Merge branch 'ab/config-based-hooks-base' into seen
[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 #include "entry.h"
20 #include "parallel-checkout.h"
21
22 /*
23  * Error messages expected by scripts out of plumbing commands such as
24  * read-tree.  Non-scripted Porcelain is not required to use these messages
25  * and in fact are encouraged to reword them to better suit their particular
26  * situation better.  See how "git checkout" and "git merge" replaces
27  * them using setup_unpack_trees_porcelain(), for example.
28  */
29 static const char *unpack_plumbing_errors[NB_UNPACK_TREES_WARNING_TYPES] = {
30         /* ERROR_WOULD_OVERWRITE */
31         "Entry '%s' would be overwritten by merge. Cannot merge.",
32
33         /* ERROR_NOT_UPTODATE_FILE */
34         "Entry '%s' not uptodate. Cannot merge.",
35
36         /* ERROR_NOT_UPTODATE_DIR */
37         "Updating '%s' would lose untracked files in it",
38
39         /* ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN */
40         "Untracked working tree file '%s' would be overwritten by merge.",
41
42         /* ERROR_WOULD_LOSE_UNTRACKED_REMOVED */
43         "Untracked working tree file '%s' would be removed by merge.",
44
45         /* ERROR_BIND_OVERLAP */
46         "Entry '%s' overlaps with '%s'.  Cannot bind.",
47
48         /* ERROR_WOULD_LOSE_SUBMODULE */
49         "Submodule '%s' cannot checkout new HEAD.",
50
51         /* NB_UNPACK_TREES_ERROR_TYPES; just a meta value */
52         "",
53
54         /* WARNING_SPARSE_NOT_UPTODATE_FILE */
55         "Path '%s' not uptodate; will not remove from working tree.",
56
57         /* WARNING_SPARSE_UNMERGED_FILE */
58         "Path '%s' unmerged; will not remove from working tree.",
59
60         /* WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN */
61         "Path '%s' already present; will not overwrite with sparse update.",
62 };
63
64 #define ERRORMSG(o,type) \
65         ( ((o) && (o)->msgs[(type)]) \
66           ? ((o)->msgs[(type)])      \
67           : (unpack_plumbing_errors[(type)]) )
68
69 static const char *super_prefixed(const char *path)
70 {
71         /*
72          * It is necessary and sufficient to have two static buffers
73          * here, as the return value of this function is fed to
74          * error() using the unpack_*_errors[] templates we see above.
75          */
76         static struct strbuf buf[2] = {STRBUF_INIT, STRBUF_INIT};
77         static int super_prefix_len = -1;
78         static unsigned idx = ARRAY_SIZE(buf) - 1;
79
80         if (super_prefix_len < 0) {
81                 const char *super_prefix = get_super_prefix();
82                 if (!super_prefix) {
83                         super_prefix_len = 0;
84                 } else {
85                         int i;
86                         for (i = 0; i < ARRAY_SIZE(buf); i++)
87                                 strbuf_addstr(&buf[i], super_prefix);
88                         super_prefix_len = buf[0].len;
89                 }
90         }
91
92         if (!super_prefix_len)
93                 return path;
94
95         if (++idx >= ARRAY_SIZE(buf))
96                 idx = 0;
97
98         strbuf_setlen(&buf[idx], super_prefix_len);
99         strbuf_addstr(&buf[idx], path);
100
101         return buf[idx].buf;
102 }
103
104 void setup_unpack_trees_porcelain(struct unpack_trees_options *opts,
105                                   const char *cmd)
106 {
107         int i;
108         const char **msgs = opts->msgs;
109         const char *msg;
110
111         strvec_init(&opts->msgs_to_free);
112
113         if (!strcmp(cmd, "checkout"))
114                 msg = advice_commit_before_merge
115                       ? _("Your local changes to the following files would be overwritten by checkout:\n%%s"
116                           "Please commit your changes or stash them before you switch branches.")
117                       : _("Your local changes to the following files would be overwritten by checkout:\n%%s");
118         else if (!strcmp(cmd, "merge"))
119                 msg = advice_commit_before_merge
120                       ? _("Your local changes to the following files would be overwritten by merge:\n%%s"
121                           "Please commit your changes or stash them before you merge.")
