "merge"))
136                 msg = advice_commit_before_merge
137                       ? _("The following untracked working tree files would be removed by merge:\n%%s"
138                           "Please move or remove them before you merge.")
139                       : _("The following untracked working tree files would be removed by merge:\n%%s");
140         else
141                 msg = advice_commit_before_merge
142                       ? _("The following untracked working tree files would be removed by %s:\n%%s"
143                           "Please move or remove them before you %s.")
144                       : _("The following untracked working tree files would be removed by %s:\n%%s");
145         msgs[ERROR_WOULD_LOSE_UNTRACKED_REMOVED] =
146                 argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
147
148         if (!strcmp(cmd, "checkout"))
149                 msg = advice_commit_before_merge
150                       ? _("The following untracked working tree files would be overwritten by checkout:\n%%s"
151                           "Please move or remove them before you switch branches.")
152                       : _("The following untracked working tree files would be overwritten by checkout:\n%%s");
153         else if (!strcmp(cmd, "merge"))
154                 msg = advice_commit_before_merge
155                       ? _("The following untracked working tree files would be overwritten by merge:\n%%s"
156                           "Please move or remove them before you merge.")
157                       : _("The following untracked working tree files would be overwritten by merge:\n%%s");
158         else
159                 msg = advice_commit_before_merge
160                       ? _("The following untracked working tree files would be overwritten by %s:\n%%s"
161                           "Please move or remove them before you %s.")
162                       : _("The following untracked working tree files would be overwritten by %s:\n%%s");
163         msgs[ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN] =
164                 argv_array_pushf(&opts->msgs_to_free, msg, cmd, cmd);
165
166         /*
167          * Special case: ERROR_BIND_OVERLAP refers to a pair of paths, we
168          * cannot easily display it as a list.
169          */
170         msgs[ERROR_BIND_OVERLAP] = _("Entry '%s' overlaps with '%s'.  Cannot bind.");
171
172         msgs[ERROR_SPARSE_NOT_UPTODATE_FILE] =
173                 _("Cannot update sparse checkout: the following entries are not up to date:\n%s");
174         msgs[ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN] =
175                 _("The following working tree files would be overwritten by sparse checkout update:\n%s");
176         msgs[ERROR_WOULD_LOSE_ORPHANED_REMOVED] =
177                 _("The following working tree files would be removed by sparse checkout update:\n%s");
178         msgs[ERROR_WOULD_LOSE_SUBMODULE] =
179                 _("Cannot update submodule:\n%s");
180
181         opts->show_all_errors = 1;
182         /* rejected paths may not have a static buffer */
183         for (i = 0; i < ARRAY_SIZE(opts->unpack_rejects); i++)
184                 opts->unpack_rejects[i].strdup_strings = 1;
185 }
186
187 void clear_unpack_trees_porcelain(struct unpack_trees_options *opts)
188 {
189         argv_array_clear(&opts->msgs_to_free);
190         memset(opts->msgs, 0, sizeof(opts->msgs));
191 }
192
193 static int do_add_entry(struct unpack_trees_options *o, struct cache_entry *ce,
194                          unsigned int set, unsigned int clear)
195 {
196         clear |= CE_HASHED;
197
198         if (set & CE_REMOVE)
199                 set |= CE_WT_REMOVE;
200
201         ce->ce_flags = (ce->ce_flags & ~clear) | set;
202         return add_index_entry(&o->result, ce,
203                                ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
204 }
205
206 static void add_entry(struct unpack_trees_options *o,
207                       const struct cache_entry *ce,
208                       unsigned int set, unsigned int clear)
209 {
210         do_add_entry(o, dup_cache_entry(ce, &o->result), set, clear);
211 }
212
213 /*
214  * add error messages on path <path>
215  * corresponding to the type <e> with the message <msg>
216  * indicating if it should be display in porcelain or not
217  */
218 static int add_rejected_path(struct unpack_trees_options *o,
219                              enum unpack_trees_error_types e,
220                              const char *path)
221 {
222         if (!o->show_all_errors)
223                 return error(ERRORMSG(o, e), super_prefixed(path));
224
225         /*
226          * Otherwise, insert in a list for future display by
227          * display_error_msgs()
228          */
229         string_list_append(&o->unpack_rejects[e], path);
230         return -1;
231 }
232
233 /*
234  * display all the error messages stored in a nice way
235  */
236 static void display_error_msgs(struct unpack_trees_options *o)
237 {
238         int e, i;
239         int something_displayed = 0;
240         for (e = 0; e < NB_UNPACK_TREES_ERROR_TYPES; e++) {
241                 struct string_list *rejects = &o->unpack_rejects[e];
242                 if (rejects->nr > 0) {
243                         struct strbuf path = STRBUF_INIT;
244                         something_displayed = 1;
245                         for (i = 0; i < rejects->nr; i++)
246                                 strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
247                         error(ERRORMSG(o, e), super_prefixed(path.buf));
248                         strbuf_release(&path);
249                 }
250                 string_list_clear(rejects, 0);
251         }
252         if (something_displayed)
253                 fprintf(stderr, _("Aborting\n"));
254 }
255
256 static int check_submodule_move_head(const struct cache_entry *ce,
257                                      const char *old_id,
258                                      const char *new_id,
259                                      struct unpack_trees_options *o)
260 {
261         unsigned flags = SUBMODULE_MOVE_HEAD_DRY_RUN;
262         const struct submodule *sub = submodule_from_ce(ce);
263
264         if (!sub)
265                 return 0;
266
267         if (o->reset)
268                 flags |= SUBMODULE_MOVE_HEAD_FORCE;
269
270         if (submodule_move_head(ce->name, old_id, new_id, flags))
271                 return o->gently ? -1 :
272                                    add_rejected_path(o, ERROR_WOULD_LOSE_SUBMODULE, ce->name);
273         return 0;
274 }
275
276 /*
277  * Preform the loading of the repository's gitmodules file.  This function is
278  * used by 'check_update()' to perform loading of the gitmodules file in two
279  * differnt situations:
280  * (1) before removing entries from the working tree if the gitmodules file has
281  *     been marked for removal.  This situation is specified by 'state' == NULL.
282  * (2) before checking out entries to the working tree if the gitmodules file
283  *     has been marked for update.  This situation is specified by 'state' != NULL.
284  */
285 static void load_gitmodules_file(struct index_state *index,
286                                  struct checkout *state)
287 {
288         int pos = index_name_pos(index, GITMODULES_FILE, strlen(GITMODULES_FILE));
289
290         if (pos >= 0) {
291                 struct cache_entry *ce = index->cache[pos];
292                 if (!state && ce->ce_flags & CE_WT_REMOVE) {
293                         repo_read_gitmodules(the_repository);
294                 } else if (state && (ce->ce_flags & CE_UPDATE)) {
295                         submodule_free(the_repository);
296                         checkout_entry(ce, state, NULL, NULL);
297                         repo_read_gitmodules(the_repository);
298                 }
299         }
300 }
301
302 static struct progress *get_progress(struct unpack_trees_options *o)
303 {
304         unsigned cnt = 0, total = 0;
305         struct index_state *index = &o->result;
306
307         if (!o->update || !o->verbose_update)
308                 return NULL;
309
310         for (; cnt < index->cache_nr; cnt++) {
311                 const struct cache_entry *ce = index->cache[cnt];
312                 if (ce->ce_flags & (CE_UPDATE | CE_WT_REMOVE))
313                         total++;
314         }
315
316         return start_delayed_progress(_("Checking out files"), total);
317 }
318
319 static void setup_collided_checkout_detection(struct checkout *state,
320                                               struct index_state *index)
321 {
322         int i;
323
324         state->clone = 1;
325         for (i = 0; i < index->cache_nr; i++)
326                 index->cache[i]->ce_flags &= ~CE_MATCHED;
327 }
328
329 static void report_collided_checkout(struct index_state *index)
330 {
331         struct string_list list = STRING_LIST_INIT_NODUP;
332         int i;
333
334         for (i = 0; i < index->cache_nr; i++) {
335                 struct cache_entry *ce = index->cache[i];
336
337                 if (!(ce->ce_flags & CE_MATCHED))
338                         continue;
339
340                 string_list_append(&list, ce->name);
341                 ce->ce_flags &= ~CE_MATCHED;
342         }
343
344         list.cmp = fspathcmp;
345         string_list_sort(&list);
346
347         if (list.nr) {
348                 warning(_("the following paths have collided (e.g. case-sensitive paths\n"
349                           "on a case-insensitive filesystem) and only one from the same\n"
350                           "colliding group is in the working tree:\n"));
351
352                 for (i = 0; i < list.nr; i++)
353                         fprintf(stderr, "  '%s'\n", list.items[i].string);
354         }
355
356         string_list_clear(&list, 0);
357 }
358
359 static int check_updates(struct unpack_trees_options *o)
360 {
361         unsigned cnt = 0;
362         int errs = 0;
363         struct progress *progress;
364         struct index_state *index = &o->result;
365         struct checkout state = CHECKOUT_INIT;
366         int i;
367
368         trace_performance_enter();
369         state.force = 1;
370         state.quiet = 1;
371         state.refresh_cache = 1;
372         state.istate = index;
373
374         if (o->clone)
375                 setup_collided_checkout_detection(&state, index);
376
377         progress = get_progress(o);
378
379         if (o->update)
380                 git_attr_set_direction(GIT_ATTR_CHECKOUT);
381
382         if (should_update_submodules() && o->update && !o->dry_run)
383                 load_gitmodules_file(index, NULL);
384
385         for (i = 0; i < index->cache_nr; i++) {
386                 const struct cache_entry *ce = index->cache[i];
387
388                 if (ce->ce_flags & CE_WT_REMOVE) {
389                         display_progress(progress, ++cnt);
390                         if (o->update && !o->dry_run)
391                                 unlink_entry(ce);
392                 }
393         }
394         remove_marked_cache_entries(index, 0);
395         remove_scheduled_dirs();
396
397         if (should_update_submodules() && o->update && !o->dry_run)
398                 load_gitmodules_file(index, &state);
399
400         enable_delayed_checkout(&state);
401         if (repository_format_partial_clone && o->update && !o->dry_run) {
402                 /*
403                  * Prefetch the objects that are to be checked out in the loop
404                  * below.
