Merge branch 'js/http-pki-credential-store'
[git] / unpack-trees.c
1 #include "cache.h"
2 #include "strvec.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 "submodule.h"
15 #include "submodule-config.h"
16 #include "fsmonitor.h"
17 #include "object-store.h"
18 #include "promisor-remote.h"
19
20 /*
21  * Error messages expected by scripts out of plumbing commands such as
22  * read-tree.  Non-scripted Porcelain is not required to use these messages
23  * and in fact are encouraged to reword them to better suit their particular
24  * situation better.  See how "git checkout" and "git merge" replaces
25  * them using setup_unpack_trees_porcelain(), for example.
26  */
27 static const char *unpack_plumbing_errors[NB_UNPACK_TREES_WARNING_TYPES] = {
28         /* ERROR_WOULD_OVERWRITE */
29         "Entry '%s' would be overwritten by merge. Cannot merge.",
30
31         /* ERROR_NOT_UPTODATE_FILE */
32         "Entry '%s' not uptodate. Cannot merge.",
33
34         /* ERROR_NOT_UPTODATE_DIR */
35         "Updating '%s' would lose untracked files in it",
36
37         /* ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN */
38         "Untracked working tree file '%s' would be overwritten by merge.",
39
40         /* ERROR_WOULD_LOSE_UNTRACKED_REMOVED */
41         "Untracked working tree file '%s' would be removed by merge.",
42
43         /* ERROR_BIND_OVERLAP */
44         "Entry '%s' overlaps with '%s'.  Cannot bind.",
45
46         /* ERROR_WOULD_LOSE_SUBMODULE */
47         "Submodule '%s' cannot checkout new HEAD.",
48
49         /* NB_UNPACK_TREES_ERROR_TYPES; just a meta value */
50         "",
51
52         /* WARNING_SPARSE_NOT_UPTODATE_FILE */
53         "Path '%s' not uptodate; will not remove from working tree.",
54
55         /* WARNING_SPARSE_UNMERGED_FILE */
56         "Path '%s' unmerged; will not remove from working tree.",
57
58         /* WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN */
59         "Path '%s' already present; will not overwrite with sparse update.",
60 };
61
62 #define ERRORMSG(o,type) \
63         ( ((o) && (o)->msgs[(type)]) \
64           ? ((o)->msgs[(type)])      \
65           : (unpack_plumbing_errors[(type)]) )
66
67 static const char *super_prefixed(const char *path)
68 {
69         /*
70          * It is necessary and sufficient to have two static buffers
71          * here, as the return value of this function is fed to
72          * error() using the unpack_*_errors[] templates we see above.
73          */
74         static struct strbuf buf[2] = {STRBUF_INIT, STRBUF_INIT};
75         static int super_prefix_len = -1;
76         static unsigned idx = ARRAY_SIZE(buf) - 1;
77
78         if (super_prefix_len < 0) {
79                 const char *super_prefix = get_super_prefix();
80                 if (!super_prefix) {
81                         super_prefix_len = 0;
82                 } else {
83                         int i;
84                         for (i = 0; i < ARRAY_SIZE(buf); i++)
85                                 strbuf_addstr(&buf[i], super_prefix);
86                         super_prefix_len = buf[0].len;
87                 }
88         }
89
90         if (!super_prefix_len)
91                 return path;
92
93         if (++idx >= ARRAY_SIZE(buf))
94                 idx = 0;
95
96         strbuf_setlen(&buf[idx], super_prefix_len);
97         strbuf_addstr(&buf[idx], path);
98
99         return buf[idx].buf;
100 }
101
102 void setup_unpack_trees_porcelain(struct unpack_trees_options *opts,
103                                   const char *cmd)
104 {
105         int i;
106         const char **msgs = opts->msgs;
107         const char *msg;
108
109         strvec_init(&opts->msgs_to_free);
110
111         if (!strcmp(cmd, "checkout"))
112                 msg = advice_commit_before_merge
113                       ? _("Your local changes to the following files would be overwritten by checkout:\n%%s"
114                           "Please commit your changes or stash them before you switch branches.")
115                       : _("Your local changes to the following files would be overwritten by checkout:\n%%s");
116         else if (!strcmp(cmd, "merge"))
117                 msg = advice_commit_before_merge
118                       ? _("Your local changes to the following files would be overwritten by merge:\n%%s"
119                           "Please commit your changes or stash them before you merge.")
120                       : _("Your local changes to the following files would be overwritten by merge:\n%%s");
121         else
122                 msg = advice_commit_before_merge
123                       ? _("Your local changes to the following files would be overwritten by %s:\n%%s"
124                           "Please commit your changes or stash them before you %s.")
125                       : _("Your local changes to the following files would be overwritten by %s:\n%%s");
126         msgs[ERROR_WOULD_OVERWRITE] = msgs[ERROR_NOT_UPTODATE_FILE] =
127                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
128
129         msgs[ERROR_NOT_UPTODATE_DIR] =
130                 _("Updating the following directories would lose untracked files in them:\n%s");
131
132         if (!strcmp(cmd, "checkout"))
133                 msg = advice_commit_before_merge
134                       ? _("The following untracked working tree files would be removed by checkout:\n%%s"
135                           "Please move or remove them before you switch branches.")
136                       : _("The following untracked working tree files would be removed by checkout:\n%%s");
137         else if (!strcmp(cmd, "merge"))
138                 msg = advice_commit_before_merge
139                       ? _("The following untracked working tree files would be removed by merge:\n%%s"
140                           "Please move or remove them before you merge.")
141                       : _("The following untracked working tree files would be removed by merge:\n%%s");
142         else
143                 msg = advice_commit_before_merge
144                       ? _("The following untracked working tree files would be removed by %s:\n%%s"
145                           "Please move or remove them before you %s.")
146                       : _("The following untracked working tree files would be removed by %s:\n%%s");
147         msgs[ERROR_WOULD_LOSE_UNTRACKED_REMOVED] =
148                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
149
150         if (!strcmp(cmd, "checkout"))
151                 msg = advice_commit_before_merge
152                       ? _("The following untracked working tree files would be overwritten by checkout:\n%%s"
153                           "Please move or remove them before you switch branches.")
154                       : _("The following untracked working tree files would be overwritten by checkout:\n%%s");
155         else if (!strcmp(cmd, "merge"))
156                 msg = advice_commit_before_merge
157                       ? _("The following untracked working tree files would be overwritten by merge:\n%%s"
158                           "Please move or remove them before you merge.")
159                       : _("The following untracked working tree files would be overwritten by merge:\n%%s");
160         else
161                 msg = advice_commit_before_merge
162                       ? _("The following untracked working tree files would be overwritten by %s:\n%%s"
163                           "Please move or remove them before you %s.")
164                       : _("The following untracked working tree files would be overwritten by %s:\n%%s");
165         msgs[ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN] =
166                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
167
168         /*
169          * Special case: ERROR_BIND_OVERLAP refers to a pair of paths, we
170          * cannot easily display it as a list.
171          */
172         msgs[ERROR_BIND_OVERLAP] = _("Entry '%s' overlaps with '%s'.  Cannot bind.");
173
174         msgs[ERROR_WOULD_LOSE_SUBMODULE] =
175                 _("Cannot update submodule:\n%s");
176
177         msgs[WARNING_SPARSE_NOT_UPTODATE_FILE] =
178                 _("The following paths are not up to date and were left despite sparse patterns:\n%s");
179         msgs[WARNING_SPARSE_UNMERGED_FILE] =
180                 _("The following paths are unmerged and were left despite sparse patterns:\n%s");
181         msgs[WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN] =
182                 _("The following paths were already present and thus not updated despite sparse patterns:\n%s");
183
184         opts->show_all_errors = 1;
185         /* rejected paths may not have a static buffer */
186         for (i = 0; i < ARRAY_SIZE(opts->unpack_rejects); i++)
187                 opts->unpack_rejects[i].strdup_strings = 1;
188 }
189
190 void clear_unpack_trees_porcelain(struct unpack_trees_options *opts)
191 {
192         strvec_clear(&opts->msgs_to_free);
193         memset(opts->msgs, 0, sizeof(opts->msgs));
194 }
195
196 static int do_add_entry(struct unpack_trees_options *o, struct cache_entry *ce,
197                          unsigned int set, unsigned int clear)
198 {
199         clear |= CE_HASHED;
200
201         if (set & CE_REMOVE)
202                 set |= CE_WT_REMOVE;
203
204         ce->ce_flags = (ce->ce_flags & ~clear) | set;
205         return add_index_entry(&o->result, ce,
206                                ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
207 }
208
209 static void add_entry(struct unpack_trees_options *o,
210                       const struct cache_entry *ce,
211                       unsigned int set, unsigned int clear)
212 {
213         do_add_entry(o, dup_cache_entry(ce, &o->result), set, clear);
214 }
215
216 /*
217  * add error messages on path <path>
218  * corresponding to the type <e> with the message <msg>
219  * indicating if it should be display in porcelain or not
220  */
221 static int add_rejected_path(struct unpack_trees_options *o,
222                              enum unpack_trees_error_types e,
223                              const char *path)
224 {
225         if (o->quiet)
226                 return -1;
227
228         if (!o->show_all_errors)
229                 return error(ERRORMSG(o, e), super_prefixed(path));
230
231         /*
232          * Otherwise, insert in a list for future display by
233          * display_(error|warning)_msgs()
234          */
235         string_list_append(&o->unpack_rejects[e], path);
236         return -1;
237 }
238
239 /*
240  * display all the error messages stored in a nice way
241  */
242 static void display_error_msgs(struct unpack_trees_options *o)
243 {
244         int e;
245         unsigned error_displayed = 0;
246         for (e = 0; e < NB_UNPACK_TREES_ERROR_TYPES; e++) {
247                 struct string_list *rejects = &o->unpack_rejects[e];
248
249                 if (rejects->nr > 0) {
250                         int i;
251                         struct strbuf path = STRBUF_INIT;
252
253                         error_displayed = 1;
254                         for (i = 0; i < rejects->nr; i++)
255                                 strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
256                         error(ERRORMSG(o, e), super_prefixed(path.buf));
257                         strbuf_release(&path);
258                 }
259                 string_list_clear(rejects, 0);
260         }
261         if (error_displayed)
262                 fprintf(stderr, _("Aborting\n"));
263 }
264
265 /*
266  * display all the warning messages stored in a nice way
267  */
268 static void display_warning_msgs(struct unpack_trees_options *o)
269 {
270         int e;
271         unsigned warning_displayed = 0;
272         for (e = NB_UNPACK_TREES_ERROR_TYPES + 1;
273              e < NB_UNPACK_TREES_WARNING_TYPES; e++) {
274                 struct string_list *rejects = &o->unpack_rejects[e];
275
276                 if (rejects->nr > 0) {
277                         int i;
278                         struct strbuf path = STRBUF_INIT;
279
280                         warning_displayed = 1;
281                         for (i = 0; i < rejects->nr; i++)
282                                 strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
283                         warning(ERRORMSG(o, e), super_prefixed(path.buf));
284                         strbuf_release(&path);
285                 }
286                 string_list_clear(rejects, 0);
287         }
288         if (warning_displayed)
289                 fprintf(stderr, _("After fixing the above paths, you may want to run `git sparse-checkout reapply`.\n"));
290 }
291 static int check_submodule_move_head(const struct cache_entry *ce,
292                                      const char *old_id,
293                                      const char *new_id,
294                                      struct unpack_trees_options *o)
295 {
296         unsigned flags = SUBMODULE_MOVE_HEAD_DRY_RUN;
297         const struct submodule *sub = submodule_from_ce(ce);
298
299         if (!sub)
300                 return 0;
301
302         if (o->reset)
303                 flags |= SUBMODULE_MOVE_HEAD_FORCE;
304
305         if (submodule_move_head(ce->name, old_id, new_id, flags))
306                 return add_rejected_path(o, ERROR_WOULD_LOSE_SUBMODULE, ce->name);
307         return 0;
308 }
309
310 /*
311  * Perform the loading of the repository's gitmodules file.  This function is
312  * used by 'check_update()' to perform loading of the gitmodules file in two
313  * different situations:
314  * (1) before removing entries from the working tree if the gitmodules file has
315  *     been marked for removal.  This situation is specified by 'state' == NULL.