122                       : _("Your local changes to the following files would be overwritten by merge:\n%%s");
123         else
124                 msg = advice_commit_before_merge
125                       ? _("Your local changes to the following files would be overwritten by %s:\n%%s"
126                           "Please commit your changes or stash them before you %s.")
127                       : _("Your local changes to the following files would be overwritten by %s:\n%%s");
128         msgs[ERROR_WOULD_OVERWRITE] = msgs[ERROR_NOT_UPTODATE_FILE] =
129                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
130
131         msgs[ERROR_NOT_UPTODATE_DIR] =
132                 _("Updating the following directories would lose untracked files in them:\n%s");
133
134         if (!strcmp(cmd, "checkout"))
135                 msg = advice_commit_before_merge
136                       ? _("The following untracked working tree files would be removed by checkout:\n%%s"
137                           "Please move or remove them before you switch branches.")
138                       : _("The following untracked working tree files would be removed by checkout:\n%%s");
139         else if (!strcmp(cmd, "merge"))
140                 msg = advice_commit_before_merge
141                       ? _("The following untracked working tree files would be removed by merge:\n%%s"
142                           "Please move or remove them before you merge.")
143                       : _("The following untracked working tree files would be removed by merge:\n%%s");
144         else
145                 msg = advice_commit_before_merge
146                       ? _("The following untracked working tree files would be removed by %s:\n%%s"
147                           "Please move or remove them before you %s.")
148                       : _("The following untracked working tree files would be removed by %s:\n%%s");
149         msgs[ERROR_WOULD_LOSE_UNTRACKED_REMOVED] =
150                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
151
152         if (!strcmp(cmd, "checkout"))
153                 msg = advice_commit_before_merge
154                       ? _("The following untracked working tree files would be overwritten by checkout:\n%%s"
155                           "Please move or remove them before you switch branches.")
156                       : _("The following untracked working tree files would be overwritten by checkout:\n%%s");
157         else if (!strcmp(cmd, "merge"))
158                 msg = advice_commit_before_merge
159                       ? _("The following untracked working tree files would be overwritten by merge:\n%%s"
160                           "Please move or remove them before you merge.")
161                       : _("The following untracked working tree files would be overwritten by merge:\n%%s");
162         else
163                 msg = advice_commit_before_merge
164                       ? _("The following untracked working tree files would be overwritten by %s:\n%%s"
165                           "Please move or remove them before you %s.")
166                       : _("The following untracked working tree files would be overwritten by %s:\n%%s");
167         msgs[ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN] =
168                 strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
169
170         /*
171          * Special case: ERROR_BIND_OVERLAP refers to a pair of paths, we
172          * cannot easily display it as a list.
173          */
174         msgs[ERROR_BIND_OVERLAP] = _("Entry '%s' overlaps with '%s'.  Cannot bind.");
175
176         msgs[ERROR_WOULD_LOSE_SUBMODULE] =
177                 _("Cannot update submodule:\n%s");
178
179         msgs[WARNING_SPARSE_NOT_UPTODATE_FILE] =
180                 _("The following paths are not up to date and were left despite sparse patterns:\n%s");
181         msgs[WARNING_SPARSE_UNMERGED_FILE] =
182                 _("The following paths are unmerged and were left despite sparse patterns:\n%s");
183         msgs[WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN] =
184                 _("The following paths were already present and thus not updated despite sparse patterns:\n%s");
185
186         opts->show_all_errors = 1;
187         /* rejected paths may not have a static buffer */
188         for (i = 0; i < ARRAY_SIZE(opts->unpack_rejects); i++)
189                 opts->unpack_rejects[i].strdup_strings = 1;
190 }
191
192 void clear_unpack_trees_porcelain(struct unpack_trees_options *opts)
193 {
194         strvec_clear(&opts->msgs_to_free);
195         memset(opts->msgs, 0, sizeof(opts->msgs));
196 }
197
198 static int do_add_entry(struct unpack_trees_options *o, struct cache_entry *ce,
199                          unsigned int set, unsigned int clear)