405                  */
406                 struct oid_array to_fetch = OID_ARRAY_INIT;
407                 int fetch_if_missing_store = fetch_if_missing;
408                 fetch_if_missing = 0;
409                 for (i = 0; i < index->cache_nr; i++) {
410                         struct cache_entry *ce = index->cache[i];
411                         if ((ce->ce_flags & CE_UPDATE) &&
412                             !S_ISGITLINK(ce->ce_mode)) {
413                                 if (!has_object_file(&ce->oid))
414                                         oid_array_append(&to_fetch, &ce->oid);
415                         }
416                 }
417                 if (to_fetch.nr)
418                         fetch_objects(repository_format_partial_clone,
419                                       to_fetch.oid, to_fetch.nr);
420                 fetch_if_missing = fetch_if_missing_store;
421                 oid_array_clear(&to_fetch);
422         }
423         for (i = 0; i < index->cache_nr; i++) {
424                 struct cache_entry *ce = index->cache[i];
425
426                 if (ce->ce_flags & CE_UPDATE) {
427                         if (ce->ce_flags & CE_WT_REMOVE)
428                                 BUG("both update and delete flags are set on %s",
429                                     ce->name);
430                         display_progress(progress, ++cnt);
431                         ce->ce_flags &= ~CE_UPDATE;
432                         if (o->update && !o->dry_run) {
433                                 errs |= checkout_entry(ce, &state, NULL, NULL);
434                         }
435                 }
436         }
437         stop_progress(&progress);
438         errs |= finish_delayed_checkout(&state, NULL);
439         if (o->update)
440                 git_attr_set_direction(GIT_ATTR_CHECKIN);
441
442         if (o->clone)
443                 report_collided_checkout(index);
444
445         trace_performance_leave("check_updates");
446         return errs != 0;
447 }
448
449 static int verify_uptodate_sparse(const struct cache_entry *ce,
450                                   struct unpack_trees_options *o);
451 static int verify_absent_sparse(const struct cache_entry *ce,
452                                 enum unpack_trees_error_types,
453                                 struct unpack_trees_options *o);
454
455 static int apply_sparse_checkout(struct index_state *istate,
456                                  struct cache_entry *ce,
457                                  struct unpack_trees_options *o)
458 {
459         int was_skip_worktree = ce_skip_worktree(ce);
460
461         if (ce->ce_flags & CE_NEW_SKIP_WORKTREE)
462                 ce->ce_flags |= CE_SKIP_WORKTREE;
463         else
464                 ce->ce_flags &= ~CE_SKIP_WORKTREE;
465         if (was_skip_worktree != ce_skip_worktree(ce)) {
466                 ce->ce_flags |= CE_UPDATE_IN_BASE;
467                 mark_fsmonitor_invalid(istate, ce);
468                 istate->cache_changed |= CE_ENTRY_CHANGED;
469         }
470
471         /*
472          * if (!was_skip_worktree && !ce_skip_worktree()) {
473          *      This is perfectly normal. Move on;
474          * }
475          */
476
477         /*
478          * Merge strategies may set CE_UPDATE|CE_REMOVE outside checkout
479          * area as a result of ce_skip_worktree() shortcuts in
480          * verify_absent() and verify_uptodate().
481          * Make sure they don't modify worktree if they are already
482          * outside checkout area
483          */
484         if (was_skip_worktree && ce_skip_worktree(ce)) {
485                 ce->ce_flags &= ~CE_UPDATE;
486
487                 /*
488                  * By default, when CE_REMOVE is on, CE_WT_REMOVE is also
489                  * on to get that file removed from both index and worktree.
490                  * If that file is already outside worktree area, don't
491                  * bother remove it.
492                  */
493                 if (ce->ce_flags & CE_REMOVE)
494                         ce->ce_flags &= ~CE_WT_REMOVE;
495         }
496
497         if (!was_skip_worktree && ce_skip_worktree(ce)) {
498                 /*
499                  * If CE_UPDATE is set, verify_uptodate() must be called already
500                  * also stat info may have lost after merged_entry() so calling
501                  * verify_uptodate() again may fail
502                  */
503                 if (!(ce->ce_flags & CE_UPDATE) && verify_uptodate_sparse(ce, o))
504                         return -1;
505                 ce->ce_flags |= CE_WT_REMOVE;
506                 ce->ce_flags &= ~CE_UPDATE;
507         }
508         if (was_skip_worktree && !ce_skip_worktree(ce)) {
509                 if (verify_absent_sparse(ce, ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o))
510                         return -1;
511                 ce->ce_flags |= CE_UPDATE;
512         }
513         return 0;
514 }
515
516 static inline int call_unpack_fn(const struct cache_entry * const *src,
517                                  struct unpack_trees_options *o)
518 {
519         int ret = o->fn(src, o);
520         if (ret > 0)
521                 ret = 0;
522         return ret;
523 }
524
525 static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
526 {
527         ce->ce_flags |= CE_UNPACKED;
528
529         if (o->cache_bottom < o->src_index->cache_nr &&
530             o->src_index->cache[o->cache_bottom] == ce) {
531                 int bottom = o->cache_bottom;
532                 while (bottom < o->src_index->cache_nr &&
533                        o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
534                         bottom++;
535                 o->cache_bottom = bottom;
536         }
537 }
538
539 static void mark_all_ce_unused(struct index_state *index)
540 {
541         int i;
542         for (i = 0; i < index->cache_nr; i++)
543                 index->cache[i]->ce_flags &= ~(CE_UNPACKED | CE_ADDED | CE_NEW_SKIP_WORKTREE);
544 }
545
546 static int locate_in_src_index(const struct cache_entry *ce,
547                                struct unpack_trees_options *o)
548 {
549         struct index_state *index = o->src_index;
550         int len = ce_namelen(ce);
551         int pos = index_name_pos(index, ce->name, len);
552         if (pos < 0)
553                 pos = -1 - pos;
554         return pos;
555 }
556
557 /*
558  * We call unpack_index_entry() with an unmerged cache entry
559  * only in diff-index, and it wants a single callback.  Skip
560  * the other unmerged entry with the same name.
561  */
562 static void mark_ce_used_same_name(struct cache_entry *ce,
563                                    struct unpack_trees_options *o)
564 {
565         struct index_state *index = o->src_index;
566         int len = ce_namelen(ce);
567         int pos;
568
569         for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
570                 struct cache_entry *next = index->cache[pos];
571                 if (len != ce_namelen(next) ||
572                     memcmp(ce->name, next->name, len))
573                         break;
574                 mark_ce_used(next, o);
575         }
576 }
577
578 static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
579 {
580         const struct index_state *index = o->src_index;
581         int pos = o->cache_bottom;
582
583         while (pos < index->cache_nr) {
584                 struct cache_entry *ce = index->cache[pos];
585                 if (!(ce->ce_flags & CE_UNPACKED))
586                         return ce;
587                 pos++;
588         }
589         return NULL;
590 }
591
592 static void add_same_unmerged(const struct cache_entry *ce,
593                               struct unpack_trees_options *o)
594 {
595         struct index_state *index = o->src_index;
596         int len = ce_namelen(ce);
597         int pos = index_name_pos(index, ce->name, len);
598
599         if (0 <= pos)
600                 die("programming error in a caller of mark_ce_used_same_name");
601         for (pos = -pos - 1; pos < index->cache_nr; pos++) {
602                 struct cache_entry *next = index->cache[pos];
603                 if (len != ce_namelen(next) ||
604                     memcmp(ce->name, next->name, len))
605                         break;
606                 add_entry(o, next, 0, 0);
607                 mark_ce_used(next, o);
608         }
609 }
610
611 static int unpack_index_entry(struct cache_entry *ce,
612                               struct unpack_trees_options *o)
613 {
614         const struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
615         int ret;
616
617         src[0] = ce;
618
619         mark_ce_used(ce, o);
620         if (ce_stage(ce)) {
621                 if (o->skip_unmerged) {
622                         add_entry(o, ce, 0, 0);
623                         return 0;
624                 }
625         }
626         ret = call_unpack_fn(src, o);
627         if (ce_stage(ce))
628                 mark_ce_used_same_name(ce, o);
629         return ret;
630 }
631
632 static int find_cache_pos(struct traverse_info *, const struct name_entry *);
633
634 static void restore_cache_bottom(struct traverse_info *info, int bottom)
635 {
636         struct unpack_trees_options *o = info->data;
637
638         if (o->diff_index_cached)
639                 return;
640         o->cache_bottom = bottom;
641 }
642
643 static int switch_cache_bottom(struct traverse_info *info)
644 {
645         struct unpack_trees_options *o = info->data;
646         int ret, pos;
647
648         if (o->diff_index_cached)
649                 return 0;
650         ret = o->cache_bottom;
651         pos = find_cache_pos(info->prev, &info->name);
652
653         if (pos < -1)
654                 o->cache_bottom = -2 - pos;
655         else if (pos < 0)
656                 o->cache_bottom = o->src_index->cache_nr;
657         return ret;
658 }
659
660 static inline int are_same_oid(struct name_entry *name_j, struct name_entry *name_k)
661 {
662         return !is_null_oid(&name_j->oid) && !is_null_oid(&name_k->oid) && oideq(&name_j->oid, &name_k->oid);
663 }
664
665 static int all_trees_same_as_cache_tree(int n, unsigned long dirmask,
666                                         struct name_entry *names,
667                                         struct traverse_info *info)
668 {
669         struct unpack_trees_options *o = info->data;
670         int i;
671
672         if (!o->merge || dirmask != ((1 << n) - 1))
673                 return 0;
674
675         for (i = 1; i < n; i++)
676                 if (!are_same_oid(names, names + i))
677                         return 0;
678
679         return cache_tree_matches_traversal(o->src_index->cache_tree, names, info);
680 }
681
682 static int index_pos_by_traverse_info(struct name_entry *names,
683                                       struct traverse_info *info)
684 {
685         struct unpack_trees_options *o = info->data;
686         int len = traverse_path_len(info, names);
687         char *name = xmalloc(len + 1 /* slash */ + 1 /* NUL */);
688         int pos;
689
690         make_traverse_path(name, info, names);
691         name[len++] = '/';
692         name[len] = '\0';
693         pos = index_name_pos(o->src_index, name, len);
694         if (pos >= 0)
695                 BUG("This is a directory and should not exist in index");
696         pos = -pos - 1;
697         if (!starts_with(o->src_index->cache[pos]->name, name) ||
698             (pos > 0 && starts_with(o->src_index->cache[pos-1]->name, name)))
699                 BUG("pos must point at the first entry in this directory");
700         free(name);
701         return pos;
702 }
703
704 /*
705  * Fast path if we detect that all trees are the same as cache-tree at this
706  * path. We'll walk these trees in an iterative loop using cache-tree/index
707  * instead of ODB since we already know what these trees contain.