316  * (2) before checking out entries to the working tree if the gitmodules file
317  *     has been marked for update.  This situation is specified by 'state' != NULL.
318  */
319 static void load_gitmodules_file(struct index_state *index,
320                                  struct checkout *state)
321 {
322         int pos = index_name_pos(index, GITMODULES_FILE, strlen(GITMODULES_FILE));
323
324         if (pos >= 0) {
325                 struct cache_entry *ce = index->cache[pos];
326                 if (!state && ce->ce_flags & CE_WT_REMOVE) {
327                         repo_read_gitmodules(the_repository, 0);
328                 } else if (state && (ce->ce_flags & CE_UPDATE)) {
329                         submodule_free(the_repository);
330                         checkout_entry(ce, state, NULL, NULL);
331                         repo_read_gitmodules(the_repository, 0);
332                 }
333         }
334 }
335
336 static struct progress *get_progress(struct unpack_trees_options *o,
337                                      struct index_state *index)
338 {
339         unsigned cnt = 0, total = 0;
340
341         if (!o->update || !o->verbose_update)
342                 return NULL;
343
344         for (; cnt < index->cache_nr; cnt++) {
345                 const struct cache_entry *ce = index->cache[cnt];
346                 if (ce->ce_flags & (CE_UPDATE | CE_WT_REMOVE))
347                         total++;
348         }
349
350         return start_delayed_progress(_("Updating files"), total);
351 }
352
353 static void setup_collided_checkout_detection(struct checkout *state,
354                                               struct index_state *index)
355 {
356         int i;
357
358         state->clone = 1;
359         for (i = 0; i < index->cache_nr; i++)
360                 index->cache[i]->ce_flags &= ~CE_MATCHED;
361 }
362
363 static void report_collided_checkout(struct index_state *index)
364 {
365         struct string_list list = STRING_LIST_INIT_NODUP;
366         int i;
367
368         for (i = 0; i < index->cache_nr; i++) {
369                 struct cache_entry *ce = index->cache[i];
370
371                 if (!(ce->ce_flags & CE_MATCHED))
372                         continue;
373
374                 string_list_append(&list, ce->name);
375                 ce->ce_flags &= ~CE_MATCHED;
376         }
377
378         list.cmp = fspathcmp;
379         string_list_sort(&list);
380
381         if (list.nr) {
382                 warning(_("the following paths have collided (e.g. case-sensitive paths\n"
383                           "on a case-insensitive filesystem) and only one from the same\n"
384                           "colliding group is in the working tree:\n"));
385
386                 for (i = 0; i < list.nr; i++)
387                         fprintf(stderr, "  '%s'\n", list.items[i].string);
388         }
389
390         string_list_clear(&list, 0);
391 }
392
393 static int check_updates(struct unpack_trees_options *o,
394                          struct index_state *index)
395 {
396         unsigned cnt = 0;
397         int errs = 0;
398         struct progress *progress;
399         struct checkout state = CHECKOUT_INIT;
400         int i;
401
402         trace_performance_enter();
403         state.force = 1;
404         state.quiet = 1;
405         state.refresh_cache = 1;
406         state.istate = index;
407         clone_checkout_metadata(&state.meta, &o->meta, NULL);
408
409         if (!o->update || o->dry_run) {
410                 remove_marked_cache_entries(index, 0);
411                 trace_performance_leave("check_updates");
412                 return 0;
413         }
414
415         if (o->clone)
416                 setup_collided_checkout_detection(&state, index);
417
418         progress = get_progress(o, index);
419
420         /* Start with clean cache to avoid using any possibly outdated info. */
421         invalidate_lstat_cache();
422
423         git_attr_set_direction(GIT_ATTR_CHECKOUT);
424
425         if (should_update_submodules())
426                 load_gitmodules_file(index, NULL);
427
428         for (i = 0; i < index->cache_nr; i++) {
429                 const struct cache_entry *ce = index->cache[i];
430
431                 if (ce->ce_flags & CE_WT_REMOVE) {
432                         display_progress(progress, ++cnt);
433                         unlink_entry(ce);
434                 }
435         }
436
437         remove_marked_cache_entries(index, 0);
438         remove_scheduled_dirs();
439
440         if (should_update_submodules())
441                 load_gitmodules_file(index, &state);
442
443         enable_delayed_checkout(&state);
444         if (has_promisor_remote()) {
445                 /*
446                  * Prefetch the objects that are to be checked out in the loop
447                  * below.
448                  */
449                 struct oid_array to_fetch = OID_ARRAY_INIT;
450                 for (i = 0; i < index->cache_nr; i++) {
451                         struct cache_entry *ce = index->cache[i];
452
453                         if (!(ce->ce_flags & CE_UPDATE) ||
454                             S_ISGITLINK(ce->ce_mode))
455                                 continue;
456                         if (!oid_object_info_extended(the_repository, &ce->oid,
457                                                       NULL,
458                                                       OBJECT_INFO_FOR_PREFETCH))
459                                 continue;
460                         oid_array_append(&to_fetch, &ce->oid);
461                 }
462                 promisor_remote_get_direct(the_repository,
463                                            to_fetch.oid, to_fetch.nr);
464                 oid_array_clear(&to_fetch);
465         }
466         for (i = 0; i < index->cache_nr; i++) {
467                 struct cache_entry *ce = index->cache[i];
468
469                 if (ce->ce_flags & CE_UPDATE) {
470                         if (ce->ce_flags & CE_WT_REMOVE)
471                                 BUG("both update and delete flags are set on %s",
472                                     ce->name);
473                         display_progress(progress, ++cnt);
474                         ce->ce_flags &= ~CE_UPDATE;
475                         errs |= checkout_entry(ce, &state, NULL, NULL);
476                 }
477         }
478         stop_progress(&progress);
479         errs |= finish_delayed_checkout(&state, NULL);
480         git_attr_set_direction(GIT_ATTR_CHECKIN);
481
482         if (o->clone)
483                 report_collided_checkout(index);
484
485         trace_performance_leave("check_updates");
486         return errs != 0;
487 }
488
489 static int verify_uptodate_sparse(const struct cache_entry *ce,
490                                   struct unpack_trees_options *o);
491 static int verify_absent_sparse(const struct cache_entry *ce,
492                                 enum unpack_trees_error_types,
493                                 struct unpack_trees_options *o);
494
495 static int apply_sparse_checkout(struct index_state *istate,
496                                  struct cache_entry *ce,
497                                  struct unpack_trees_options *o)
498 {
499         int was_skip_worktree = ce_skip_worktree(ce);
500
501         if (ce->ce_flags & CE_NEW_SKIP_WORKTREE)
502                 ce->ce_flags |= CE_SKIP_WORKTREE;
503         else
504                 ce->ce_flags &= ~CE_SKIP_WORKTREE;
505         if (was_skip_worktree != ce_skip_worktree(ce)) {
506                 ce->ce_flags |= CE_UPDATE_IN_BASE;
507                 mark_fsmonitor_invalid(istate, ce);
508                 istate->cache_changed |= CE_ENTRY_CHANGED;
509         }
510
511         /*
512          * if (!was_skip_worktree && !ce_skip_worktree()) {
513          *      This is perfectly normal. Move on;
514          * }
515          */
516
517         /*
518          * Merge strategies may set CE_UPDATE|CE_REMOVE outside checkout
519          * area as a result of ce_skip_worktree() shortcuts in
520          * verify_absent() and verify_uptodate().
521          * Make sure they don't modify worktree if they are already
522          * outside checkout area
523          */
524         if (was_skip_worktree && ce_skip_worktree(ce)) {
525                 ce->ce_flags &= ~CE_UPDATE;
526
527                 /*
528                  * By default, when CE_REMOVE is on, CE_WT_REMOVE is also
529                  * on to get that file removed from both index and worktree.
530                  * If that file is already outside worktree area, don't
531                  * bother remove it.
532                  */
533                 if (ce->ce_flags & CE_REMOVE)
534                         ce->ce_flags &= ~CE_WT_REMOVE;
535         }
536
537         if (!was_skip_worktree && ce_skip_worktree(ce)) {
538                 /*
539                  * If CE_UPDATE is set, verify_uptodate() must be called already
540                  * also stat info may have lost after merged_entry() so calling
541                  * verify_uptodate() again may fail
542                  */
543                 if (!(ce->ce_flags & CE_UPDATE) &&
544                     verify_uptodate_sparse(ce, o)) {
545                         ce->ce_flags &= ~CE_SKIP_WORKTREE;
546                         return -1;
547                 }
548                 ce->ce_flags |= CE_WT_REMOVE;
549                 ce->ce_flags &= ~CE_UPDATE;
550         }
551         if (was_skip_worktree && !ce_skip_worktree(ce)) {
552                 if (verify_absent_sparse(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
553                         return -1;
554                 ce->ce_flags |= CE_UPDATE;
555         }
556         return 0;
557 }
558
559 static int warn_conflicted_path(struct index_state *istate,
560                                 int i,
561                                 struct unpack_trees_options *o)
562 {
563         char *conflicting_path = istate->cache[i]->name;
564         int count = 0;
565
566         add_rejected_path(o, WARNING_SPARSE_UNMERGED_FILE, conflicting_path);
567
568         /* Find out how many higher stage entries are at same path */
569         while ((++count) + i < istate->cache_nr &&
570                !strcmp(conflicting_path, istate->cache[count + i]->name))
571                 ; /* do nothing */
572
573         return count;
574 }
575
576 static inline int call_unpack_fn(const struct cache_entry * const *src,
577                                  struct unpack_trees_options *o)
578 {
579         int ret = o->fn(src, o);
580         if (ret > 0)
581                 ret = 0;
582         return ret;
583 }
584
585 static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
586 {
587         ce->ce_flags |= CE_UNPACKED;
588
589         if (o->cache_bottom < o->src_index->cache_nr &&
590             o->src_index->cache[o->cache_bottom] == ce) {
591                 int bottom = o->cache_bottom;
592                 while (bottom < o->src_index->cache_nr &&
593                        o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
594                         bottom++;
595                 o->cache_bottom = bottom;
596         }
597 }
598
599 static void mark_all_ce_unused(struct index_state *index)
600 {
601         int i;
602         for (i = 0; i < index->cache_nr; i++)
603                 index->cache[i]->ce_flags &= ~(CE_UNPACKED | CE_ADDED | CE_NEW_SKIP_WORKTREE);
604 }
605
606 static int locate_in_src_index(const struct cache_entry *ce,
607                                struct unpack_trees_options *o)
608 {
609         struct index_state *index = o->src_index;
610         int len = ce_namelen(ce);
611         int pos = index_name_pos(index, ce->name, len);
612         if (pos < 0)
613                 pos = -1 - pos;
614         return pos;
615 }
616
617 /*
618  * We call unpack_index_entry() with an unmerged cache entry
619  * only in diff-index, and it wants a single callback.  Skip
620  * the other unmerged entry with the same name.