200 {
201         clear |= CE_HASHED;
202
203         if (set & CE_REMOVE)
204                 set |= CE_WT_REMOVE;
205
206         ce->ce_flags = (ce->ce_flags & ~clear) | set;
207         return add_index_entry(&o->result, ce,
208                                ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
209 }
210
211 static void add_entry(struct unpack_trees_options *o,
212                       const struct cache_entry *ce,
213                       unsigned int set, unsigned int clear)
214 {
215         do_add_entry(o, dup_cache_entry(ce, &o->result), set, clear);
216 }
217
218 /*
219  * add error messages on path <path>
220  * corresponding to the type <e> with the message <msg>
221  * indicating if it should be display in porcelain or not
222  */
223 static int add_rejected_path(struct unpack_trees_options *o,
224                              enum unpack_trees_error_types e,
225                              const char *path)
226 {
227         if (o->quiet)
228                 return -1;
229
230         if (!o->show_all_errors)
231                 return error(ERRORMSG(o, e), super_prefixed(path));
232
233         /*
234          * Otherwise, insert in a list for future display by
235          * display_(error|warning)_msgs()
236          */
237         string_list_append(&o->unpack_rejects[e], path);
238         return -1;
239 }
240
241 /*
242  * display all the error messages stored in a nice way
243  */
244 static void display_error_msgs(struct unpack_trees_options *o)
245 {
246         int e;
247         unsigned error_displayed = 0;
248         for (e = 0; e < NB_UNPACK_TREES_ERROR_TYPES; e++) {
249                 struct string_list *rejects = &o->unpack_rejects[e];
250
251                 if (rejects->nr > 0) {
252                         int i;
253                         struct strbuf path = STRBUF_INIT;
254
255                         error_displayed = 1;
256                         for (i = 0; i < rejects->nr; i++)
257                                 strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
258                         error(ERRORMSG(o, e), super_prefixed(path.buf));
259                         strbuf_release(&path);
260                 }
261                 string_list_clear(rejects, 0);
262         }
263         if (error_displayed)
264                 fprintf(stderr, _("Aborting\n"));
265 }
266
267 /*
268  * display all the warning messages stored in a nice way
269  */
270 static void display_warning_msgs(struct unpack_trees_options *o)
271 {
272         int e;
273         unsigned warning_displayed = 0;
274         for (e = NB_UNPACK_TREES_ERROR_TYPES + 1;
275              e < NB_UNPACK_TREES_WARNING_TYPES; e++) {
276                 struct string_list *rejects = &o->unpack_rejects[e];
277
278                 if (rejects->nr > 0) {
279                         int i;
280                         struct strbuf path = STRBUF_INIT;
281
282                         warning_displayed = 1;
283                         for (i = 0; i < rejects->nr; i++)
284                                 strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
285                         warning(ERRORMSG(o, e), super_prefixed(path.buf));
286                         strbuf_release(&path);
287                 }
288                 string_list_clear(rejects, 0);
289         }
290         if (warning_displayed)
291                 fprintf(stderr, _("After fixing the above paths, you may want to run `git sparse-checkout reapply`.\n"));
292 }
293 static int check_submodule_move_head(const struct cache_entry *ce,
294                                      const char *old_id,
295                                      const char *new_id,
296                                      struct unpack_trees_options *o)
297 {
298         unsigned flags = SUBMODULE_MOVE_HEAD_DRY_RUN;
299         const struct submodule *sub = submodule_from_ce(ce);
300
301         if (!sub)
302                 return 0;
303
304         if (o->reset)
305                 flags |= SUBMODULE_MOVE_HEAD_FORCE;
306
307         if (submodule_move_head(ce->name, old_id, new_id, flags))
308                 return add_rejected_path(o, ERROR_WOULD_LOSE_SUBMODULE, ce->name);
309         return 0;
310 }
311
312 /*
313  * Perform the loading of the repository's gitmodules file.  This function is
314  * used by 'check_update()' to perform loading of the gitmodules file in two
315  * different situations:
316  * (1) before removing entries from the working tree if the gitmodules file has
317  *     been marked for removal.  This situation is specified by 'state' == NULL.
318  * (2) before checking out entries to the working tree if the gitmodules file
319  *     has been marked for update.  This situation is specified by 'state' != NULL.