708  */
709 static int traverse_by_cache_tree(int pos, int nr_entries, int nr_names,
710                                   struct traverse_info *info)
711 {
712         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
713         struct unpack_trees_options *o = info->data;
714         struct cache_entry *tree_ce = NULL;
715         int ce_len = 0;
716         int i, d;
717
718         if (!o->merge)
719                 BUG("We need cache-tree to do this optimization");
720
721         /*
722          * Do what unpack_callback() and unpack_nondirectories() normally
723          * do. But we walk all paths in an iterative loop instead.
724          *
725          * D/F conflicts and higher stage entries are not a concern
726          * because cache-tree would be invalidated and we would never
727          * get here in the first place.
728          */
729         for (i = 0; i < nr_entries; i++) {
730                 int new_ce_len, len, rc;
731
732                 src[0] = o->src_index->cache[pos + i];
733
734                 len = ce_namelen(src[0]);
735                 new_ce_len = cache_entry_size(len);
736
737                 if (new_ce_len > ce_len) {
738                         new_ce_len <<= 1;
739                         tree_ce = xrealloc(tree_ce, new_ce_len);
740                         memset(tree_ce, 0, new_ce_len);
741                         ce_len = new_ce_len;
742
743                         tree_ce->ce_flags = create_ce_flags(0);
744
745                         for (d = 1; d <= nr_names; d++)
746                                 src[d] = tree_ce;
747                 }
748
749                 tree_ce->ce_mode = src[0]->ce_mode;
750                 tree_ce->ce_namelen = len;
751                 oidcpy(&tree_ce->oid, &src[0]->oid);
752                 memcpy(tree_ce->name, src[0]->name, len + 1);
753
754                 rc = call_unpack_fn((const struct cache_entry * const *)src, o);
755                 if (rc < 0) {
756                         free(tree_ce);
757                         return rc;
758                 }
759
760                 mark_ce_used(src[0], o);
761         }
762         free(tree_ce);
763         if (o->debug_unpack)
764                 printf("Unpacked %d entries from %s to %s using cache-tree\n",
765                        nr_entries,
766                        o->src_index->cache[pos]->name,
767                        o->src_index->cache[pos + nr_entries - 1]->name);
768         return 0;
769 }
770
771 static int traverse_trees_recursive(int n, unsigned long dirmask,
772                                     unsigned long df_conflicts,
773                                     struct name_entry *names,
774                                     struct traverse_info *info)
775 {
776         struct unpack_trees_options *o = info->data;
777         int i, ret, bottom;
778         int nr_buf = 0;
779         struct tree_desc t[MAX_UNPACK_TREES];
780         void *buf[MAX_UNPACK_TREES];
781         struct traverse_info newinfo;
782         struct name_entry *p;
783         int nr_entries;
784
785         nr_entries = all_trees_same_as_cache_tree(n, dirmask, names, info);
786         if (nr_entries > 0) {
787                 int pos = index_pos_by_traverse_info(names, info);
788
789                 if (!o->merge || df_conflicts)
790                         BUG("Wrong condition to get here buddy");
791
792                 /*
793                  * All entries up to 'pos' must have been processed
794                  * (i.e. marked CE_UNPACKED) at this point. But to be safe,
795                  * save and restore cache_bottom anyway to not miss
796                  * unprocessed entries before 'pos'.
797                  */
798                 bottom = o->cache_bottom;
799                 ret = traverse_by_cache_tree(pos, nr_entries, n, info);
800                 o->cache_bottom = bottom;
801                 return ret;
802         }
803
804         p = names;
805         while (!p->mode)
806                 p++;
807
808         newinfo = *info;
809         newinfo.prev = info;
810         newinfo.pathspec = info->pathspec;
811         newinfo.name = *p;
812         newinfo.pathlen += tree_entry_len(p) + 1;
813         newinfo.df_conflicts |= df_conflicts;
814
815         /*
816          * Fetch the tree from the ODB for each peer directory in the
817          * n commits.
818          *
819          * For 2- and 3-way traversals, we try to avoid hitting the
820          * ODB twice for the same OID.  This should yield a nice speed
821          * up in checkouts and merges when the commits are similar.
822          *
823          * We don't bother doing the full O(n^2) search for larger n,
824          * because wider traversals don't happen that often and we
825          * avoid the search setup.
826          *
827          * When 2 peer OIDs are the same, we just copy the tree
828          * descriptor data.  This implicitly borrows the buffer
829          * data from the earlier cell.
830          */
831         for (i = 0; i < n; i++, dirmask >>= 1) {
832                 if (i > 0 && are_same_oid(&names[i], &names[i - 1]))
833                         t[i] = t[i - 1];
834                 else if (i > 1 && are_same_oid(&names[i], &names[i - 2]))
835                         t[i] = t[i - 2];
836                 else {
837                         const struct object_id *oid = NULL;
838                         if (dirmask & 1)
839                                 oid = &names[i].oid;
840                         buf[nr_buf++] = fill_tree_descriptor(t + i, oid);
841                 }
842         }
843
844         bottom = switch_cache_bottom(&newinfo);
845         ret = traverse_trees(o->src_index, n, t, &newinfo);
846         restore_cache_bottom(&newinfo, bottom);
847
848         for (i = 0; i < nr_buf; i++)
849                 free(buf[i]);
850
851         return ret;
852 }
853
854 /*
855  * Compare the traverse-path to the cache entry without actually
856  * having to generate the textual representation of the traverse
857  * path.
858  *
859  * NOTE! This *only* compares up to the size of the traverse path
860  * itself - the caller needs to do the final check for the cache
861  * entry having more data at the end!
862  */
863 static int do_compare_entry_piecewise(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
864 {
865         int len, pathlen, ce_len;
866         const char *ce_name;
867
868         if (info->prev) {
869                 int cmp = do_compare_entry_piecewise(ce, info->prev,
870                                                      &info->name);
871                 if (cmp)
872                         return cmp;
873         }
874         pathlen = info->pathlen;
875         ce_len = ce_namelen(ce);
876
877         /* If ce_len < pathlen then we must have previously hit "name == directory" entry */
878         if (ce_len < pathlen)
879                 return -1;
880
881         ce_len -= pathlen;
882         ce_name = ce->name + pathlen;
883
884         len = tree_entry_len(n);
885         return df_name_compare(ce_name, ce_len, S_IFREG, n->path, len, n->mode);
886 }
887
888 static int do_compare_entry(const struct cache_entry *ce,
889                             const struct traverse_info *info,
890                             const struct name_entry *n)
891 {
892         int len, pathlen, ce_len;
893         const char *ce_name;
894         int cmp;
895
896         /*
897          * If we have not precomputed the traverse path, it is quicker
898          * to avoid doing so.  But if we have precomputed it,
899          * it is quicker to use the precomputed version.
900          */
901         if (!info->traverse_path)
902                 return do_compare_entry_piecewise(ce, info, n);
903
904         cmp = strncmp(ce->name, info->traverse_path, info->pathlen);
905         if (cmp)
906                 return cmp;
907
908         pathlen = info->pathlen;
909         ce_len = ce_namelen(ce);
910
911         if (ce_len < pathlen)
912                 return -1;
913
914         ce_len -= pathlen;
915         ce_name = ce->name + pathlen;
916
917         len = tree_entry_len(n);
918         return df_name_compare(ce_name, ce_len, S_IFREG, n->path, len, n->mode);
919 }
920
921 static int compare_entry(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
922 {
923         int cmp = do_compare_entry(ce, info, n);
924         if (cmp)
925                 return cmp;
926
927         /*
928          * Even if the beginning compared identically, the ce should
929          * compare as bigger than a directory leading up to it!
930          */
931         return ce_namelen(ce) > traverse_path_len(info, n);
932 }
933
934 static int ce_in_traverse_path(const struct cache_entry *ce,
935                                const struct traverse_info *info)
936 {
937         if (!info->prev)
938                 return 1;
939         if (do_compare_entry(ce, info->prev, &info->name))
940                 return 0;
941         /*
942          * If ce (blob) is the same name as the path (which is a tree
943          * we will be descending into), it won't be inside it.
944          */
945         return (info->pathlen < ce_namelen(ce));
946 }
947
948 static struct cache_entry *create_ce_entry(const struct traverse_info *info,
949         const struct name_entry *n,
950         int stage,
951         struct index_state *istate,
952         int is_transient)
953 {
954         int len = traverse_path_len(info, n);
955         struct cache_entry *ce =
956                 is_transient ?
957                 make_empty_transient_cache_entry(len) :
958                 make_empty_cache_entry(istate, len);
959
960         ce->ce_mode = create_ce_mode(n->mode);
961         ce->ce_flags = create_ce_flags(stage);
962         ce->ce_namelen = len;
963         oidcpy(&ce->oid, &n->oid);
964         make_traverse_path(ce->name, info, n);
965
966         return ce;
967 }
968
969 /*
970  * Note that traverse_by_cache_tree() duplicates some logic in this function
971  * without actually calling it. If you change the logic here you may need to
972  * check and change there as well.