621  */
622 static void mark_ce_used_same_name(struct cache_entry *ce,
623                                    struct unpack_trees_options *o)
624 {
625         struct index_state *index = o->src_index;
626         int len = ce_namelen(ce);
627         int pos;
628
629         for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
630                 struct cache_entry *next = index->cache[pos];
631                 if (len != ce_namelen(next) ||
632                     memcmp(ce->name, next->name, len))
633                         break;
634                 mark_ce_used(next, o);
635         }
636 }
637
638 static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
639 {
640         const struct index_state *index = o->src_index;
641         int pos = o->cache_bottom;
642
643         while (pos < index->cache_nr) {
644                 struct cache_entry *ce = index->cache[pos];
645                 if (!(ce->ce_flags & CE_UNPACKED))
646                         return ce;
647                 pos++;
648         }
649         return NULL;
650 }
651
652 static void add_same_unmerged(const struct cache_entry *ce,
653                               struct unpack_trees_options *o)
654 {
655         struct index_state *index = o->src_index;
656         int len = ce_namelen(ce);
657         int pos = index_name_pos(index, ce->name, len);
658
659         if (0 <= pos)
660                 die("programming error in a caller of mark_ce_used_same_name");
661         for (pos = -pos - 1; pos < index->cache_nr; pos++) {
662                 struct cache_entry *next = index->cache[pos];
663                 if (len != ce_namelen(next) ||
664                     memcmp(ce->name, next->name, len))
665                         break;
666                 add_entry(o, next, 0, 0);
667                 mark_ce_used(next, o);
668         }
669 }
670
671 static int unpack_index_entry(struct cache_entry *ce,
672                               struct unpack_trees_options *o)
673 {
674         const struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
675         int ret;
676
677         src[0] = ce;
678
679         mark_ce_used(ce, o);
680         if (ce_stage(ce)) {
681                 if (o->skip_unmerged) {
682                         add_entry(o, ce, 0, 0);
683                         return 0;
684                 }
685         }
686         ret = call_unpack_fn(src, o);
687         if (ce_stage(ce))
688                 mark_ce_used_same_name(ce, o);
689         return ret;
690 }
691
692 static int find_cache_pos(struct traverse_info *, const char *p, size_t len);
693
694 static void restore_cache_bottom(struct traverse_info *info, int bottom)
695 {
696         struct unpack_trees_options *o = info->data;
697
698         if (o->diff_index_cached)
699                 return;
700         o->cache_bottom = bottom;
701 }
702
703 static int switch_cache_bottom(struct traverse_info *info)
704 {
705         struct unpack_trees_options *o = info->data;
706         int ret, pos;
707
708         if (o->diff_index_cached)
709                 return 0;
710         ret = o->cache_bottom;
711         pos = find_cache_pos(info->prev, info->name, info->namelen);
712
713         if (pos < -1)
714                 o->cache_bottom = -2 - pos;
715         else if (pos < 0)
716                 o->cache_bottom = o->src_index->cache_nr;
717         return ret;
718 }
719
720 static inline int are_same_oid(struct name_entry *name_j, struct name_entry *name_k)
721 {
722         return !is_null_oid(&name_j->oid) && !is_null_oid(&name_k->oid) && oideq(&name_j->oid, &name_k->oid);
723 }
724
725 static int all_trees_same_as_cache_tree(int n, unsigned long dirmask,
726                                         struct name_entry *names,
727                                         struct traverse_info *info)
728 {
729         struct unpack_trees_options *o = info->data;
730         int i;
731
732         if (!o->merge || dirmask != ((1 << n) - 1))
733                 return 0;
734
735         for (i = 1; i < n; i++)
736                 if (!are_same_oid(names, names + i))
737                         return 0;
738
739         return cache_tree_matches_traversal(o->src_index->cache_tree, names, info);
740 }
741
742 static int index_pos_by_traverse_info(struct name_entry *names,
743                                       struct traverse_info *info)
744 {
745         struct unpack_trees_options *o = info->data;
746         struct strbuf name = STRBUF_INIT;
747         int pos;
748
749         strbuf_make_traverse_path(&name, info, names->path, names->pathlen);
750         strbuf_addch(&name, '/');
751         pos = index_name_pos(o->src_index, name.buf, name.len);
752         if (pos >= 0)
753                 BUG("This is a directory and should not exist in index");
754         pos = -pos - 1;
755         if (pos >= o->src_index->cache_nr ||
756             !starts_with(o->src_index->cache[pos]->name, name.buf) ||
757             (pos > 0 && starts_with(o->src_index->cache[pos-1]->name, name.buf)))
758                 BUG("pos %d doesn't point to the first entry of %s in index",
759                     pos, name.buf);
760         strbuf_release(&name);
761         return pos;
762 }
763
764 /*
765  * Fast path if we detect that all trees are the same as cache-tree at this
766  * path. We'll walk these trees in an iterative loop using cache-tree/index
767  * instead of ODB since we already know what these trees contain.
768  */
769 static int traverse_by_cache_tree(int pos, int nr_entries, int nr_names,
770                                   struct traverse_info *info)
771 {
772         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
773         struct unpack_trees_options *o = info->data;
774         struct cache_entry *tree_ce = NULL;
775         int ce_len = 0;
776         int i, d;
777
778         if (!o->merge)
779                 BUG("We need cache-tree to do this optimization");
780
781         /*
782          * Do what unpack_callback() and unpack_nondirectories() normally
783          * do. But we walk all paths in an iterative loop instead.
784          *
785          * D/F conflicts and higher stage entries are not a concern
786          * because cache-tree would be invalidated and we would never
787          * get here in the first place.
788          */
789         for (i = 0; i < nr_entries; i++) {
790                 int new_ce_len, len, rc;
791
792                 src[0] = o->src_index->cache[pos + i];
793
794                 len = ce_namelen(src[0]);
795                 new_ce_len = cache_entry_size(len);
796
797                 if (new_ce_len > ce_len) {
798                         new_ce_len <<= 1;
799                         tree_ce = xrealloc(tree_ce, new_ce_len);
800                         memset(tree_ce, 0, new_ce_len);
801                         ce_len = new_ce_len;
802
803                         tree_ce->ce_flags = create_ce_flags(0);
804
805                         for (d = 1; d <= nr_names; d++)
806                                 src[d] = tree_ce;
807                 }
808
809                 tree_ce->ce_mode = src[0]->ce_mode;
810                 tree_ce->ce_namelen = len;
811                 oidcpy(&tree_ce->oid, &src[0]->oid);
812                 memcpy(tree_ce->name, src[0]->name, len + 1);
813
814                 rc = call_unpack_fn((const struct cache_entry * const *)src, o);
815                 if (rc < 0) {
816                         free(tree_ce);
817                         return rc;
818                 }
819
820                 mark_ce_used(src[0], o);
821         }
822         free(tree_ce);
823         if (o->debug_unpack)
824                 printf("Unpacked %d entries from %s to %s using cache-tree\n",
825                        nr_entries,
826                        o->src_index->cache[pos]->name,
827                        o->src_index->cache[pos + nr_entries - 1]->name);
828         return 0;
829 }
830
831 static int traverse_trees_recursive(int n, unsigned long dirmask,
832                                     unsigned long df_conflicts,
833                                     struct name_entry *names,
834                                     struct traverse_info *info)
835 {
836         struct unpack_trees_options *o = info->data;
837         int i, ret, bottom;
838         int nr_buf = 0;
839         struct tree_desc t[MAX_UNPACK_TREES];
840         void *buf[MAX_UNPACK_TREES];
841         struct traverse_info newinfo;
842         struct name_entry *p;
843         int nr_entries;
844
845         nr_entries = all_trees_same_as_cache_tree(n, dirmask, names, info);
846         if (nr_entries > 0) {
847                 int pos = index_pos_by_traverse_info(names, info);
848
849                 if (!o->merge || df_conflicts)
850                         BUG("Wrong condition to get here buddy");
851
852                 /*
853                  * All entries up to 'pos' must have been processed
854                  * (i.e. marked CE_UNPACKED) at this point. But to be safe,
855                  * save and restore cache_bottom anyway to not miss
856                  * unprocessed entries before 'pos'.
857                  */
858                 bottom = o->cache_bottom;
859                 ret = traverse_by_cache_tree(pos, nr_entries, n, info);
860                 o->cache_bottom = bottom;
861                 return ret;
862         }
863
864         p = names;
865         while (!p->mode)
866                 p++;
867
868         newinfo = *info;
869         newinfo.prev = info;
870         newinfo.pathspec = info->pathspec;
871         newinfo.name = p->path;
872         newinfo.namelen = p->pathlen;
873         newinfo.mode = p->mode;
874         newinfo.pathlen = st_add3(newinfo.pathlen, tree_entry_len(p), 1);
875         newinfo.df_conflicts |= df_conflicts;
876
877         /*
878          * Fetch the tree from the ODB for each peer directory in the
879          * n commits.
880          *
881          * For 2- and 3-way traversals, we try to avoid hitting the
882          * ODB twice for the same OID.  This should yield a nice speed
883          * up in checkouts and merges when the commits are similar.
884          *
885          * We don't bother doing the full O(n^2) search for larger n,
886          * because wider traversals don't happen that often and we
887          * avoid the search setup.
888          *
889          * When 2 peer OIDs are the same, we just copy the tree
890          * descriptor data.  This implicitly borrows the buffer
891          * data from the earlier cell.
892          */
893         for (i = 0; i < n; i++, dirmask >>= 1) {
894                 if (i > 0 && are_same_oid(&names[i], &names[i - 1]))
895                         t[i] = t[i - 1];
896                 else if (i > 1 && are_same_oid(&names[i], &names[i - 2]))
897                         t[i] = t[i - 2];
898                 else {
899                         const struct object_id *oid = NULL;
900                         if (dirmask & 1)
901                                 oid = &names[i].oid;
902                         buf[nr_buf++] = fill_tree_descriptor(the_repository, t + i, oid);
903                 }
904         }
905
906         bottom = switch_cache_bottom(&newinfo);
907         ret = traverse_trees(o->src_index, n, t, &newinfo);
908         restore_cache_bottom(&newinfo, bottom);
909
910         for (i = 0; i < nr_buf; i++)
911                 free(buf[i]);
912
913         return ret;
914 }
915
916 /*
917  * Compare the traverse-path to the cache entry without actually
918  * having to generate the textual representation of the traverse
919  * path.
920  *
921  * NOTE! This *only* compares up to the size of the traverse path
922  * itself - the caller needs to do the final check for the cache
923  * entry having more data at the end!
924  */
925 static int do_compare_entry_piecewise(const struct cache_entry *ce,
926                                       const struct traverse_info *info,
927                                       const char *name, size_t namelen,
928                                       unsigned mode)
929 {
930         int pathlen, ce_len;
931         const char *ce_name;
932
933         if (info->prev) {
934                 int cmp = do_compare_entry_piecewise(ce, info->prev,
935                                                      info->name, info->namelen,
936                                                      info->mode);
937                 if (cmp)
938                         return cmp;
939         }
940         pathlen = info->pathlen;
941         ce_len = ce_namelen(ce);
942
943         /* If ce_len < pathlen then we must have previously hit "name == directory" entry */
944         if (ce_len < pathlen)
945                 return -1;
946
947         ce_len -= pathlen;
948         ce_name = ce->name + pathlen;
949
950         return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
951 }
952
953 static int do_compare_entry(const struct cache_entry *ce,
954                             const struct traverse_info *info,
955                             const char *name, size_t namelen,
956                             unsigned mode)
957 {
958         int pathlen, ce_len;
959         const char *ce_name;
960         int cmp;
961
962         /*
963          * If we have not precomputed the traverse path, it is quicker
964          * to avoid doing so.  But if we have precomputed it,
965          * it is quicker to use the precomputed version.