320  */
321 static void load_gitmodules_file(struct index_state *index,
322                                  struct checkout *state)
323 {
324         int pos = index_name_pos(index, GITMODULES_FILE, strlen(GITMODULES_FILE));
325
326         if (pos >= 0) {
327                 struct cache_entry *ce = index->cache[pos];
328                 if (!state && ce->ce_flags & CE_WT_REMOVE) {
329                         repo_read_gitmodules(the_repository, 0);
330                 } else if (state && (ce->ce_flags & CE_UPDATE)) {
331                         submodule_free(the_repository);
332                         checkout_entry(ce, state, NULL, NULL);
333                         repo_read_gitmodules(the_repository, 0);
334                 }
335         }
336 }
337
338 static struct progress *get_progress(struct unpack_trees_options *o,
339                                      struct index_state *index)
340 {
341         unsigned cnt = 0, total = 0;
342
343         if (!o->update || !o->verbose_update)
344                 return NULL;
345
346         for (; cnt < index->cache_nr; cnt++) {
347                 const struct cache_entry *ce = index->cache[cnt];
348                 if (ce->ce_flags & (CE_UPDATE | CE_WT_REMOVE))
349                         total++;
350         }
351
352         return start_delayed_progress(_("Updating files"), total);
353 }
354
355 static void setup_collided_checkout_detection(struct checkout *state,
356                                               struct index_state *index)
357 {
358         int i;
359
360         state->clone = 1;
361         for (i = 0; i < index->cache_nr; i++)
362                 index->cache[i]->ce_flags &= ~CE_MATCHED;
363 }
364
365 static void report_collided_checkout(struct index_state *index)
366 {
367         struct string_list list = STRING_LIST_INIT_NODUP;
368         int i;
369
370         for (i = 0; i < index->cache_nr; i++) {
371                 struct cache_entry *ce = index->cache[i];
372
373                 if (!(ce->ce_flags & CE_MATCHED))
374                         continue;
375
376                 string_list_append(&list, ce->name);
377                 ce->ce_flags &= ~CE_MATCHED;
378         }
379
380         list.cmp = fspathcmp;
381         string_list_sort(&list);
382
383         if (list.nr) {
384                 warning(_("the following paths have collided (e.g. case-sensitive paths\n"
385                           "on a case-insensitive filesystem) and only one from the same\n"
386                           "colliding group is in the working tree:\n"));
387
388                 for (i = 0; i < list.nr; i++)
389                         fprintf(stderr, "  '%s'\n", list.items[i].string);
390         }
391
392         string_list_clear(&list, 0);
393 }
394
395 static int check_updates(struct unpack_trees_options *o,
396                          struct index_state *index)
397 {
398         unsigned cnt = 0;
399         int errs = 0;
400         struct progress *progress;
401         struct checkout state = CHECKOUT_INIT;
402         int i, pc_workers, pc_threshold;
403
404         trace_performance_enter();
405         state.force = 1;
406         state.quiet = 1;
407         state.refresh_cache = 1;
408         state.istate = index;
409         clone_checkout_metadata(&state.meta, &o->meta, NULL);
410
411         if (!o->update || o->dry_run) {
412                 remove_marked_cache_entries(index, 0);
413                 trace_performance_leave("check_updates");
414                 return 0;
415         }
416
417         if (o->clone)
418                 setup_collided_checkout_detection(&state, index);
419
420         progress = get_progress(o, index);
421
422         /* Start with clean cache to avoid using any possibly outdated info. */
423         invalidate_lstat_cache();
424
425         git_attr_set_direction(GIT_ATTR_CHECKOUT);
426
427         if (should_update_submodules())
428                 load_gitmodules_file(index, NULL);
429
430         for (i = 0; i < index->cache_nr; i++) {
431                 const struct cache_entry *ce = index->cache[i];
432
433                 if (ce->ce_flags & CE_WT_REMOVE) {
434                         display_progress(progress, ++cnt);
435                         unlink_entry(ce);
436                 }
437         }
438
439         remove_marked_cache_entries(index, 0);
440         remove_scheduled_dirs();
441
442         if (should_update_submodules())
443                 load_gitmodules_file(index, &state);
444
445         if (has_promisor_remote()) {
446                 /*
447                  * Prefetch the objects that are to be checked out in the loop
448                  * below.