973  */
974 static int unpack_nondirectories(int n, unsigned long mask,
975                                  unsigned long dirmask,
976                                  struct cache_entry **src,
977                                  const struct name_entry *names,
978                                  const struct traverse_info *info)
979 {
980         int i;
981         struct unpack_trees_options *o = info->data;
982         unsigned long conflicts = info->df_conflicts | dirmask;
983
984         /* Do we have *only* directories? Nothing to do */
985         if (mask == dirmask && !src[0])
986                 return 0;
987
988         /*
989          * Ok, we've filled in up to any potential index entry in src[0],
990          * now do the rest.
991          */
992         for (i = 0; i < n; i++) {
993                 int stage;
994                 unsigned int bit = 1ul << i;
995                 if (conflicts & bit) {
996                         src[i + o->merge] = o->df_conflict_entry;
997                         continue;
998                 }
999                 if (!(mask & bit))
1000                         continue;
1001                 if (!o->merge)
1002                         stage = 0;
1003                 else if (i + 1 < o->head_idx)
1004                         stage = 1;
1005                 else if (i + 1 > o->head_idx)
1006                         stage = 3;
1007                 else
1008                         stage = 2;
1009
1010                 /*
1011                  * If the merge bit is set, then the cache entries are
1012                  * discarded in the following block.  In this case,
1013                  * construct "transient" cache_entries, as they are
1014                  * not stored in the index.  otherwise construct the
1015                  * cache entry from the index aware logic.
1016                  */
1017                 src[i + o->merge] = create_ce_entry(info, names + i, stage, &o->result, o->merge);
1018         }
1019
1020         if (o->merge) {
1021                 int rc = call_unpack_fn((const struct cache_entry * const *)src,
1022                                         o);
1023                 for (i = 0; i < n; i++) {
1024                         struct cache_entry *ce = src[i + o->merge];
1025                         if (ce != o->df_conflict_entry)
1026                                 discard_cache_entry(ce);
1027                 }
1028                 return rc;
1029         }
1030
1031         for (i = 0; i < n; i++)
1032                 if (src[i] && src[i] != o->df_conflict_entry)
1033                         if (do_add_entry(o, src[i], 0, 0))
1034                                 return -1;
1035
1036         return 0;
1037 }
1038
1039 static int unpack_failed(struct unpack_trees_options *o, const char *message)
1040 {
1041         discard_index(&o->result);
1042         if (!o->gently && !o->exiting_early) {
1043                 if (message)
1044                         return error("%s", message);
1045                 return -1;
1046         }
1047         return -1;
1048 }
1049
1050 /*
1051  * The tree traversal is looking at name p.  If we have a matching entry,
1052  * return it.  If name p is a directory in the index, do not return
1053  * anything, as we will want to match it when the traversal descends into
1054  * the directory.
1055  */
1056 static int find_cache_pos(struct traverse_info *info,
1057                           const struct name_entry *p)
1058 {
1059         int pos;
1060         struct unpack_trees_options *o = info->data;
1061         struct index_state *index = o->src_index;
1062         int pfxlen = info->pathlen;
1063         int p_len = tree_entry_len(p);
1064
1065         for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
1066                 const struct cache_entry *ce = index->cache[pos];
1067                 const char *ce_name, *ce_slash;
1068                 int cmp, ce_len;
1069
1070                 if (ce->ce_flags & CE_UNPACKED) {
1071                         /*
1072                          * cache_bottom entry is already unpacked, so
1073                          * we can never match it; don't check it
1074                          * again.
1075                          */
1076                         if (pos == o->cache_bottom)
1077                                 ++o->cache_bottom;
1078                         continue;
1079                 }
1080                 if (!ce_in_traverse_path(ce, info)) {
1081                         /*
1082                          * Check if we can skip future cache checks
1083                          * (because we're already past all possible
1084                          * entries in the traverse path).
1085                          */
1086                         if (info->traverse_path) {
1087                                 if (strncmp(ce->name, info->traverse_path,
1088                                             info->pathlen) > 0)
1089                                         break;
1090                         }
1091                         continue;
1092                 }
1093                 ce_name = ce->name + pfxlen;
1094                 ce_slash = strchr(ce_name, '/');
1095                 if (ce_slash)
1096                         ce_len = ce_slash - ce_name;
1097                 else
1098                         ce_len = ce_namelen(ce) - pfxlen;
1099                 cmp = name_compare(p->path, p_len, ce_name, ce_len);
1100                 /*
1101                  * Exact match; if we have a directory we need to
1102                  * delay returning it.
1103                  */
1104                 if (!cmp)
1105                         return ce_slash ? -2 - pos : pos;
1106                 if (0 < cmp)
1107                         continue; /* keep looking */
1108                 /*
1109                  * ce_name sorts after p->path; could it be that we
1110                  * have files under p->path directory in the index?
1111                  * E.g.  ce_name == "t-i", and p->path == "t"; we may
1112                  * have "t/a" in the index.
1113                  */
1114                 if (p_len < ce_len && !memcmp(ce_name, p->path, p_len) &&
1115                     ce_name[p_len] < '/')
1116                         continue; /* keep looking */
1117                 break;
1118         }
1119         return -1;
1120 }
1121
1122 static struct cache_entry *find_cache_entry(struct traverse_info *info,
1123                                             const struct name_entry *p)
1124 {
1125         int pos = find_cache_pos(info, p);
1126         struct unpack_trees_options *o = info->data;
1127
1128         if (0 <= pos)
1129                 return o->src_index->cache[pos];
1130         else
1131                 return NULL;
1132 }
1133
1134 static void debug_path(struct traverse_info *info)
1135 {
1136         if (info->prev) {
1137                 debug_path(info->prev);
1138                 if (*info->prev->name.path)
1139                         putchar('/');
1140         }
1141         printf("%s", info->name.path);
1142 }
1143
1144 static void debug_name_entry(int i, struct name_entry *n)
1145 {
1146         printf("ent#%d %06o %s\n", i,
1147                n->path ? n->mode : 0,
1148                n->path ? n->path : "(missing)");
1149 }
1150
1151 static void debug_unpack_callback(int n,
1152                                   unsigned long mask,
1153                                   unsigned long dirmask,
1154                                   struct name_entry *names,
1155                                   struct traverse_info *info)
1156 {
1157         int i;
1158         printf("* unpack mask %lu, dirmask %lu, cnt %d ",
1159                mask, dirmask, n);
1160         debug_path(info);
1161         putchar('\n');
1162         for (i = 0; i < n; i++)
1163                 debug_name_entry(i, names + i);
1164 }
1165
1166 /*
1167  * Note that traverse_by_cache_tree() duplicates some logic in this function
1168  * without actually calling it. If you change the logic here you may need to
1169  * check and change there as well.
1170  */
1171 static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
1172 {
1173         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
1174         struct unpack_trees_options *o = info->data;
1175         const struct name_entry *p = names;
1176
1177         /* Find first entry with a real name (we could use "mask" too) */
1178         while (!p->mode)
1179                 p++;
1180
1181         if (o->debug_unpack)
1182                 debug_unpack_callback(n, mask, dirmask, names, info);
1183
1184         /* Are we supposed to look at the index too? */
1185         if (o->merge) {
1186                 while (1) {
1187                         int cmp;
1188                         struct cache_entry *ce;
1189
1190                         if (o->diff_index_cached)
1191                                 ce = next_cache_entry(o);
1192                         else
1193                                 ce = find_cache_entry(info, p);
1194
1195                         if (!ce)
1196                                 break;
1197                         cmp = compare_entry(ce, info, p);
1198                         if (cmp < 0) {
1199                                 if (unpack_index_entry(ce, o) < 0)
1200                                         return unpack_failed(o, NULL);
1201                                 continue;
1202                         }
1203                         if (!cmp) {
1204                                 if (ce_stage(ce)) {
1205                                         /*
1206                                          * If we skip unmerged index
1207                                          * entries, we'll skip this
1208                                          * entry *and* the tree
1209                                          * entries associated with it!
1210                                          */
1211                                         if (o->skip_unmerged) {
1212                                                 add_same_unmerged(ce, o);
1213                                                 return mask;
1214                                         }
1215                                 }
1216                                 src[0] = ce;
1217                         }
1218                         break;
1219                 }
1220         }
1221
1222         if (unpack_nondirectories(n, mask, dirmask, src, names, info) < 0)
1223                 return -1;
1224
1225         if (o->merge && src[0]) {
1226                 if (ce_stage(src[0]))
1227                         mark_ce_used_same_name(src[0], o);
1228                 else
1229                         mark_ce_used(src[0], o);
1230         }
1231
1232         /* Now handle any directories.. */
1233         if (dirmask) {
1234                 /* special case: "diff-index --cached" looking at a tree */
1235                 if (o->diff_index_cached &&
1236                     n == 1 && dirmask == 1 && S_ISDIR(names->mode)) {
1237                         int matches;
1238                         matches = cache_tree_matches_traversal(o->src_index->cache_tree,
1239                                                                names, info);
1240                         /*
1241                          * Everything under the name matches; skip the
1242                          * entire hierarchy.  diff_index_cached codepath
1243                          * special cases D/F conflicts in such a way that
1244                          * it does not do any look-ahead, so this is safe.