966          */
967         if (!info->traverse_path)
968                 return do_compare_entry_piecewise(ce, info, name, namelen, mode);
969
970         cmp = strncmp(ce->name, info->traverse_path, info->pathlen);
971         if (cmp)
972                 return cmp;
973
974         pathlen = info->pathlen;
975         ce_len = ce_namelen(ce);
976
977         if (ce_len < pathlen)
978                 return -1;
979
980         ce_len -= pathlen;
981         ce_name = ce->name + pathlen;
982
983         return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
984 }
985
986 static int compare_entry(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
987 {
988         int cmp = do_compare_entry(ce, info, n->path, n->pathlen, n->mode);
989         if (cmp)
990                 return cmp;
991
992         /*
993          * Even if the beginning compared identically, the ce should
994          * compare as bigger than a directory leading up to it!
995          */
996         return ce_namelen(ce) > traverse_path_len(info, tree_entry_len(n));
997 }
998
999 static int ce_in_traverse_path(const struct cache_entry *ce,
1000                                const struct traverse_info *info)
1001 {
1002         if (!info->prev)
1003                 return 1;
1004         if (do_compare_entry(ce, info->prev,
1005                              info->name, info->namelen, info->mode))
1006                 return 0;
1007         /*
1008          * If ce (blob) is the same name as the path (which is a tree
1009          * we will be descending into), it won't be inside it.
1010          */
1011         return (info->pathlen < ce_namelen(ce));
1012 }
1013
1014 static struct cache_entry *create_ce_entry(const struct traverse_info *info,
1015         const struct name_entry *n,
1016         int stage,
1017         struct index_state *istate,
1018         int is_transient)
1019 {
1020         size_t len = traverse_path_len(info, tree_entry_len(n));
1021         struct cache_entry *ce =
1022                 is_transient ?
1023                 make_empty_transient_cache_entry(len) :
1024                 make_empty_cache_entry(istate, len);
1025
1026         ce->ce_mode = create_ce_mode(n->mode);
1027         ce->ce_flags = create_ce_flags(stage);
1028         ce->ce_namelen = len;
1029         oidcpy(&ce->oid, &n->oid);
1030         /* len+1 because the cache_entry allocates space for NUL */
1031         make_traverse_path(ce->name, len + 1, info, n->path, n->pathlen);
1032
1033         return ce;
1034 }
1035
1036 /*
1037  * Note that traverse_by_cache_tree() duplicates some logic in this function
1038  * without actually calling it. If you change the logic here you may need to
1039  * check and change there as well.
1040  */
1041 static int unpack_nondirectories(int n, unsigned long mask,
1042                                  unsigned long dirmask,
1043                                  struct cache_entry **src,
1044                                  const struct name_entry *names,
1045                                  const struct traverse_info *info)
1046 {
1047         int i;
1048         struct unpack_trees_options *o = info->data;
1049         unsigned long conflicts = info->df_conflicts | dirmask;
1050
1051         /* Do we have *only* directories? Nothing to do */
1052         if (mask == dirmask && !src[0])
1053                 return 0;
1054
1055         /*
1056          * Ok, we've filled in up to any potential index entry in src[0],
1057          * now do the rest.
1058          */
1059         for (i = 0; i < n; i++) {
1060                 int stage;
1061                 unsigned int bit = 1ul << i;
1062                 if (conflicts & bit) {
1063                         src[i + o->merge] = o->df_conflict_entry;
1064                         continue;
1065                 }
1066                 if (!(mask & bit))
1067                         continue;
1068                 if (!o->merge)
1069                         stage = 0;
1070                 else if (i + 1 < o->head_idx)
1071                         stage = 1;
1072                 else if (i + 1 > o->head_idx)
1073                         stage = 3;
1074                 else
1075                         stage = 2;
1076
1077                 /*
1078                  * If the merge bit is set, then the cache entries are
1079                  * discarded in the following block.  In this case,
1080                  * construct "transient" cache_entries, as they are
1081                  * not stored in the index.  otherwise construct the
1082                  * cache entry from the index aware logic.
1083                  */
1084                 src[i + o->merge] = create_ce_entry(info, names + i, stage, &o->result, o->merge);
1085         }
1086
1087         if (o->merge) {
1088                 int rc = call_unpack_fn((const struct cache_entry * const *)src,
1089                                         o);
1090                 for (i = 0; i < n; i++) {
1091                         struct cache_entry *ce = src[i + o->merge];
1092                         if (ce != o->df_conflict_entry)
1093                                 discard_cache_entry(ce);
1094                 }
1095                 return rc;
1096         }
1097
1098         for (i = 0; i < n; i++)
1099                 if (src[i] && src[i] != o->df_conflict_entry)
1100                         if (do_add_entry(o, src[i], 0, 0))
1101                                 return -1;
1102
1103         return 0;
1104 }
1105
1106 static int unpack_failed(struct unpack_trees_options *o, const char *message)
1107 {
1108         discard_index(&o->result);
1109         if (!o->quiet && !o->exiting_early) {
1110                 if (message)
1111                         return error("%s", message);
1112                 return -1;
1113         }
1114         return -1;
1115 }
1116
1117 /*
1118  * The tree traversal is looking at name p.  If we have a matching entry,
1119  * return it.  If name p is a directory in the index, do not return
1120  * anything, as we will want to match it when the traversal descends into
1121  * the directory.
1122  */
1123 static int find_cache_pos(struct traverse_info *info,
1124                           const char *p, size_t p_len)
1125 {
1126         int pos;
1127         struct unpack_trees_options *o = info->data;
1128         struct index_state *index = o->src_index;
1129         int pfxlen = info->pathlen;
1130
1131         for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
1132                 const struct cache_entry *ce = index->cache[pos];
1133                 const char *ce_name, *ce_slash;
1134                 int cmp, ce_len;
1135
1136                 if (ce->ce_flags & CE_UNPACKED) {
1137                         /*
1138                          * cache_bottom entry is already unpacked, so
1139                          * we can never match it; don't check it
1140                          * again.
1141                          */
1142                         if (pos == o->cache_bottom)
1143                                 ++o->cache_bottom;
1144                         continue;
1145                 }
1146                 if (!ce_in_traverse_path(ce, info)) {
1147                         /*
1148                          * Check if we can skip future cache checks
1149                          * (because we're already past all possible
1150                          * entries in the traverse path).
1151                          */
1152                         if (info->traverse_path) {
1153                                 if (strncmp(ce->name, info->traverse_path,
1154                                             info->pathlen) > 0)
1155                                         break;
1156                         }
1157                         continue;
1158                 }
1159                 ce_name = ce->name + pfxlen;
1160                 ce_slash = strchr(ce_name, '/');
1161                 if (ce_slash)
1162                         ce_len = ce_slash - ce_name;
1163                 else
1164                         ce_len = ce_namelen(ce) - pfxlen;
1165                 cmp = name_compare(p, p_len, ce_name, ce_len);
1166                 /*
1167                  * Exact match; if we have a directory we need to
1168                  * delay returning it.
1169                  */
1170                 if (!cmp)
1171                         return ce_slash ? -2 - pos : pos;
1172                 if (0 < cmp)
1173                         continue; /* keep looking */
1174                 /*
1175                  * ce_name sorts after p->path; could it be that we
1176                  * have files under p->path directory in the index?
1177                  * E.g.  ce_name == "t-i", and p->path == "t"; we may
1178                  * have "t/a" in the index.
1179                  */
1180                 if (p_len < ce_len && !memcmp(ce_name, p, p_len) &&
1181                     ce_name[p_len] < '/')
1182                         continue; /* keep looking */
1183                 break;
1184         }
1185         return -1;
1186 }
1187
1188 static struct cache_entry *find_cache_entry(struct traverse_info *info,
1189                                             const struct name_entry *p)
1190 {
1191         int pos = find_cache_pos(info, p->path, p->pathlen);
1192         struct unpack_trees_options *o = info->data;
1193
1194         if (0 <= pos)
1195                 return o->src_index->cache[pos];
1196         else
1197                 return NULL;
1198 }
1199
1200 static void debug_path(struct traverse_info *info)
1201 {
1202         if (info->prev) {
1203                 debug_path(info->prev);
1204                 if (*info->prev->name)
1205                         putchar('/');
1206         }
1207         printf("%s", info->name);
1208 }
1209
1210 static void debug_name_entry(int i, struct name_entry *n)
1211 {
1212         printf("ent#%d %06o %s\n", i,
1213                n->path ? n->mode : 0,
1214                n->path ? n->path : "(missing)");
1215 }
1216
1217 static void debug_unpack_callback(int n,
1218                                   unsigned long mask,
1219                                   unsigned long dirmask,
1220                                   struct name_entry *names,
1221                                   struct traverse_info *info)
1222 {
1223         int i;
1224         printf("* unpack mask %lu, dirmask %lu, cnt %d ",
1225                mask, dirmask, n);
1226         debug_path(info);
1227         putchar('\n');
1228         for (i = 0; i < n; i++)
1229                 debug_name_entry(i, names + i);
1230 }
1231
1232 /*
1233  * Note that traverse_by_cache_tree() duplicates some logic in this function
1234  * without actually calling it. If you change the logic here you may need to
1235  * check and change there as well.
1236  */
1237 static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
1238 {
1239         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
1240         struct unpack_trees_options *o = info->data;
1241         const struct name_entry *p = names;
1242
1243         /* Find first entry with a real name (we could use "mask" too) */
1244         while (!p->mode)
1245                 p++;
1246
1247         if (o->debug_unpack)
1248                 debug_unpack_callback(n, mask, dirmask, names, info);
1249
1250         /* Are we supposed to look at the index too? */
1251         if (o->merge) {
1252                 while (1) {
1253                         int cmp;
1254                         struct cache_entry *ce;
1255
1256                         if (o->diff_index_cached)
1257                                 ce = next_cache_entry(o);
1258                         else
1259                                 ce = find_cache_entry(info, p);
1260
1261                         if (!ce)
1262                                 break;
1263                         cmp = compare_entry(ce, info, p);
1264                         if (cmp < 0) {
1265                                 if (unpack_index_entry(ce, o) < 0)
1266                                         return unpack_failed(o, NULL);
1267                                 continue;
1268                         }
1269                         if (!cmp) {
1270                                 if (ce_stage(ce)) {
1271                                         /*
1272                                          * If we skip unmerged index
1273                                          * entries, we'll skip this
1274                                          * entry *and* the tree
1275                                          * entries associated with it!
1276                                          */
1277                                         if (o->skip_unmerged) {
1278                                                 add_same_unmerged(ce, o);
1279                                                 return mask;
1280                                         }
1281                                 }
1282                                 src[0] = ce;
1283                         }
1284                         break;
1285                 }
1286         }
1287
1288         if (unpack_nondirectories(n, mask, dirmask, src, names, info) < 0)
1289                 return -1;
1290
1291         if (o->merge && src[0]) {
1292                 if (ce_stage(src[0]))
1293                         mark_ce_used_same_name(src[0], o);
1294                 else
1295                         mark_ce_used(src[0], o);
1296         }
1297
1298         /* Now handle any directories.. */
1299         if (dirmask) {
1300                 /* special case: "diff-index --cached" looking at a tree */
1301                 if (o->diff_index_cached &&
1302                     n == 1 && dirmask == 1 && S_ISDIR(names->mode)) {
1303                         int matches;
1304                         matches = cache_tree_matches_traversal(o->src_index->cache_tree,
1305                                                                names, info);
1306                         /*
1307                          * Everything under the name matches; skip the
1308                          * entire hierarchy.  diff_index_cached codepath
1309                          * special cases D/F conflicts in such a way that
1310                          * it does not do any look-ahead, so this is safe.