449                  */
450                 struct oid_array to_fetch = OID_ARRAY_INIT;
451                 for (i = 0; i < index->cache_nr; i++) {
452                         struct cache_entry *ce = index->cache[i];
453
454                         if (!(ce->ce_flags & CE_UPDATE) ||
455                             S_ISGITLINK(ce->ce_mode))
456                                 continue;
457                         if (!oid_object_info_extended(the_repository, &ce->oid,
458                                                       NULL,
459                                                       OBJECT_INFO_FOR_PREFETCH))
460                                 continue;
461                         oid_array_append(&to_fetch, &ce->oid);
462                 }
463                 promisor_remote_get_direct(the_repository,
464                                            to_fetch.oid, to_fetch.nr);
465                 oid_array_clear(&to_fetch);
466         }
467
468         get_parallel_checkout_configs(&pc_workers, &pc_threshold);
469
470         enable_delayed_checkout(&state);
471         if (pc_workers > 1)
472                 init_parallel_checkout();
473         for (i = 0; i < index->cache_nr; i++) {
474                 struct cache_entry *ce = index->cache[i];
475
476                 if (ce->ce_flags & CE_UPDATE) {
477                         size_t last_pc_queue_size = pc_queue_size();
478
479                         if (ce->ce_flags & CE_WT_REMOVE)
480                                 BUG("both update and delete flags are set on %s",
481                                     ce->name);
482                         ce->ce_flags &= ~CE_UPDATE;
483                         errs |= checkout_entry(ce, &state, NULL, NULL);
484
485                         if (last_pc_queue_size == pc_queue_size())
486                                 display_progress(progress, ++cnt);
487                 }
488         }
489         if (pc_workers > 1)
490                 errs |= run_parallel_checkout(&state, pc_workers, pc_threshold,
491                                               progress, &cnt);
492         stop_progress(&progress);
493         errs |= finish_delayed_checkout(&state, NULL);
494         git_attr_set_direction(GIT_ATTR_CHECKIN);
495
496         if (o->clone)
497                 report_collided_checkout(index);
498
499         trace_performance_leave("check_updates");
500         return errs != 0;
501 }
502
503 static int verify_uptodate_sparse(const struct cache_entry *ce,
504                                   struct unpack_trees_options *o);
505 static int verify_absent_sparse(const struct cache_entry *ce,
506                                 enum unpack_trees_error_types,
507                                 struct unpack_trees_options *o);
508
509 static int apply_sparse_checkout(struct index_state *istate,
510                                  struct cache_entry *ce,
511                                  struct unpack_trees_options *o)
512 {
513         int was_skip_worktree = ce_skip_worktree(ce);
514
515         if (ce->ce_flags & CE_NEW_SKIP_WORKTREE)
516                 ce->ce_flags |= CE_SKIP_WORKTREE;
517         else
518                 ce->ce_flags &= ~CE_SKIP_WORKTREE;
519         if (was_skip_worktree != ce_skip_worktree(ce)) {
520                 ce->ce_flags |= CE_UPDATE_IN_BASE;
521                 mark_fsmonitor_invalid(istate, ce);
522                 istate->cache_changed |= CE_ENTRY_CHANGED;
523         }
524
525         /*
526          * if (!was_skip_worktree && !ce_skip_worktree()) {
527          *      This is perfectly normal. Move on;
528          * }
529          */
530
531         /*
532          * Merge strategies may set CE_UPDATE|CE_REMOVE outside checkout
533          * area as a result of ce_skip_worktree() shortcuts in
534          * verify_absent() and verify_uptodate().
535          * Make sure they don't modify worktree if they are already
536          * outside checkout area
537          */
538         if (was_skip_worktree && ce_skip_worktree(ce)) {
539                 ce->ce_flags &= ~CE_UPDATE;
540
541                 /*
542                  * By default, when CE_REMOVE is on, CE_WT_REMOVE is also
543                  * on to get that file removed from both index and worktree.
544                  * If that file is already outside worktree area, don't
545                  * bother remove it.