1245                          */
1246                         if (matches) {
1247                                 o->cache_bottom += matches;
1248                                 return mask;
1249                         }
1250                 }
1251
1252                 if (traverse_trees_recursive(n, dirmask, mask & ~dirmask,
1253                                              names, info) < 0)
1254                         return -1;
1255                 return mask;
1256         }
1257
1258         return mask;
1259 }
1260
1261 static int clear_ce_flags_1(struct index_state *istate,
1262                             struct cache_entry **cache, int nr,
1263                             struct strbuf *prefix,
1264                             int select_mask, int clear_mask,
1265                             struct exclude_list *el, int defval);
1266
1267 /* Whole directory matching */
1268 static int clear_ce_flags_dir(struct index_state *istate,
1269                               struct cache_entry **cache, int nr,
1270                               struct strbuf *prefix,
1271                               char *basename,
1272                               int select_mask, int clear_mask,
1273                               struct exclude_list *el, int defval)
1274 {
1275         struct cache_entry **cache_end;
1276         int dtype = DT_DIR;
1277         int ret = is_excluded_from_list(prefix->buf, prefix->len,
1278                                         basename, &dtype, el, istate);
1279         int rc;
1280
1281         strbuf_addch(prefix, '/');
1282
1283         /* If undecided, use matching result of parent dir in defval */
1284         if (ret < 0)
1285                 ret = defval;
1286
1287         for (cache_end = cache; cache_end != cache + nr; cache_end++) {
1288                 struct cache_entry *ce = *cache_end;
1289                 if (strncmp(ce->name, prefix->buf, prefix->len))
1290                         break;
1291         }
1292
1293         /*
1294          * TODO: check el, if there are no patterns that may conflict
1295          * with ret (iow, we know in advance the incl/excl
1296          * decision for the entire directory), clear flag here without
1297          * calling clear_ce_flags_1(). That function will call
1298          * the expensive is_excluded_from_list() on every entry.
1299          */
1300         rc = clear_ce_flags_1(istate, cache, cache_end - cache,
1301                               prefix,
1302                               select_mask, clear_mask,
1303                               el, ret);
1304         strbuf_setlen(prefix, prefix->len - 1);
1305         return rc;
1306 }
1307
1308 /*
1309  * Traverse the index, find every entry that matches according to
1310  * o->el. Do "ce_flags &= ~clear_mask" on those entries. Return the
1311  * number of traversed entries.
1312  *
1313  * If select_mask is non-zero, only entries whose ce_flags has on of
1314  * those bits enabled are traversed.
1315  *
1316  * cache        : pointer to an index entry
1317  * prefix_len   : an offset to its path
1318  *
1319  * The current path ("prefix") including the trailing '/' is
1320  *   cache[0]->name[0..(prefix_len-1)]
1321  * Top level path has prefix_len zero.
1322  */
1323 static int clear_ce_flags_1(struct index_state *istate,
1324                             struct cache_entry **cache, int nr,
1325                             struct strbuf *prefix,
1326                             int select_mask, int clear_mask,
1327                             struct exclude_list *el, int defval)
1328 {
1329         struct cache_entry **cache_end = cache + nr;
1330
1331         /*
1332          * Process all entries that have the given prefix and meet
1333          * select_mask condition
1334          */
1335         while(cache != cache_end) {
1336                 struct cache_entry *ce = *cache;
1337                 const char *name, *slash;
1338                 int len, dtype, ret;
1339
1340                 if (select_mask && !(ce->ce_flags & select_mask)) {
1341                         cache++;
1342                         continue;
1343                 }
1344
1345                 if (prefix->len && strncmp(ce->name, prefix->buf, prefix->len))
1346                         break;
1347
1348                 name = ce->name + prefix->len;
1349                 slash = strchr(name, '/');
1350
1351                 /* If it's a directory, try whole directory match first */
1352                 if (slash) {
1353                         int processed;
1354
1355                         len = slash - name;
1356                         strbuf_add(prefix, name, len);
1357
1358                         processed = clear_ce_flags_dir(istate, cache, cache_end - cache,
1359                                                        prefix,
1360                                                        prefix->buf + prefix->len - len,
1361                                                        select_mask, clear_mask,
1362                                                        el, defval);
1363
1364                         /* clear_c_f_dir eats a whole dir already? */
1365                         if (processed) {
1366                                 cache += processed;
1367                                 strbuf_setlen(prefix, prefix->len - len);
1368                                 continue;
1369                         }
1370
1371                         strbuf_addch(prefix, '/');
1372                         cache += clear_ce_flags_1(istate, cache, cache_end - cache,
1373                                                   prefix,
1374                                                   select_mask, clear_mask, el, defval);
1375                         strbuf_setlen(prefix, prefix->len - len - 1);
1376                         continue;
1377                 }
1378
1379                 /* Non-directory */
1380                 dtype = ce_to_dtype(ce);
1381                 ret = is_excluded_from_list(ce->name, ce_namelen(ce),
1382                                             name, &dtype, el, istate);
1383                 if (ret < 0)
1384                         ret = defval;
1385                 if (ret > 0)
1386                         ce->ce_flags &= ~clear_mask;
1387                 cache++;
1388         }
1389         return nr - (cache_end - cache);
1390 }
1391
1392 static int clear_ce_flags(struct index_state *istate,
1393                           int select_mask, int clear_mask,
1394                           struct exclude_list *el)
1395 {
1396         static struct strbuf prefix = STRBUF_INIT;
1397
1398         strbuf_reset(&prefix);
1399
1400         return clear_ce_flags_1(istate,
1401                                 istate->cache,
1402                                 istate->cache_nr,
1403                                 &prefix,
1404                                 select_mask, clear_mask,
1405                                 el, 0);
1406 }
1407
1408 /*
1409  * Set/Clear CE_NEW_SKIP_WORKTREE according to $GIT_DIR/info/sparse-checkout
1410  */
1411 static void mark_new_skip_worktree(struct exclude_list *el,
1412                                    struct index_state *istate,
1413                                    int select_flag, int skip_wt_flag)
1414 {
1415         int i;
1416
1417         /*
1418          * 1. Pretend the narrowest worktree: only unmerged entries
1419          * are checked out
1420          */
1421         for (i = 0; i < istate->cache_nr; i++) {
1422                 struct cache_entry *ce = istate->cache[i];
1423
1424                 if (select_flag && !(ce->ce_flags & select_flag))
1425                         continue;
1426
1427                 if (!ce_stage(ce) && !(ce->ce_flags & CE_CONFLICTED))
1428                         ce->ce_flags |= skip_wt_flag;
1429                 else
1430                         ce->ce_flags &= ~skip_wt_flag;
1431         }
1432
1433         /*
1434          * 2. Widen worktree according to sparse-checkout file.
1435          * Matched entries will have skip_wt_flag cleared (i.e. "in")
1436          */
1437         clear_ce_flags(istate, select_flag, skip_wt_flag, el);
1438 }
1439
1440 static int verify_absent(const struct cache_entry *,
1441                          enum unpack_trees_error_types,
1442                          struct unpack_trees_options *);
1443 /*
1444  * N-way merge "len" trees.  Returns 0 on success, -1 on failure to manipulate the
1445  * resulting index, -2 on failure to reflect the changes to the work tree.
1446  *
1447  * CE_ADDED, CE_UNPACKED and CE_NEW_SKIP_WORKTREE are used internally
1448  */
1449 int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options *o)
1450 {
1451         int i, ret;
1452         static struct cache_entry *dfc;
1453         struct exclude_list el;
1454
1455         if (len > MAX_UNPACK_TREES)
1456                 die("unpack_trees takes at most %d trees", MAX_UNPACK_TREES);
1457
1458         trace_performance_enter();
1459         memset(&el, 0, sizeof(el));
1460         if (!core_apply_sparse_checkout || !o->update)
1461                 o->skip_sparse_checkout = 1;
1462         if (!o->skip_sparse_checkout) {
1463                 char *sparse = git_pathdup("info/sparse-checkout");
1464                 if (add_excludes_from_file_to_list(sparse, "", 0, &el, NULL) < 0)
1465                         o->skip_sparse_checkout = 1;
1466                 else
1467                         o->el = &el;
1468                 free(sparse);
1469         }
1470
1471         memset(&o->result, 0, sizeof(o->result));
1472         o->result.initialized = 1;
1473         o->result.timestamp.sec = o->src_index->timestamp.sec;
1474         o->result.timestamp.nsec = o->src_index->timestamp.nsec;
1475         o->result.version = o->src_index->version;
1476         if (!o->src_index->split_index) {
1477                 o->result.split_index = NULL;
1478         } else if (o->src_index == o->dst_index) {
1479                 /*
1480                  * o->dst_index (and thus o->src_index) will be discarded
1481                  * and overwritten with o->result at the end of this function,
1482                  * so just use src_index's split_index to avoid having to
1483                  * create a new one.
1484                  */
1485                 o->result.split_index = o->src_index->split_index;
1486                 o->result.split_index->refcount++;
1487         } else {
1488                 o->result.split_index = init_split_index(&o->result);
1489         }
1490         oidcpy(&o->result.oid, &o->src_index->oid);
1491         o->merge_size = len;
1492         mark_all_ce_unused(o->src_index);
1493
1494         /*
1495          * Sparse checkout loop #1: set NEW_SKIP_WORKTREE on existing entries
1496          */
1497         if (!o->skip_sparse_checkout)
1498                 mark_new_skip_worktree(o->el, o->src_index, 0, CE_NEW_SKIP_WORKTREE);
1499
1500         if (!dfc)
1501                 dfc = xcalloc(1, cache_entry_size(0));
1502         o->df_conflict_entry = dfc;
1503
1504         if (len) {
1505                 const char *prefix = o->prefix ? o->prefix : "";
1506                 struct traverse_info info;
1507
1508                 setup_traverse_info(&info, prefix);
1509                 info.fn = unpack_callback;
1510                 info.data = o;
1511                 info.show_all_errors = o->show_all_errors;
1512                 info.pathspec = o->pathspec;
1513
1514                 if (o->prefix) {
1515                         /*
1516                          * Unpack existing index entries that sort before the
1517                          * prefix the tree is spliced into.  Note that o->merge
1518                          * is always true in this case.