1311                          */
1312                         if (matches) {
1313                                 o->cache_bottom += matches;
1314                                 return mask;
1315                         }
1316                 }
1317
1318                 if (traverse_trees_recursive(n, dirmask, mask & ~dirmask,
1319                                              names, info) < 0)
1320                         return -1;
1321                 return mask;
1322         }
1323
1324         return mask;
1325 }
1326
1327 static int clear_ce_flags_1(struct index_state *istate,
1328                             struct cache_entry **cache, int nr,
1329                             struct strbuf *prefix,
1330                             int select_mask, int clear_mask,
1331                             struct pattern_list *pl,
1332                             enum pattern_match_result default_match,
1333                             int progress_nr);
1334
1335 /* Whole directory matching */
1336 static int clear_ce_flags_dir(struct index_state *istate,
1337                               struct cache_entry **cache, int nr,
1338                               struct strbuf *prefix,
1339                               char *basename,
1340                               int select_mask, int clear_mask,
1341                               struct pattern_list *pl,
1342                               enum pattern_match_result default_match,
1343                               int progress_nr)
1344 {
1345         struct cache_entry **cache_end;
1346         int dtype = DT_DIR;
1347         int rc;
1348         enum pattern_match_result ret, orig_ret;
1349         orig_ret = path_matches_pattern_list(prefix->buf, prefix->len,
1350                                              basename, &dtype, pl, istate);
1351
1352         strbuf_addch(prefix, '/');
1353
1354         /* If undecided, use matching result of parent dir in defval */
1355         if (orig_ret == UNDECIDED)
1356                 ret = default_match;
1357         else
1358                 ret = orig_ret;
1359
1360         for (cache_end = cache; cache_end != cache + nr; cache_end++) {
1361                 struct cache_entry *ce = *cache_end;
1362                 if (strncmp(ce->name, prefix->buf, prefix->len))
1363                         break;
1364         }
1365
1366         if (pl->use_cone_patterns && orig_ret == MATCHED_RECURSIVE) {
1367                 struct cache_entry **ce = cache;
1368                 rc = cache_end - cache;
1369
1370                 while (ce < cache_end) {
1371                         (*ce)->ce_flags &= ~clear_mask;
1372                         ce++;
1373                 }
1374         } else if (pl->use_cone_patterns && orig_ret == NOT_MATCHED) {
1375                 rc = cache_end - cache;
1376         } else {
1377                 rc = clear_ce_flags_1(istate, cache, cache_end - cache,
1378                                       prefix,
1379                                       select_mask, clear_mask,
1380                                       pl, ret,
1381                                       progress_nr);
1382         }
1383
1384         strbuf_setlen(prefix, prefix->len - 1);
1385         return rc;
1386 }
1387
1388 /*
1389  * Traverse the index, find every entry that matches according to
1390  * o->pl. Do "ce_flags &= ~clear_mask" on those entries. Return the
1391  * number of traversed entries.
1392  *
1393  * If select_mask is non-zero, only entries whose ce_flags has on of
1394  * those bits enabled are traversed.
1395  *
1396  * cache        : pointer to an index entry
1397  * prefix_len   : an offset to its path
1398  *
1399  * The current path ("prefix") including the trailing '/' is
1400  *   cache[0]->name[0..(prefix_len-1)]
1401  * Top level path has prefix_len zero.
1402  */
1403 static int clear_ce_flags_1(struct index_state *istate,
1404                             struct cache_entry **cache, int nr,
1405                             struct strbuf *prefix,
1406                             int select_mask, int clear_mask,
1407                             struct pattern_list *pl,
1408                             enum pattern_match_result default_match,
1409                             int progress_nr)
1410 {
1411         struct cache_entry **cache_end = nr ? cache + nr : cache;
1412
1413         /*
1414          * Process all entries that have the given prefix and meet
1415          * select_mask condition
1416          */
1417         while(cache != cache_end) {
1418                 struct cache_entry *ce = *cache;
1419                 const char *name, *slash;
1420                 int len, dtype;
1421                 enum pattern_match_result ret;
1422
1423                 display_progress(istate->progress, progress_nr);
1424
1425                 if (select_mask && !(ce->ce_flags & select_mask)) {
1426                         cache++;
1427                         progress_nr++;
1428                         continue;
1429                 }
1430
1431                 if (prefix->len && strncmp(ce->name, prefix->buf, prefix->len))
1432                         break;
1433
1434                 name = ce->name + prefix->len;
1435                 slash = strchr(name, '/');
1436
1437                 /* If it's a directory, try whole directory match first */
1438                 if (slash) {
1439                         int processed;
1440
1441                         len = slash - name;
1442                         strbuf_add(prefix, name, len);
1443
1444                         processed = clear_ce_flags_dir(istate, cache, cache_end - cache,
1445                                                        prefix,
1446                                                        prefix->buf + prefix->len - len,
1447                                                        select_mask, clear_mask,
1448                                                        pl, default_match,
1449                                                        progress_nr);
1450
1451                         /* clear_c_f_dir eats a whole dir already? */
1452                         if (processed) {
1453                                 cache += processed;
1454                                 progress_nr += processed;
1455                                 strbuf_setlen(prefix, prefix->len - len);
1456                                 continue;
1457                         }
1458
1459                         strbuf_addch(prefix, '/');
1460                         processed = clear_ce_flags_1(istate, cache, cache_end - cache,
1461                                                      prefix,
1462                                                      select_mask, clear_mask, pl,
1463                                                      default_match, progress_nr);
1464
1465                         cache += processed;
1466                         progress_nr += processed;
1467
1468                         strbuf_setlen(prefix, prefix->len - len - 1);
1469                         continue;
1470                 }
1471
1472                 /* Non-directory */
1473                 dtype = ce_to_dtype(ce);
1474                 ret = path_matches_pattern_list(ce->name,
1475                                                 ce_namelen(ce),
1476                                                 name, &dtype, pl, istate);
1477                 if (ret == UNDECIDED)
1478                         ret = default_match;
1479                 if (ret == MATCHED || ret == MATCHED_RECURSIVE)
1480                         ce->ce_flags &= ~clear_mask;
1481                 cache++;
1482                 progress_nr++;
1483         }
1484
1485         display_progress(istate->progress, progress_nr);
1486         return nr - (cache_end - cache);
1487 }
1488
1489 static int clear_ce_flags(struct index_state *istate,
1490                           int select_mask, int clear_mask,
1491                           struct pattern_list *pl,
1492                           int show_progress)
1493 {
1494         static struct strbuf prefix = STRBUF_INIT;
1495         char label[100];
1496         int rval;
1497
1498         strbuf_reset(&prefix);
1499         if (show_progress)
1500                 istate->progress = start_delayed_progress(
1501                                         _("Updating index flags"),
1502                                         istate->cache_nr);
1503
1504         xsnprintf(label, sizeof(label), "clear_ce_flags(0x%08lx,0x%08lx)",
1505                   (unsigned long)select_mask, (unsigned long)clear_mask);
1506         trace2_region_enter("unpack_trees", label, the_repository);
1507         rval = clear_ce_flags_1(istate,
1508                                 istate->cache,
1509                                 istate->cache_nr,
1510                                 &prefix,
1511                                 select_mask, clear_mask,
1512                                 pl, 0, 0);
1513         trace2_region_leave("unpack_trees", label, the_repository);
1514
1515         stop_progress(&istate->progress);
1516         return rval;
1517 }
1518
1519 /*
1520  * Set/Clear CE_NEW_SKIP_WORKTREE according to $GIT_DIR/info/sparse-checkout
1521  */
1522 static void mark_new_skip_worktree(struct pattern_list *pl,
1523                                    struct index_state *istate,
1524                                    int select_flag, int skip_wt_flag,
1525                                    int show_progress)
1526 {
1527         int i;
1528
1529         /*
1530          * 1. Pretend the narrowest worktree: only unmerged entries
1531          * are checked out
1532          */
1533         for (i = 0; i < istate->cache_nr; i++) {
1534                 struct cache_entry *ce = istate->cache[i];
1535
1536                 if (select_flag && !(ce->ce_flags & select_flag))
1537                         continue;
1538
1539                 if (!ce_stage(ce) && !(ce->ce_flags & CE_CONFLICTED))
1540                         ce->ce_flags |= skip_wt_flag;
1541                 else
1542                         ce->ce_flags &= ~skip_wt_flag;
1543         }
1544
1545         /*
1546          * 2. Widen worktree according to sparse-checkout file.
1547          * Matched entries will have skip_wt_flag cleared (i.e. "in")
1548          */
1549         clear_ce_flags(istate, select_flag, skip_wt_flag, pl, show_progress);
1550 }
1551
1552 static void populate_from_existing_patterns(struct unpack_trees_options *o,
1553                                             struct pattern_list *pl)
1554 {
1555         if (get_sparse_checkout_patterns(pl) < 0)
1556                 o->skip_sparse_checkout = 1;
1557         else
1558                 o->pl = pl;
1559 }
1560
1561
1562 static int verify_absent(const struct cache_entry *,
1563                          enum unpack_trees_error_types,
1564                          struct unpack_trees_options *);
1565 /*
1566  * N-way merge "len" trees.  Returns 0 on success, -1 on failure to manipulate the
1567  * resulting index, -2 on failure to reflect the changes to the work tree.
1568  *
1569  * CE_ADDED, CE_UNPACKED and CE_NEW_SKIP_WORKTREE are used internally
1570  */
1571 int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options *o)
1572 {
1573         int i, ret;
1574         static struct cache_entry *dfc;
1575         struct pattern_list pl;
1576         int free_pattern_list = 0;
1577
1578         if (len > MAX_UNPACK_TREES)
1579                 die("unpack_trees takes at most %d trees", MAX_UNPACK_TREES);
1580
1581         trace_performance_enter();
1582         trace2_region_enter("unpack_trees", "unpack_trees", the_repository);
1583
1584         if (!core_apply_sparse_checkout || !o->update)
1585                 o->skip_sparse_checkout = 1;
1586         if (!o->skip_sparse_checkout && !o->pl) {
1587                 memset(&pl, 0, sizeof(pl));
1588                 free_pattern_list = 1;
1589                 populate_from_existing_patterns(o, &pl);
1590         }
1591
1592         memset(&o->result, 0, sizeof(o->result));
1593         o->result.initialized = 1;
1594         o->result.timestamp.sec = o->src_index->timestamp.sec;
1595         o->result.timestamp.nsec = o->src_index->timestamp.nsec;
1596         o->result.version = o->src_index->version;
1597         if (!o->src_index->split_index) {
1598                 o->result.split_index = NULL;
1599         } else if (o->src_index == o->dst_index) {
1600                 /*
1601                  * o->dst_index (and thus o->src_index) will be discarded
1602                  * and overwritten with o->result at the end of this function,
1603                  * so just use src_index's split_index to avoid having to
1604                  * create a new one.
