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