1519                          */
1520                         while (1) {
1521                                 struct cache_entry *ce = next_cache_entry(o);
1522                                 if (!ce)
1523                                         break;
1524                                 if (ce_in_traverse_path(ce, &info))
1525                                         break;
1526                                 if (unpack_index_entry(ce, o) < 0)
1527                                         goto return_failed;
1528                         }
1529                 }
1530
1531                 trace_performance_enter();
1532                 ret = traverse_trees(o->src_index, len, t, &info);
1533                 trace_performance_leave("traverse_trees");
1534                 if (ret < 0)
1535                         goto return_failed;
1536         }
1537
1538         /* Any left-over entries in the index? */
1539         if (o->merge) {
1540                 while (1) {
1541                         struct cache_entry *ce = next_cache_entry(o);
1542                         if (!ce)
1543                                 break;
1544                         if (unpack_index_entry(ce, o) < 0)
1545                                 goto return_failed;
1546                 }
1547         }
1548         mark_all_ce_unused(o->src_index);
1549
1550         if (o->trivial_merges_only && o->nontrivial_merge) {
1551                 ret = unpack_failed(o, "Merge requires file-level merging");
1552                 goto done;
1553         }
1554
1555         if (!o->skip_sparse_checkout) {
1556                 int empty_worktree = 1;
1557
1558                 /*
1559                  * Sparse checkout loop #2: set NEW_SKIP_WORKTREE on entries not in loop #1
1560                  * If the will have NEW_SKIP_WORKTREE, also set CE_SKIP_WORKTREE
1561                  * so apply_sparse_checkout() won't attempt to remove it from worktree
1562                  */
1563                 mark_new_skip_worktree(o->el, &o->result, CE_ADDED, CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
1564
1565                 ret = 0;
1566                 for (i = 0; i < o->result.cache_nr; i++) {
1567                         struct cache_entry *ce = o->result.cache[i];
1568
1569                         /*
1570                          * Entries marked with CE_ADDED in merged_entry() do not have
1571                          * verify_absent() check (the check is effectively disabled
1572                          * because CE_NEW_SKIP_WORKTREE is set unconditionally).
1573                          *
1574                          * Do the real check now because we have had
1575                          * correct CE_NEW_SKIP_WORKTREE
1576                          */
1577                         if (ce->ce_flags & CE_ADDED &&
1578                             verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
1579                                 if (!o->show_all_errors)
1580                                         goto return_failed;
1581                                 ret = -1;
1582                         }
1583
1584                         if (apply_sparse_checkout(&o->result, ce, o)) {
1585                                 if (!o->show_all_errors)
1586                                         goto return_failed;
1587                                 ret = -1;
1588                         }
1589                         if (!ce_skip_worktree(ce))
1590                                 empty_worktree = 0;
1591
1592                 }
1593                 if (ret < 0)
1594                         goto return_failed;
1595                 /*
1596                  * Sparse checkout is meant to narrow down checkout area
1597                  * but it does not make sense to narrow down to empty working
1598                  * tree. This is usually a mistake in sparse checkout rules.
1599                  * Do not allow users to do that.
1600                  */
1601                 if (o->result.cache_nr && empty_worktree) {
1602                         ret = unpack_failed(o, "Sparse checkout leaves no entry on working directory");
1603                         goto done;
1604                 }
1605         }
1606
1607         ret = check_updates(o) ? (-2) : 0;
1608         if (o->dst_index) {
1609                 move_index_extensions(&o->result, o->src_index);
1610                 if (!ret) {
1611                         if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
1612                                 cache_tree_verify(the_repository, &o->result);
1613                         if (!o->result.cache_tree)
1614                                 o->result.cache_tree = cache_tree();
1615                         if (!cache_tree_fully_valid(o->result.cache_tree))
1616                                 cache_tree_update(&o->result,
1617                                                   WRITE_TREE_SILENT |
1618                                                   WRITE_TREE_REPAIR);
1619                 }
1620
1621                 o->result.updated_workdir = 1;
1622                 discard_index(o->dst_index);
1623                 *o->dst_index = o->result;
1624         } else {
1625                 discard_index(&o->result);
1626         }
1627         o->src_index = NULL;
1628
1629 done:
1630         trace_performance_leave("unpack_trees");
1631         clear_exclude_list(&el);
1632         return ret;
1633
1634 return_failed:
1635         if (o->show_all_errors)
1636                 display_error_msgs(o);
1637         mark_all_ce_unused(o->src_index);
1638         ret = unpack_failed(o, NULL);
1639         if (o->exiting_early)
1640                 ret = 0;
1641         goto done;
1642 }
1643
1644 /* Here come the merge functions */
1645
1646 static int reject_merge(const struct cache_entry *ce,
1647                         struct unpack_trees_options *o)
1648 {
1649         return o->gently ? -1 :
1650                 add_rejected_path(o, ERROR_WOULD_OVERWRITE, ce->name);
1651 }
1652
1653 static int same(const struct cache_entry *a, const struct cache_entry *b)
1654 {
1655         if (!!a != !!b)
1656                 return 0;
1657         if (!a && !b)
1658                 return 1;
1659         if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
1660                 return 0;
1661         return a->ce_mode == b->ce_mode &&
1662                oideq(&a->oid, &b->oid);
1663 }
1664
1665
1666 /*
1667  * When a CE gets turned into an unmerged entry, we
1668  * want it to be up-to-date
1669  */
1670 static int verify_uptodate_1(const struct cache_entry *ce,
1671                              struct unpack_trees_options *o,
1672                              enum unpack_trees_error_types error_type)
1673 {
1674         struct stat st;
1675
1676         if (o->index_only)
1677                 return 0;
1678
1679         /*
1680          * CE_VALID and CE_SKIP_WORKTREE cheat, we better check again
1681          * if this entry is truly up-to-date because this file may be
1682          * overwritten.
1683          */
1684         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
1685                 ; /* keep checking */
1686         else if (o->reset || ce_uptodate(ce))
1687                 return 0;
1688
1689         if (!lstat(ce->name, &st)) {
1690                 int flags = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE;
1691                 unsigned changed = ie_match_stat(o->src_index, ce, &st, flags);
1692
1693                 if (submodule_from_ce(ce)) {
1694                         int r = check_submodule_move_head(ce,
1695                                 "HEAD", oid_to_hex(&ce->oid), o);
1696                         if (r)
1697                                 return o->gently ? -1 :
1698                                         add_rejected_path(o, error_type, ce->name);
1699                         return 0;
1700                 }
1701
1702                 if (!changed)
1703                         return 0;
1704                 /*
1705                  * Historic default policy was to allow submodule to be out
1706                  * of sync wrt the superproject index. If the submodule was
1707                  * not considered interesting above, we don't care here.
1708                  */
1709                 if (S_ISGITLINK(ce->ce_mode))
1710                         return 0;
1711
1712                 errno = 0;
1713         }
1714         if (errno == ENOENT)
1715                 return 0;
1716         return o->gently ? -1 :
1717                 add_rejected_path(o, error_type, ce->name);
1718 }
1719
1720 int verify_uptodate(const struct cache_entry *ce,
1721                     struct unpack_trees_options *o)
1722 {
1723         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1724                 return 0;
1725         return verify_uptodate_1(ce, o, ERROR_NOT_UPTODATE_FILE);
1726 }
1727
1728 static int verify_uptodate_sparse(const struct cache_entry *ce,
1729                                   struct unpack_trees_options *o)
1730 {
1731         return verify_uptodate_1(ce, o, ERROR_SPARSE_NOT_UPTODATE_FILE);
1732 }
1733
1734 /*
1735  * TODO: We should actually invalidate o->result, not src_index [1].
1736  * But since cache tree and untracked cache both are not copied to
1737  * o->result until unpacking is complete, we invalidate them on
1738  * src_index instead with the assumption that they will be copied to
1739  * dst_index at the end.
1740  *
1741  * [1] src_index->cache_tree is also used in unpack_callback() so if
1742  * we invalidate o->result, we need to update it to use
1743  * o->result.cache_tree as well.
1744  */
1745 static void invalidate_ce_path(const struct cache_entry *ce,
1746                                struct unpack_trees_options *o)
1747 {
1748         if (!ce)
1749                 return;
1750         cache_tree_invalidate_path(o->src_index, ce->name);
1751         untracked_cache_invalidate_path(o->src_index, ce->name, 1);
1752 }
1753
1754 /*
1755  * Check that checking out ce->sha1 in subdir ce->name is not
1756  * going to overwrite any working files.
1757  *
1758  * Currently, git does not checkout subprojects during a superproject
1759  * checkout, so it is not going to overwrite anything.
1760  */
1761 static int verify_clean_submodule(const char *old_sha1,
1762                                   const struct cache_entry *ce,
1763                                   struct unpack_trees_options *o)
1764 {
1765         if (!submodule_from_ce(ce))
1766                 return 0;
1767
1768         return check_submodule_move_head(ce, old_sha1,
1769                                          oid_to_hex(&ce->oid), o);
1770 }
1771
1772 static int verify_clean_subdirectory(const struct cache_entry *ce,
1773                                      struct unpack_trees_options *o)
1774 {
1775         /*
1776          * we are about to extract "ce->name"; we would not want to lose
1777          * anything in the existing directory there.
1778          */
1779         int namelen;
1780         int i;
1781         struct dir_struct d;
1782         char *pathbuf;
1783         int cnt = 0;
1784
1785         if (S_ISGITLINK(ce->ce_mode)) {
1786                 struct object_id oid;
1787                 int sub_head = resolve_gitlink_ref(ce->name, "HEAD", &oid);
1788                 /*
1789                  * If we are not going to update the submodule, then
1790                  * we don't care.
1791                  */
1792                 if (!sub_head && oideq(&oid, &ce->oid))
1793                         return 0;
1794                 return verify_clean_submodule(sub_head ? NULL : oid_to_hex(&oid),
1795                                               ce, o);
1796         }
1797
1798         /*
1799          * First let's make sure we do not have a local modification
1800          * in that directory.
1801          */
1802         namelen = ce_namelen(ce);
1803         for (i = locate_in_src_index(ce, o);
1804              i < o->src_index->cache_nr;
1805              i++) {
1806                 struct cache_entry *ce2 = o->src_index->cache[i];
1807                 int len = ce_namelen(ce2);
1808                 if (len < namelen ||
1809                     strncmp(ce->name, ce2->name, namelen) ||
1810                     ce2->name[namelen] != '/')
1811                         break;
1812                 /*
1813                  * ce2->name is an entry in the subdirectory to be
1814                  * removed.