1605                  */
1606                 o->result.split_index = o->src_index->split_index;
1607                 o->result.split_index->refcount++;
1608         } else {
1609                 o->result.split_index = init_split_index(&o->result);
1610         }
1611         oidcpy(&o->result.oid, &o->src_index->oid);
1612         o->merge_size = len;
1613         mark_all_ce_unused(o->src_index);
1614
1615         o->result.fsmonitor_last_update =
1616                 xstrdup_or_null(o->src_index->fsmonitor_last_update);
1617
1618         /*
1619          * Sparse checkout loop #1: set NEW_SKIP_WORKTREE on existing entries
1620          */
1621         if (!o->skip_sparse_checkout)
1622                 mark_new_skip_worktree(o->pl, o->src_index, 0,
1623                                        CE_NEW_SKIP_WORKTREE, o->verbose_update);
1624
1625         if (!dfc)
1626                 dfc = xcalloc(1, cache_entry_size(0));
1627         o->df_conflict_entry = dfc;
1628
1629         if (len) {
1630                 const char *prefix = o->prefix ? o->prefix : "";
1631                 struct traverse_info info;
1632
1633                 setup_traverse_info(&info, prefix);
1634                 info.fn = unpack_callback;
1635                 info.data = o;
1636                 info.show_all_errors = o->show_all_errors;
1637                 info.pathspec = o->pathspec;
1638
1639                 if (o->prefix) {
1640                         /*
1641                          * Unpack existing index entries that sort before the
1642                          * prefix the tree is spliced into.  Note that o->merge
1643                          * is always true in this case.
1644                          */
1645                         while (1) {
1646                                 struct cache_entry *ce = next_cache_entry(o);
1647                                 if (!ce)
1648                                         break;
1649                                 if (ce_in_traverse_path(ce, &info))
1650                                         break;
1651                                 if (unpack_index_entry(ce, o) < 0)
1652                                         goto return_failed;
1653                         }
1654                 }
1655
1656                 trace_performance_enter();
1657                 trace2_region_enter("unpack_trees", "traverse_trees", the_repository);
1658                 ret = traverse_trees(o->src_index, len, t, &info);
1659                 trace2_region_leave("unpack_trees", "traverse_trees", the_repository);
1660                 trace_performance_leave("traverse_trees");
1661                 if (ret < 0)
1662                         goto return_failed;
1663         }
1664
1665         /* Any left-over entries in the index? */
1666         if (o->merge) {
1667                 while (1) {
1668                         struct cache_entry *ce = next_cache_entry(o);
1669                         if (!ce)
1670                                 break;
1671                         if (unpack_index_entry(ce, o) < 0)
1672                                 goto return_failed;
1673                 }
1674         }
1675         mark_all_ce_unused(o->src_index);
1676
1677         if (o->trivial_merges_only && o->nontrivial_merge) {
1678                 ret = unpack_failed(o, "Merge requires file-level merging");
1679                 goto done;
1680         }
1681
1682         if (!o->skip_sparse_checkout) {
1683                 /*
1684                  * Sparse checkout loop #2: set NEW_SKIP_WORKTREE on entries not in loop #1
1685                  * If they will have NEW_SKIP_WORKTREE, also set CE_SKIP_WORKTREE
1686                  * so apply_sparse_checkout() won't attempt to remove it from worktree
1687                  */
1688                 mark_new_skip_worktree(o->pl, &o->result,
1689                                        CE_ADDED, CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE,
1690                                        o->verbose_update);
1691
1692                 ret = 0;
1693                 for (i = 0; i < o->result.cache_nr; i++) {
1694                         struct cache_entry *ce = o->result.cache[i];
1695
1696                         /*
1697                          * Entries marked with CE_ADDED in merged_entry() do not have
1698                          * verify_absent() check (the check is effectively disabled
1699                          * because CE_NEW_SKIP_WORKTREE is set unconditionally).
1700                          *
1701                          * Do the real check now because we have had
1702                          * correct CE_NEW_SKIP_WORKTREE
1703                          */
1704                         if (ce->ce_flags & CE_ADDED &&
1705                             verify_absent(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
1706                                 ret = 1;
1707
1708                         if (apply_sparse_checkout(&o->result, ce, o))
1709                                 ret = 1;
1710                 }
1711                 if (ret == 1) {
1712                         /*
1713                          * Inability to sparsify or de-sparsify individual
1714                          * paths is not an error, but just a warning.
1715                          */
1716                         if (o->show_all_errors)
1717                                 display_warning_msgs(o);
1718                         ret = 0;
1719                 }
1720         }
1721
1722         ret = check_updates(o, &o->result) ? (-2) : 0;
1723         if (o->dst_index) {
1724                 move_index_extensions(&o->result, o->src_index);
1725                 if (!ret) {
1726                         if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
1727                                 cache_tree_verify(the_repository, &o->result);
1728                         if (!cache_tree_fully_valid(o->result.cache_tree))
1729                                 cache_tree_update(&o->result,
1730                                                   WRITE_TREE_SILENT |
1731                                                   WRITE_TREE_REPAIR);
1732                 }
1733
1734                 o->result.updated_workdir = 1;
1735                 discard_index(o->dst_index);
1736                 *o->dst_index = o->result;
1737         } else {
1738                 discard_index(&o->result);
1739         }
1740         o->src_index = NULL;
1741
1742 done:
1743         if (free_pattern_list)
1744                 clear_pattern_list(&pl);
1745         trace2_region_leave("unpack_trees", "unpack_trees", the_repository);
1746         trace_performance_leave("unpack_trees");
1747         return ret;
1748
1749 return_failed:
1750         if (o->show_all_errors)
1751                 display_error_msgs(o);
1752         mark_all_ce_unused(o->src_index);
1753         ret = unpack_failed(o, NULL);
1754         if (o->exiting_early)
1755                 ret = 0;
1756         goto done;
1757 }
1758
1759 /*
1760  * Update SKIP_WORKTREE bits according to sparsity patterns, and update
1761  * working directory to match.
1762  *
1763  * CE_NEW_SKIP_WORKTREE is used internally.
1764  */
1765 enum update_sparsity_result update_sparsity(struct unpack_trees_options *o)
1766 {
1767         enum update_sparsity_result ret = UPDATE_SPARSITY_SUCCESS;
1768         struct pattern_list pl;
1769         int i;
1770         unsigned old_show_all_errors;
1771         int free_pattern_list = 0;
1772
1773         old_show_all_errors = o->show_all_errors;
1774         o->show_all_errors = 1;
1775
1776         /* Sanity checks */
1777         if (!o->update || o->index_only || o->skip_sparse_checkout)
1778                 BUG("update_sparsity() is for reflecting sparsity patterns in working directory");
1779         if (o->src_index != o->dst_index || o->fn)
1780                 BUG("update_sparsity() called wrong");
1781
1782         trace_performance_enter();
1783
1784         /* If we weren't given patterns, use the recorded ones */
1785         if (!o->pl) {
1786                 memset(&pl, 0, sizeof(pl));
1787                 free_pattern_list = 1;
1788                 populate_from_existing_patterns(o, &pl);
1789                 if (o->skip_sparse_checkout)
1790                         goto skip_sparse_checkout;
1791         }
1792
1793         /* Set NEW_SKIP_WORKTREE on existing entries. */
1794         mark_all_ce_unused(o->src_index);
1795         mark_new_skip_worktree(o->pl, o->src_index, 0,
1796                                CE_NEW_SKIP_WORKTREE, o->verbose_update);
1797
1798         /* Then loop over entries and update/remove as needed */
1799         ret = UPDATE_SPARSITY_SUCCESS;
1800         for (i = 0; i < o->src_index->cache_nr; i++) {
1801                 struct cache_entry *ce = o->src_index->cache[i];
1802
1803
1804                 if (ce_stage(ce)) {
1805                         /* -1 because for loop will increment by 1 */
1806                         i += warn_conflicted_path(o->src_index, i, o) - 1;
1807                         ret = UPDATE_SPARSITY_WARNINGS;
1808                         continue;
1809                 }
1810
1811                 if (apply_sparse_checkout(o->src_index, ce, o))
1812                         ret = UPDATE_SPARSITY_WARNINGS;
1813         }
1814
1815 skip_sparse_checkout:
1816         if (check_updates(o, o->src_index))
1817                 ret = UPDATE_SPARSITY_WORKTREE_UPDATE_FAILURES;
1818
1819         display_warning_msgs(o);
1820         o->show_all_errors = old_show_all_errors;
1821         if (free_pattern_list)
1822                 clear_pattern_list(&pl);
1823         trace_performance_leave("update_sparsity");
1824         return ret;
1825 }
1826
1827 /* Here come the merge functions */
1828
1829 static int reject_merge(const struct cache_entry *ce,
1830                         struct unpack_trees_options *o)
1831 {
1832         return add_rejected_path(o, ERROR_WOULD_OVERWRITE, ce->name);
1833 }
1834
1835 static int same(const struct cache_entry *a, const struct cache_entry *b)
1836 {
1837         if (!!a != !!b)
1838                 return 0;
1839         if (!a && !b)
1840                 return 1;
1841         if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
1842                 return 0;
1843         return a->ce_mode == b->ce_mode &&
1844                oideq(&a->oid, &b->oid);
1845 }
1846
1847
1848 /*
1849  * When a CE gets turned into an unmerged entry, we
1850  * want it to be up-to-date
1851  */
1852 static int verify_uptodate_1(const struct cache_entry *ce,
1853                              struct unpack_trees_options *o,
1854                              enum unpack_trees_error_types error_type)
1855 {
1856         struct stat st;
1857
1858         if (o->index_only)
1859                 return 0;
1860
1861         /*
1862          * CE_VALID and CE_SKIP_WORKTREE cheat, we better check again
1863          * if this entry is truly up-to-date because this file may be
1864          * overwritten.
1865          */
1866         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
1867                 ; /* keep checking */
1868         else if (o->reset || ce_uptodate(ce))
1869                 return 0;
1870
1871         if (!lstat(ce->name, &st)) {
1872                 int flags = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE;
1873                 unsigned changed = ie_match_stat(o->src_index, ce, &st, flags);
1874
1875                 if (submodule_from_ce(ce)) {
1876                         int r = check_submodule_move_head(ce,
1877                                 "HEAD", oid_to_hex(&ce->oid), o);
1878                         if (r)
1879                                 return add_rejected_path(o, error_type, ce->name);
1880                         return 0;
1881                 }
1882
1883                 if (!changed)
1884                         return 0;
1885                 /*
1886                  * Historic default policy was to allow submodule to be out
1887                  * of sync wrt the superproject index. If the submodule was
1888                  * not considered interesting above, we don't care here.
1889                  */
1890                 if (S_ISGITLINK(ce->ce_mode))
1891                         return 0;
1892
1893                 errno = 0;
1894         }
1895         if (errno == ENOENT)
1896                 return 0;
1897         return add_rejected_path(o, error_type, ce->name);
1898 }
1899
1900 int verify_uptodate(const struct cache_entry *ce,
1901                     struct unpack_trees_options *o)
1902 {
1903         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1904                 return 0;
1905         return verify_uptodate_1(ce, o, ERROR_NOT_UPTODATE_FILE);
1906 }
1907
1908 static int verify_uptodate_sparse(const struct cache_entry *ce,
1909                                   struct unpack_trees_options *o)
1910 {
1911         return verify_uptodate_1(ce, o, WARNING_SPARSE_NOT_UPTODATE_FILE);
1912 }
1913
1914 /*
1915  * TODO: We should actually invalidate o->result, not src_index [1].
1916  * But since cache tree and untracked cache both are not copied to
1917  * o->result until unpacking is complete, we invalidate them on
1918  * src_index instead with the assumption that they will be copied to
1919  * dst_index at the end.
1920  *
1921  * [1] src_index->cache_tree is also used in unpack_callback() so if
1922  * we invalidate o->result, we need to update it to use
1923  * o->result.cache_tree as well.
1924  */
1925 static void invalidate_ce_path(const struct cache_entry *ce,
1926                                struct unpack_trees_options *o)
1927 {
1928         if (!ce)
1929                 return;
1930         cache_tree_invalidate_path(o->src_index, ce->name);
1931         untracked_cache_invalidate_path(o->src_index, ce->name, 1);
1932 }
1933
1934 /*
1935  * Check that checking out ce->sha1 in subdir ce->name is not
1936  * going to overwrite any working files.
1937  */
1938 static int verify_clean_submodule(const char *old_sha1,
1939                                   const struct cache_entry *ce,
1940                                   struct unpack_trees_options *o)
1941 {
1942         if (!submodule_from_ce(ce))
1943                 return 0;
1944
1945         return check_submodule_move_head(ce, old_sha1,
1946                                          oid_to_hex(&ce->oid), o);
1947 }
1948
1949 static int verify_clean_subdirectory(const struct cache_entry *ce,
1950                                      struct unpack_trees_options *o)
1951 {
1952         /*
1953          * we are about to extract "ce->name"; we would not want to lose
1954          * anything in the existing directory there.
1955          */
1956         int namelen;
1957         int i;
1958         struct dir_struct d;
1959         char *pathbuf;
1960         int cnt = 0;
1961
1962         if (S_ISGITLINK(ce->ce_mode)) {
1963                 struct object_id oid;
1964                 int sub_head = resolve_gitlink_ref(ce->name, "HEAD", &oid);
1965                 /*
1966                  * If we are not going to update the submodule, then
1967                  * we don't care.
1968                  */
1969                 if (!sub_head && oideq(&oid, &ce->oid))
1970                         return 0;
1971                 return verify_clean_submodule(sub_head ? NULL : oid_to_hex(&oid),
1972                                               ce, o);
1973         }
1974
1975         /*
1976          * First let's make sure we do not have a local modification
1977          * in that directory.
1978          */
1979         namelen = ce_namelen(ce);
1980         for (i = locate_in_src_index(ce, o);
1981              i < o->src_index->cache_nr;
1982              i++) {
1983                 struct cache_entry *ce2 = o->src_index->cache[i];
1984                 int len = ce_namelen(ce2);
1985                 if (len < namelen ||
1986                     strncmp(ce->name, ce2->name, namelen) ||
1987                     ce2->name[namelen] != '/')
1988                         break;
1989                 /*
1990                  * ce2->name is an entry in the subdirectory to be
1991                  * removed.
1992                  */
1993                 if (!ce_stage(ce2)) {
1994                         if (verify_uptodate(ce2, o))
1995                                 return -1;
1996                         add_entry(o, ce2, CE_REMOVE, 0);
1997                         invalidate_ce_path(ce, o);
1998                         mark_ce_used(ce2, o);
1999                 }
2000                 cnt++;
2001         }
2002
2003         /*
2004          * Then we need to make sure that we do not lose a locally
2005          * present file that is not ignored.
2006          */
2007         pathbuf = xstrfmt("%.*s/", namelen, ce->name);
2008
2009         memset(&d, 0, sizeof(d));
2010         if (o->dir)
2011                 d.exclude_per_dir = o->dir->exclude_per_dir;
2012         i = read_directory(&d, o->src_index, pathbuf, namelen+1, NULL);
2013         if (i)
2014                 return add_rejected_path(o, ERROR_NOT_UPTODATE_DIR, ce->name);
2015         free(pathbuf);
2016         return cnt;
2017 }
2018
2019 /*
2020  * This gets called when there was no index entry for the tree entry 'dst',
2021  * but we found a file in the working tree that 'lstat()' said was fine,
2022  * and we're on a case-insensitive filesystem.
2023  *
2024  * See if we can find a case-insensitive match in the index that also
2025  * matches the stat information, and assume it's that other file!
2026  */
2027 static int icase_exists(struct unpack_trees_options *o, const char *name, int len, struct stat *st)
2028 {
2029         const struct cache_entry *src;
2030
2031         src = index_file_exists(o->src_index, name, len, 1);
2032         return src && !ie_match_stat(o->src_index, src, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2033 }
2034
2035 static int check_ok_to_remove(const char *name, int len, int dtype,
2036                               const struct cache_entry *ce, struct stat *st,
2037                               enum unpack_trees_error_types error_type,
2038                               struct unpack_trees_options *o)
2039 {
2040         const struct cache_entry *result;
2041
2042         /*
2043          * It may be that the 'lstat()' succeeded even though
2044          * target 'ce' was absent, because there is an old
2045          * entry that is different only in case..
2046          *
2047          * Ignore that lstat() if it matches.
2048          */
2049         if (ignore_case && icase_exists(o, name, len, st))
2050                 return 0;
2051
2052         if (o->dir &&
2053             is_excluded(o->dir, o->src_index, name, &dtype))
2054                 /*
2055                  * ce->name is explicitly excluded, so it is Ok to
2056                  * overwrite it.
2057                  */
2058                 return 0;
2059         if (S_ISDIR(st->st_mode)) {
2060                 /*
2061                  * We are checking out path "foo" and
2062                  * found "foo/." in the working tree.
2063                  * This is tricky -- if we have modified
2064                  * files that are in "foo/" we would lose
2065                  * them.
2066                  */
2067                 if (verify_clean_subdirectory(ce, o) < 0)
2068                         return -1;
2069                 return 0;
2070         }
2071
2072         /*
2073          * The previous round may already have decided to
2074          * delete this path, which is in a subdirectory that
2075          * is being replaced with a blob.
2076          */
2077         result = index_file_exists(&o->result, name, len, 0);
2078         if (result) {
2079                 if (result->ce_flags & CE_REMOVE)
2080                         return 0;
2081         }
2082
2083         return add_rejected_path(o, error_type, name);
2084 }
2085
2086 /*
2087  * We do not want to remove or overwrite a working tree file that
2088  * is not tracked, unless it is ignored.
2089  */
2090 static int verify_absent_1(const struct cache_entry *ce,
2091                            enum unpack_trees_error_types error_type,
2092                            struct unpack_trees_options *o)
2093 {
2094         int len;
2095         struct stat st;
2096
2097         if (o->index_only || o->reset || !o->update)
2098                 return 0;
2099
2100         len = check_leading_path(ce->name, ce_namelen(ce));
2101         if (!len)
2102                 return 0;
2103         else if (len > 0) {
2104                 char *path;
2105                 int ret;
2106
2107                 path = xmemdupz(ce->name, len);
2108                 if (lstat(path, &st))
2109                         ret = error_errno("cannot stat '%s'", path);
2110                 else {
2111                         if (submodule_from_ce(ce))
2112                                 ret = check_submodule_move_head(ce,
2113                                                                 oid_to_hex(&ce->oid),
2114                                                                 NULL, o);
2115                         else
2116                                 ret = check_ok_to_remove(path, len, DT_UNKNOWN, NULL,
2117                                                          &st, error_type, o);
2118                 }
2119                 free(path);
2120                 return ret;
2121         } else if (lstat(ce->name, &st)) {
2122                 if (errno != ENOENT)
2123                         return error_errno("cannot stat '%s'", ce->name);
2124                 return 0;
2125         } else {
2126                 if (submodule_from_ce(ce))
2127                         return check_submodule_move_head(ce, oid_to_hex(&ce->oid),
2128                                                          NULL, o);
2129
2130                 return check_ok_to_remove(ce->name, ce_namelen(ce),
2131                                           ce_to_dtype(ce), ce, &st,
2132                                           error_type, o);
2133         }
2134 }
2135
2136 static int verify_absent(const struct cache_entry *ce,
2137                          enum unpack_trees_error_types error_type,
2138                          struct unpack_trees_options *o)
2139 {
2140         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
2141                 return 0;
2142         return verify_absent_1(ce, error_type, o);
2143 }
2144
2145 static int verify_absent_sparse(const struct cache_entry *ce,
2146                                 enum unpack_trees_error_types error_type,
2147                                 struct unpack_trees_options *o)
2148 {
2149         return verify_absent_1(ce, error_type, o);
2150 }
2151
2152 static int merged_entry(const struct cache_entry *ce,
2153                         const struct cache_entry *old,
2154                         struct unpack_trees_options *o)
2155 {
2156         int update = CE_UPDATE;
2157         struct cache_entry *merge = dup_cache_entry(ce, &o->result);
2158
2159         if (!old) {
2160                 /*
2161                  * New index entries. In sparse checkout, the following
2162                  * verify_absent() will be delayed until after
2163                  * traverse_trees() finishes in unpack_trees(), then:
2164                  *
2165                  *  - CE_NEW_SKIP_WORKTREE will be computed correctly
2166                  *  - verify_absent() be called again, this time with
2167                  *    correct CE_NEW_SKIP_WORKTREE
2168                  *
2169                  * verify_absent() call here does nothing in sparse
2170                  * checkout (i.e. o->skip_sparse_checkout == 0)
2171                  */
2172                 update |= CE_ADDED;
2173                 merge->ce_flags |= CE_NEW_SKIP_WORKTREE;
2174
2175                 if (verify_absent(merge,
2176                                   ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2177                         discard_cache_entry(merge);
2178                         return -1;
2179                 }
2180                 invalidate_ce_path(merge, o);
2181
2182                 if (submodule_from_ce(ce) && file_exists(ce->name)) {
2183                         int ret = check_submodule_move_head(ce, NULL,
2184                                                             oid_to_hex(&ce->oid),
2185                                                             o);
2186                         if (ret)
2187                                 return ret;
2188                 }
2189
2190         } else if (!(old->ce_flags & CE_CONFLICTED)) {
2191                 /*
2192                  * See if we can re-use the old CE directly?
2193                  * That way we get the uptodate stat info.
2194                  *
2195                  * This also removes the UPDATE flag on a match; otherwise
2196                  * we will end up overwriting local changes in the work tree.