1815                  */
1816                 if (!ce_stage(ce2)) {
1817                         if (verify_uptodate(ce2, o))
1818                                 return -1;
1819                         add_entry(o, ce2, CE_REMOVE, 0);
1820                         invalidate_ce_path(ce, o);
1821                         mark_ce_used(ce2, o);
1822                 }
1823                 cnt++;
1824         }
1825
1826         /*
1827          * Then we need to make sure that we do not lose a locally
1828          * present file that is not ignored.
1829          */
1830         pathbuf = xstrfmt("%.*s/", namelen, ce->name);
1831
1832         memset(&d, 0, sizeof(d));
1833         if (o->dir)
1834                 d.exclude_per_dir = o->dir->exclude_per_dir;
1835         i = read_directory(&d, o->src_index, pathbuf, namelen+1, NULL);
1836         if (i)
1837                 return o->gently ? -1 :
1838                         add_rejected_path(o, ERROR_NOT_UPTODATE_DIR, ce->name);
1839         free(pathbuf);
1840         return cnt;
1841 }
1842
1843 /*
1844  * This gets called when there was no index entry for the tree entry 'dst',
1845  * but we found a file in the working tree that 'lstat()' said was fine,
1846  * and we're on a case-insensitive filesystem.
1847  *
1848  * See if we can find a case-insensitive match in the index that also
1849  * matches the stat information, and assume it's that other file!
1850  */
1851 static int icase_exists(struct unpack_trees_options *o, const char *name, int len, struct stat *st)
1852 {
1853         const struct cache_entry *src;
1854
1855         src = index_file_exists(o->src_index, name, len, 1);
1856         return src && !ie_match_stat(o->src_index, src, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
1857 }
1858
1859 static int check_ok_to_remove(const char *name, int len, int dtype,
1860                               const struct cache_entry *ce, struct stat *st,
1861                               enum unpack_trees_error_types error_type,
1862                               struct unpack_trees_options *o)
1863 {
1864         const struct cache_entry *result;
1865
1866         /*
1867          * It may be that the 'lstat()' succeeded even though
1868          * target 'ce' was absent, because there is an old
1869          * entry that is different only in case..
1870          *
1871          * Ignore that lstat() if it matches.
1872          */
1873         if (ignore_case && icase_exists(o, name, len, st))
1874                 return 0;
1875
1876         if (o->dir &&
1877             is_excluded(o->dir, o->src_index, name, &dtype))
1878                 /*
1879                  * ce->name is explicitly excluded, so it is Ok to
1880                  * overwrite it.
1881                  */
1882                 return 0;
1883         if (S_ISDIR(st->st_mode)) {
1884                 /*
1885                  * We are checking out path "foo" and
1886                  * found "foo/." in the working tree.
1887                  * This is tricky -- if we have modified
1888                  * files that are in "foo/" we would lose
1889                  * them.
1890                  */
1891                 if (verify_clean_subdirectory(ce, o) < 0)
1892                         return -1;
1893                 return 0;
1894         }
1895
1896         /*
1897          * The previous round may already have decided to
1898          * delete this path, which is in a subdirectory that
1899          * is being replaced with a blob.
1900          */
1901         result = index_file_exists(&o->result, name, len, 0);
1902         if (result) {
1903                 if (result->ce_flags & CE_REMOVE)
1904                         return 0;
1905         }
1906
1907         return o->gently ? -1 :
1908                 add_rejected_path(o, error_type, name);
1909 }
1910
1911 /*
1912  * We do not want to remove or overwrite a working tree file that
1913  * is not tracked, unless it is ignored.
1914  */
1915 static int verify_absent_1(const struct cache_entry *ce,
1916                            enum unpack_trees_error_types error_type,
1917                            struct unpack_trees_options *o)
1918 {
1919         int len;
1920         struct stat st;
1921
1922         if (o->index_only || o->reset || !o->update)
1923                 return 0;
1924
1925         len = check_leading_path(ce->name, ce_namelen(ce));
1926         if (!len)
1927                 return 0;
1928         else if (len > 0) {
1929                 char *path;
1930                 int ret;
1931
1932                 path = xmemdupz(ce->name, len);
1933                 if (lstat(path, &st))
1934                         ret = error_errno("cannot stat '%s'", path);
1935                 else {
1936                         if (submodule_from_ce(ce))
1937                                 ret = check_submodule_move_head(ce,
1938                                                                 oid_to_hex(&ce->oid),
1939                                                                 NULL, o);
1940                         else
1941                                 ret = check_ok_to_remove(path, len, DT_UNKNOWN, NULL,
1942                                                          &st, error_type, o);
1943                 }
1944                 free(path);
1945                 return ret;
1946         } else if (lstat(ce->name, &st)) {
1947                 if (errno != ENOENT)
1948                         return error_errno("cannot stat '%s'", ce->name);
1949                 return 0;
1950         } else {
1951                 if (submodule_from_ce(ce))
1952                         return check_submodule_move_head(ce, oid_to_hex(&ce->oid),
1953                                                          NULL, o);
1954
1955                 return check_ok_to_remove(ce->name, ce_namelen(ce),
1956                                           ce_to_dtype(ce), ce, &st,
1957                                           error_type, o);
1958         }
1959 }
1960
1961 static int verify_absent(const struct cache_entry *ce,
1962                          enum unpack_trees_error_types error_type,
1963                          struct unpack_trees_options *o)
1964 {
1965         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1966                 return 0;
1967         return verify_absent_1(ce, error_type, o);
1968 }
1969
1970 static int verify_absent_sparse(const struct cache_entry *ce,
1971                                 enum unpack_trees_error_types error_type,
1972                                 struct unpack_trees_options *o)
1973 {
1974         enum unpack_trees_error_types orphaned_error = error_type;
1975         if (orphaned_error == ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN)
1976                 orphaned_error = ERROR_WOULD_LOSE_ORPHANED_OVERWRITTEN;
1977
1978         return verify_absent_1(ce, orphaned_error, o);
1979 }
1980
1981 static int merged_entry(const struct cache_entry *ce,
1982                         const struct cache_entry *old,
1983                         struct unpack_trees_options *o)
1984 {
1985         int update = CE_UPDATE;
1986         struct cache_entry *merge = dup_cache_entry(ce, &o->result);
1987
1988         if (!old) {
1989                 /*
1990                  * New index entries. In sparse checkout, the following
1991                  * verify_absent() will be delayed until after
1992                  * traverse_trees() finishes in unpack_trees(), then:
1993                  *
1994                  *  - CE_NEW_SKIP_WORKTREE will be computed correctly
1995                  *  - verify_absent() be called again, this time with
1996                  *    correct CE_NEW_SKIP_WORKTREE
1997                  *
1998                  * verify_absent() call here does nothing in sparse
1999                  * checkout (i.e. o->skip_sparse_checkout == 0)
2000                  */
2001                 update |= CE_ADDED;
2002                 merge->ce_flags |= CE_NEW_SKIP_WORKTREE;
2003
2004                 if (verify_absent(merge,
2005                                   ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2006                         discard_cache_entry(merge);
2007                         return -1;
2008                 }
2009                 invalidate_ce_path(merge, o);
2010
2011                 if (submodule_from_ce(ce)) {
2012                         int ret = check_submodule_move_head(ce, NULL,
2013                                                             oid_to_hex(&ce->oid),
2014                                                             o);
2015                         if (ret)
2016                                 return ret;
2017                 }
2018
2019         } else if (!(old->ce_flags & CE_CONFLICTED)) {
2020                 /*
2021                  * See if we can re-use the old CE directly?
2022                  * That way we get the uptodate stat info.
2023                  *
2024                  * This also removes the UPDATE flag on a match; otherwise
2025                  * we will end up overwriting local changes in the work tree.
2026                  */
2027                 if (same(old, merge)) {
2028                         copy_cache_entry(merge, old);
2029                         update = 0;
2030                 } else {
2031                         if (verify_uptodate(old, o)) {
2032                                 discard_cache_entry(merge);
2033                                 return -1;
2034                         }
2035                         /* Migrate old flags over */
2036                         update |= old->ce_flags & (CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
2037                         invalidate_ce_path(old, o);
2038                 }
2039
2040                 if (submodule_from_ce(ce)) {
2041                         int ret = check_submodule_move_head(ce, oid_to_hex(&old->oid),
2042                                                             oid_to_hex(&ce->oid),
2043                                                             o);
2044                         if (ret)
2045                                 return ret;
2046                 }
2047         } else {
2048                 /*
2049                  * Previously unmerged entry left as an existence
2050                  * marker by read_index_unmerged();
2051                  */
2052                 invalidate_ce_path(old, o);
2053         }
2054
2055         do_add_entry(o, merge, update, CE_STAGEMASK);
2056         return 1;
2057 }
2058
2059 static int deleted_entry(const struct cache_entry *ce,
2060                          const struct cache_entry *old,
2061                          struct unpack_trees_options *o)
2062 {
2063         /* Did it exist in the index? */
2064         if (!old) {
2065                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2066                         return -1;
2067                 return 0;
2068         }
2069         if (!(old->ce_flags & CE_CONFLICTED) && verify_uptodate(old, o))
2070                 return -1;
2071         add_entry(o, ce, CE_REMOVE, 0);
2072         invalidate_ce_path(ce, o);
2073         return 1;
2074 }
2075
2076 static int keep_entry(const struct cache_entry *ce,
2077                       struct unpack_trees_options *o)
2078 {
2079         add_entry(o, ce, 0, 0);
2080         if (ce_stage(ce))
2081                 invalidate_ce_path(ce, o);
2082         return 1;
2083 }
2084
2085 #if DBRT_DEBUG
2086 static void show_stage_entry(FILE *o,
2087                              const char *label, const struct cache_entry *ce)
2088 {
2089         if (!ce)
2090                 fprintf(o, "%s (missing)\n", label);
2091         else
2092                 fprintf(o, "%s%06o %s %d\t%s\n",
2093                         label,
2094                         ce->ce_mode,
2095                         oid_to_hex(&ce->oid),
2096                         ce_stage(ce),
2097                         ce->name);
2098 }
2099 #endif
2100
2101 int threeway_merge(const struct cache_entry * const *stages,
2102                    struct unpack_trees_options *o)
2103 {
2104         const struct cache_entry *index;
2105         const struct cache_entry *head;
2106         const struct cache_entry *remote = stages[o->head_idx + 1];
2107         int count;
2108         int head_match = 0;
2109         int remote_match = 0;
2110
2111         int df_conflict_head = 0;
2112         int df_conflict_remote = 0;
2113
2114         int any_anc_missing = 0;
2115         int no_anc_exists = 1;
2116         int i;
2117
2118         for (i = 1; i < o->head_idx; i++) {
2119                 if (!stages[i] || stages[i] == o->df_conflict_entry)
2120                         any_anc_missing = 1;
2121                 else
2122                         no_anc_exists = 0;
2123         }
2124
2125         index = stages[0];
2126         head = stages[o->head_idx];
2127
2128         if (head == o->df_conflict_entry) {
2129                 df_conflict_head = 1;
2130                 head = NULL;
2131         }
2132
2133         if (remote == o->df_conflict_entry) {
2134                 df_conflict_remote = 1;
2135                 remote = NULL;
2136         }
2137
2138         /*
2139          * First, if there's a #16 situation, note that to prevent #13
2140          * and #14.