2197                  */
2198                 if (same(old, merge)) {
2199                         copy_cache_entry(merge, old);
2200                         update = 0;
2201                 } else {
2202                         if (verify_uptodate(old, o)) {
2203                                 discard_cache_entry(merge);
2204                                 return -1;
2205                         }
2206                         /* Migrate old flags over */
2207                         update |= old->ce_flags & (CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
2208                         invalidate_ce_path(old, o);
2209                 }
2210
2211                 if (submodule_from_ce(ce) && file_exists(ce->name)) {
2212                         int ret = check_submodule_move_head(ce, oid_to_hex(&old->oid),
2213                                                             oid_to_hex(&ce->oid),
2214                                                             o);
2215                         if (ret)
2216                                 return ret;
2217                 }
2218         } else {
2219                 /*
2220                  * Previously unmerged entry left as an existence
2221                  * marker by read_index_unmerged();
2222                  */
2223                 invalidate_ce_path(old, o);
2224         }
2225
2226         if (do_add_entry(o, merge, update, CE_STAGEMASK) < 0)
2227                 return -1;
2228         return 1;
2229 }
2230
2231 static int deleted_entry(const struct cache_entry *ce,
2232                          const struct cache_entry *old,
2233                          struct unpack_trees_options *o)
2234 {
2235         /* Did it exist in the index? */
2236         if (!old) {
2237                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2238                         return -1;
2239                 return 0;
2240         }
2241         if (!(old->ce_flags & CE_CONFLICTED) && verify_uptodate(old, o))
2242                 return -1;
2243         add_entry(o, ce, CE_REMOVE, 0);
2244         invalidate_ce_path(ce, o);
2245         return 1;
2246 }
2247
2248 static int keep_entry(const struct cache_entry *ce,
2249                       struct unpack_trees_options *o)
2250 {
2251         add_entry(o, ce, 0, 0);
2252         if (ce_stage(ce))
2253                 invalidate_ce_path(ce, o);
2254         return 1;
2255 }
2256
2257 #if DBRT_DEBUG
2258 static void show_stage_entry(FILE *o,
2259                              const char *label, const struct cache_entry *ce)
2260 {
2261         if (!ce)
2262                 fprintf(o, "%s (missing)\n", label);
2263         else
2264                 fprintf(o, "%s%06o %s %d\t%s\n",
2265                         label,
2266                         ce->ce_mode,
2267                         oid_to_hex(&ce->oid),
2268                         ce_stage(ce),
2269                         ce->name);
2270 }
2271 #endif
2272
2273 int threeway_merge(const struct cache_entry * const *stages,
2274                    struct unpack_trees_options *o)
2275 {
2276         const struct cache_entry *index;
2277         const struct cache_entry *head;
2278         const struct cache_entry *remote = stages[o->head_idx + 1];
2279         int count;
2280         int head_match = 0;
2281         int remote_match = 0;
2282
2283         int df_conflict_head = 0;
2284         int df_conflict_remote = 0;
2285
2286         int any_anc_missing = 0;
2287         int no_anc_exists = 1;
2288         int i;
2289
2290         for (i = 1; i < o->head_idx; i++) {
2291                 if (!stages[i] || stages[i] == o->df_conflict_entry)
2292                         any_anc_missing = 1;
2293                 else
2294                         no_anc_exists = 0;
2295         }
2296
2297         index = stages[0];
2298         head = stages[o->head_idx];
2299
2300         if (head == o->df_conflict_entry) {
2301                 df_conflict_head = 1;
2302                 head = NULL;
2303         }
2304
2305         if (remote == o->df_conflict_entry) {
2306                 df_conflict_remote = 1;
2307                 remote = NULL;
2308         }
2309
2310         /*
2311          * First, if there's a #16 situation, note that to prevent #13
2312          * and #14.
2313          */
2314         if (!same(remote, head)) {
2315                 for (i = 1; i < o->head_idx; i++) {
2316                         if (same(stages[i], head)) {
2317                                 head_match = i;
2318                         }
2319                         if (same(stages[i], remote)) {
2320                                 remote_match = i;
2321                         }
2322                 }
2323         }
2324
2325         /*
2326          * We start with cases where the index is allowed to match
2327          * something other than the head: #14(ALT) and #2ALT, where it
2328          * is permitted to match the result instead.
2329          */
2330         /* #14, #14ALT, #2ALT */
2331         if (remote && !df_conflict_head && head_match && !remote_match) {
2332                 if (index && !same(index, remote) && !same(index, head))
2333                         return reject_merge(index, o);
2334                 return merged_entry(remote, index, o);
2335         }
2336         /*
2337          * If we have an entry in the index cache, then we want to
2338          * make sure that it matches head.
2339          */
2340         if (index && !same(index, head))
2341                 return reject_merge(index, o);
2342
2343         if (head) {
2344                 /* #5ALT, #15 */
2345                 if (same(head, remote))
2346                         return merged_entry(head, index, o);
2347                 /* #13, #3ALT */
2348                 if (!df_conflict_remote && remote_match && !head_match)
2349                         return merged_entry(head, index, o);
2350         }
2351
2352         /* #1 */
2353         if (!head && !remote && any_anc_missing)
2354                 return 0;
2355
2356         /*
2357          * Under the "aggressive" rule, we resolve mostly trivial
2358          * cases that we historically had git-merge-one-file resolve.
2359          */
2360         if (o->aggressive) {
2361                 int head_deleted = !head;
2362                 int remote_deleted = !remote;
2363                 const struct cache_entry *ce = NULL;
2364
2365                 if (index)
2366                         ce = index;
2367                 else if (head)
2368                         ce = head;
2369                 else if (remote)
2370                         ce = remote;
2371                 else {
2372                         for (i = 1; i < o->head_idx; i++) {
2373                                 if (stages[i] && stages[i] != o->df_conflict_entry) {
2374                                         ce = stages[i];
2375                                         break;
2376                                 }
2377                         }
2378                 }
2379
2380                 /*
2381                  * Deleted in both.
2382                  * Deleted in one and unchanged in the other.
2383                  */
2384                 if ((head_deleted && remote_deleted) ||
2385                     (head_deleted && remote && remote_match) ||
2386                     (remote_deleted && head && head_match)) {
2387                         if (index)
2388                                 return deleted_entry(index, index, o);
2389                         if (ce && !head_deleted) {
2390                                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2391                                         return -1;
2392                         }
2393                         return 0;
2394                 }
2395                 /*
2396                  * Added in both, identically.
2397                  */
2398                 if (no_anc_exists && head && remote && same(head, remote))
2399                         return merged_entry(head, index, o);
2400
2401         }
2402
2403         /* Below are "no merge" cases, which require that the index be
2404          * up-to-date to avoid the files getting overwritten with
2405          * conflict resolution files.
2406          */
2407         if (index) {
2408                 if (verify_uptodate(index, o))
2409                         return -1;
2410         }
2411
2412         o->nontrivial_merge = 1;
2413
2414         /* #2, #3, #4, #6, #7, #9, #10, #11. */
2415         count = 0;
2416         if (!head_match || !remote_match) {
2417                 for (i = 1; i < o->head_idx; i++) {
2418                         if (stages[i] && stages[i] != o->df_conflict_entry) {
2419                                 keep_entry(stages[i], o);
2420                                 count++;
2421                                 break;
2422                         }
2423                 }
2424         }
2425 #if DBRT_DEBUG
2426         else {
2427                 fprintf(stderr, "read-tree: warning #16 detected\n");
2428                 show_stage_entry(stderr, "head   ", stages[head_match]);
2429                 show_stage_entry(stderr, "remote ", stages[remote_match]);
2430         }
2431 #endif
2432         if (head) { count += keep_entry(head, o); }
2433         if (remote) { count += keep_entry(remote, o); }
2434         return count;
2435 }
2436
2437 /*
2438  * Two-way merge.
2439  *
2440  * The rule is to "carry forward" what is in the index without losing
2441  * information across a "fast-forward", favoring a successful merge
2442  * over a merge failure when it makes sense.  For details of the
2443  * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
2444  *
2445  */
2446 int twoway_merge(const struct cache_entry * const *src,
2447                  struct unpack_trees_options *o)
2448 {
2449         const struct cache_entry *current = src[0];
2450         const struct cache_entry *oldtree = src[1];
2451         const struct cache_entry *newtree = src[2];
2452
2453         if (o->merge_size != 2)
2454                 return error("Cannot do a twoway merge of %d trees",
2455                              o->merge_size);
2456
2457         if (oldtree == o->df_conflict_entry)
2458                 oldtree = NULL;
2459         if (newtree == o->df_conflict_entry)
2460                 newtree = NULL;
2461
2462         if (current) {
2463                 if (current->ce_flags & CE_CONFLICTED) {
2464                         if (same(oldtree, newtree) || o->reset) {
2465                                 if (!newtree)
2466                                         return deleted_entry(current, current, o);
2467                                 else
2468                                         return merged_entry(newtree, current, o);
2469                         }
2470                         return reject_merge(current, o);
2471                 } else if ((!oldtree && !newtree) || /* 4 and 5 */
2472                          (!oldtree && newtree &&
2473                           same(current, newtree)) || /* 6 and 7 */
2474                          (oldtree && newtree &&
2475                           same(oldtree, newtree)) || /* 14 and 15 */
2476                          (oldtree && newtree &&
2477                           !same(oldtree, newtree) && /* 18 and 19 */
2478                           same(current, newtree))) {
2479                         return keep_entry(current, o);
2480                 } else if (oldtree && !newtree && same(current, oldtree)) {
2481                         /* 10 or 11 */
2482                         return deleted_entry(oldtree, current, o);
2483                 } else if (oldtree && newtree &&
2484                          same(current, oldtree) && !same(current, newtree)) {
2485                         /* 20 or 21 */
2486                         return merged_entry(newtree, current, o);
2487                 } else
2488                         return reject_merge(current, o);
2489         }
2490         else if (newtree) {
2491                 if (oldtree && !o->initial_checkout) {
2492                         /*
2493                          * deletion of the path was staged;
2494                          */
2495                         if (same(oldtree, newtree))
2496                                 return 1;
2497                         return reject_merge(oldtree, o);
2498                 }
2499                 return merged_entry(newtree, current, o);
2500         }
2501         return deleted_entry(oldtree, current, o);
2502 }
2503
2504 /*
2505  * Bind merge.
2506  *
2507  * Keep the index entries at stage0, collapse stage1 but make sure
2508  * stage0 does not have anything there.
2509  */
2510 int bind_merge(const struct cache_entry * const *src,
2511                struct unpack_trees_options *o)
2512 {
2513         const struct cache_entry *old = src[0];
2514         const struct cache_entry *a = src[1];
2515
2516         if (o->merge_size != 1)
2517                 return error("Cannot do a bind merge of %d trees",
2518                              o->merge_size);
2519         if (a && old)
2520                 return o->quiet ? -1 :
2521                         error(ERRORMSG(o, ERROR_BIND_OVERLAP),
2522                               super_prefixed(a->name),
2523                               super_prefixed(old->name));
2524         if (!a)
2525                 return keep_entry(old, o);
2526         else
2527                 return merged_entry(a, NULL, o);
2528 }
2529
2530 /*
2531  * One-way merge.
2532  *
2533  * The rule is:
2534  * - take the stat information from stage0, take the data from stage1
2535  */
2536 int oneway_merge(const struct cache_entry * const *src,
2537                  struct unpack_trees_options *o)
2538 {
2539         const struct cache_entry *old = src[0];
2540         const struct cache_entry *a = src[1];
2541
2542         if (o->merge_size != 1)
2543                 return error("Cannot do a oneway merge of %d trees",
2544                              o->merge_size);
2545
2546         if (!a || a == o->df_conflict_entry)
2547                 return deleted_entry(old, old, o);
2548
2549         if (old && same(old, a)) {
2550                 int update = 0;
2551                 if (o->reset && o->update && !ce_uptodate(old) && !ce_skip_worktree(old) &&
2552                         !(old->ce_flags & CE_FSMONITOR_VALID)) {
2553                         struct stat st;
2554                         if (lstat(old->name, &st) ||
2555                             ie_match_stat(o->src_index, old, &st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE))
2556                                 update |= CE_UPDATE;
2557                 }
2558                 if (o->update && S_ISGITLINK(old->ce_mode) &&
2559                     should_update_submodules() && !verify_uptodate(old, o))
2560                         update |= CE_UPDATE;
2561                 add_entry(o, old, update, CE_STAGEMASK);
2562                 return 0;
2563         }
2564         return merged_entry(a, old, o);
2565 }
2566
2567 /*
2568  * Merge worktree and untracked entries in a stash entry.
2569  *
2570  * Ignore all index entries. Collapse remaining trees but make sure that they
2571  * don't have any conflicting files.
2572  */
2573 int stash_worktree_untracked_merge(const struct cache_entry * const *src,
2574                                    struct unpack_trees_options *o)
2575 {
2576         const struct cache_entry *worktree = src[1];
2577         const struct cache_entry *untracked = src[2];
2578
2579         if (o->merge_size != 2)
2580                 BUG("invalid merge_size: %d", o->merge_size);
2581
2582         if (worktree && untracked)
2583                 return error(_("worktree and untracked commit have duplicate entries: %s"),
2584                              super_prefixed(worktree->name));
2585
2586         return merged_entry(worktree ? worktree : untracked, NULL, o);
2587 }