2141          */
2142         if (!same(remote, head)) {
2143                 for (i = 1; i < o->head_idx; i++) {
2144                         if (same(stages[i], head)) {
2145                                 head_match = i;
2146                         }
2147                         if (same(stages[i], remote)) {
2148                                 remote_match = i;
2149                         }
2150                 }
2151         }
2152
2153         /*
2154          * We start with cases where the index is allowed to match
2155          * something other than the head: #14(ALT) and #2ALT, where it
2156          * is permitted to match the result instead.
2157          */
2158         /* #14, #14ALT, #2ALT */
2159         if (remote && !df_conflict_head && head_match && !remote_match) {
2160                 if (index && !same(index, remote) && !same(index, head))
2161                         return reject_merge(index, o);
2162                 return merged_entry(remote, index, o);
2163         }
2164         /*
2165          * If we have an entry in the index cache, then we want to
2166          * make sure that it matches head.
2167          */
2168         if (index && !same(index, head))
2169                 return reject_merge(index, o);
2170
2171         if (head) {
2172                 /* #5ALT, #15 */
2173                 if (same(head, remote))
2174                         return merged_entry(head, index, o);
2175                 /* #13, #3ALT */
2176                 if (!df_conflict_remote && remote_match && !head_match)
2177                         return merged_entry(head, index, o);
2178         }
2179
2180         /* #1 */
2181         if (!head && !remote && any_anc_missing)
2182                 return 0;
2183
2184         /*
2185          * Under the "aggressive" rule, we resolve mostly trivial
2186          * cases that we historically had git-merge-one-file resolve.
2187          */
2188         if (o->aggressive) {
2189                 int head_deleted = !head;
2190                 int remote_deleted = !remote;
2191                 const struct cache_entry *ce = NULL;
2192
2193                 if (index)
2194                         ce = index;
2195                 else if (head)
2196                         ce = head;
2197                 else if (remote)
2198                         ce = remote;
2199                 else {
2200                         for (i = 1; i < o->head_idx; i++) {
2201                                 if (stages[i] && stages[i] != o->df_conflict_entry) {
2202                                         ce = stages[i];
2203                                         break;
2204                                 }
2205                         }
2206                 }
2207
2208                 /*
2209                  * Deleted in both.
2210                  * Deleted in one and unchanged in the other.
2211                  */
2212                 if ((head_deleted && remote_deleted) ||
2213                     (head_deleted && remote && remote_match) ||
2214                     (remote_deleted && head && head_match)) {
2215                         if (index)
2216                                 return deleted_entry(index, index, o);
2217                         if (ce && !head_deleted) {
2218                                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2219                                         return -1;
2220                         }
2221                         return 0;
2222                 }
2223                 /*
2224                  * Added in both, identically.
2225                  */
2226                 if (no_anc_exists && head && remote && same(head, remote))
2227                         return merged_entry(head, index, o);
2228
2229         }
2230
2231         /* Below are "no merge" cases, which require that the index be
2232          * up-to-date to avoid the files getting overwritten with
2233          * conflict resolution files.
2234          */
2235         if (index) {
2236                 if (verify_uptodate(index, o))
2237                         return -1;
2238         }
2239
2240         o->nontrivial_merge = 1;
2241
2242         /* #2, #3, #4, #6, #7, #9, #10, #11. */
2243         count = 0;
2244         if (!head_match || !remote_match) {
2245                 for (i = 1; i < o->head_idx; i++) {
2246                         if (stages[i] && stages[i] != o->df_conflict_entry) {
2247                                 keep_entry(stages[i], o);
2248                                 count++;
2249                                 break;
2250                         }
2251                 }
2252         }
2253 #if DBRT_DEBUG
2254         else {
2255                 fprintf(stderr, "read-tree: warning #16 detected\n");
2256                 show_stage_entry(stderr, "head   ", stages[head_match]);
2257                 show_stage_entry(stderr, "remote ", stages[remote_match]);
2258         }
2259 #endif
2260         if (head) { count += keep_entry(head, o); }
2261         if (remote) { count += keep_entry(remote, o); }
2262         return count;
2263 }
2264
2265 /*
2266  * Two-way merge.
2267  *
2268  * The rule is to "carry forward" what is in the index without losing
2269  * information across a "fast-forward", favoring a successful merge
2270  * over a merge failure when it makes sense.  For details of the
2271  * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
2272  *
2273  */
2274 int twoway_merge(const struct cache_entry * const *src,
2275                  struct unpack_trees_options *o)
2276 {
2277         const struct cache_entry *current = src[0];
2278         const struct cache_entry *oldtree = src[1];
2279         const struct cache_entry *newtree = src[2];
2280
2281         if (o->merge_size != 2)
2282                 return error("Cannot do a twoway merge of %d trees",
2283                              o->merge_size);
2284
2285         if (oldtree == o->df_conflict_entry)
2286                 oldtree = NULL;
2287         if (newtree == o->df_conflict_entry)
2288                 newtree = NULL;
2289
2290         if (current) {
2291                 if (current->ce_flags & CE_CONFLICTED) {
2292                         if (same(oldtree, newtree) || o->reset) {
2293                                 if (!newtree)
2294                                         return deleted_entry(current, current, o);
2295                                 else
2296                                         return merged_entry(newtree, current, o);
2297                         }
2298                         return reject_merge(current, o);
2299                 } else if ((!oldtree && !newtree) || /* 4 and 5 */
2300                          (!oldtree && newtree &&
2301                           same(current, newtree)) || /* 6 and 7 */
2302                          (oldtree && newtree &&
2303                           same(oldtree, newtree)) || /* 14 and 15 */
2304                          (oldtree && newtree &&
2305                           !same(oldtree, newtree) && /* 18 and 19 */
2306                           same(current, newtree))) {
2307                         return keep_entry(current, o);
2308                 } else if (oldtree && !newtree && same(current, oldtree)) {
2309                         /* 10 or 11 */
2310                         return deleted_entry(oldtree, current, o);
2311                 } else if (oldtree && newtree &&
2312                          same(current, oldtree) && !same(current, newtree)) {
2313                         /* 20 or 21 */
2314                         return merged_entry(newtree, current, o);
2315                 } else
2316                         return reject_merge(current, o);
2317         }
2318         else if (newtree) {
2319                 if (oldtree && !o->initial_checkout) {
2320                         /*
2321                          * deletion of the path was staged;
2322                          */
2323                         if (same(oldtree, newtree))
2324                                 return 1;
2325                         return reject_merge(oldtree, o);
2326                 }
2327                 return merged_entry(newtree, current, o);
2328         }
2329         return deleted_entry(oldtree, current, o);
2330 }
2331
2332 /*
2333  * Bind merge.
2334  *
2335  * Keep the index entries at stage0, collapse stage1 but make sure
2336  * stage0 does not have anything there.
2337  */
2338 int bind_merge(const struct cache_entry * const *src,
2339                struct unpack_trees_options *o)
2340 {
2341         const struct cache_entry *old = src[0];
2342         const struct cache_entry *a = src[1];
2343
2344         if (o->merge_size != 1)
2345                 return error("Cannot do a bind merge of %d trees",
2346                              o->merge_size);
2347         if (a && old)
2348                 return o->gently ? -1 :
2349                         error(ERRORMSG(o, ERROR_BIND_OVERLAP),
2350                               super_prefixed(a->name),
2351                               super_prefixed(old->name));
2352         if (!a)
2353                 return keep_entry(old, o);
2354         else
2355                 return merged_entry(a, NULL, o);
2356 }
2357
2358 /*
2359  * One-way merge.
2360  *
2361  * The rule is:
2362  * - take the stat information from stage0, take the data from stage1
2363  */
2364 int oneway_merge(const struct cache_entry * const *src,
2365                  struct unpack_trees_options *o)
2366 {
2367         const struct cache_entry *old = src[0];
2368         const struct cache_entry *a = src[1];
2369
2370         if (o->merge_size != 1)
2371                 return error("Cannot do a oneway merge of %d trees",
2372                              o->merge_size);
2373
2374         if (!a || a == o->df_conflict_entry)
2375                 return deleted_entry(old, old, o);
2376
2377         if (old && same(old, a)) {
2378                 int update = 0;
2379                 if (o->reset && o->update && !ce_uptodate(old) && !ce_skip_worktree(old)) {
2380                         struct stat st;
2381                         if (lstat(old->name, &st) ||
2382                             ie_match_stat(o->src_index, old, &st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE))
2383                                 update |= CE_UPDATE;
2384                 }
2385                 if (o->update && S_ISGITLINK(old->ce_mode) &&
2386                     should_update_submodules() && !verify_uptodate(old, o))
2387                         update |= CE_UPDATE;
2388                 add_entry(o, old, update, CE_STAGEMASK);
2389                 return 0;
2390         }
2391         return merged_entry(a, old, o);
2392 }