546                  */
547                 if (ce->ce_flags & CE_REMOVE)
548                         ce->ce_flags &= ~CE_WT_REMOVE;
549         }
550
551         if (!was_skip_worktree && ce_skip_worktree(ce)) {
552                 /*
553                  * If CE_UPDATE is set, verify_uptodate() must be called already
554                  * also stat info may have lost after merged_entry() so calling
555                  * verify_uptodate() again may fail
556                  */
557                 if (!(ce->ce_flags & CE_UPDATE) &&
558                     verify_uptodate_sparse(ce, o)) {
559                         ce->ce_flags &= ~CE_SKIP_WORKTREE;
560                         return -1;
561                 }
562                 ce->ce_flags |= CE_WT_REMOVE;
563                 ce->ce_flags &= ~CE_UPDATE;
564         }
565         if (was_skip_worktree && !ce_skip_worktree(ce)) {
566                 if (verify_absent_sparse(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
567                         return -1;
568                 ce->ce_flags |= CE_UPDATE;
569         }
570         return 0;
571 }
572
573 static int warn_conflicted_path(struct index_state *istate,
574                                 int i,
575                                 struct unpack_trees_options *o)
576 {
577         char *conflicting_path = istate->cache[i]->name;
578         int count = 0;
579
580         add_rejected_path(o, WARNING_SPARSE_UNMERGED_FILE, conflicting_path);
581
582         /* Find out how many higher stage entries are at same path */
583         while ((++count) + i < istate->cache_nr &&
584                !strcmp(conflicting_path, istate->cache[count + i]->name))
585                 ; /* do nothing */
586
587         return count;
588 }
589
590 static inline int call_unpack_fn(const struct cache_entry * const *src,
591                                  struct unpack_trees_options *o)
592 {
593         int ret = o->fn(src, o);
594         if (ret > 0)
595                 ret = 0;
596         return ret;
597 }
598
599 static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
600 {
601         ce->ce_flags |= CE_UNPACKED;
602
603         /*
604          * If this is a sparse directory, don't advance cache_bottom.
605          * That will be advanced later using the cache-tree data.
606          */
607         if (S_ISSPARSEDIR(ce->ce_mode))
608                 return;
609
610         if (o->cache_bottom < o->src_index->cache_nr &&
611             o->src_index->cache[o->cache_bottom] == ce) {
612                 int bottom = o->cache_bottom;
613                 while (bottom < o->src_index->cache_nr &&
614                        o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
615                         bottom++;
616                 o->cache_bottom = bottom;
617         }
618 }
619
620 static void mark_all_ce_unused(struct index_state *index)
621 {
622         int i;
623         for (i = 0; i < index->cache_nr; i++)
624                 index->cache[i]->ce_flags &= ~(CE_UNPACKED | CE_ADDED | CE_NEW_SKIP_WORKTREE);
625 }
626
627 static int locate_in_src_index(const struct cache_entry *ce,
628                                struct unpack_trees_options *o)
629 {
630         struct index_state *index = o->src_index;
631         int len = ce_namelen(ce);
632         int pos = index_name_pos(index, ce->name, len);
633         if (pos < 0)
634                 pos = -1 - pos;
635         return pos;
636 }
637
638 /*
639  * We call unpack_index_entry() with an unmerged cache entry
640  * only in diff-index, and it wants a single callback.  Skip
641  * the other unmerged entry with the same name.
642  */
643 static void mark_ce_used_same_name(struct cache_entry *ce,
644                                    struct unpack_trees_options *o)
645 {
646         struct index_state *index = o->src_index;
647         int len = ce_namelen(ce);
648         int pos;
649
650         for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
651                 struct cache_entry *next = index->cache[pos];
652                 if (len != ce_namelen(next) ||
653                     memcmp(ce->name, next->name, len))
654                         break;
655                 mark_ce_used(next, o);
656         }
657 }
658
659 static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
660 {
661         const struct index_state *index = o->src_index;
662         int pos = o->cache_bottom;
663
664         while (pos < index->cache_nr) {
665                 struct cache_entry *ce = index->cache[pos];
666                 if (!(ce->ce_flags & CE_UNPACKED))
667                         return ce;
668                 pos++;
669         }
670         return NULL;
671 }
672
673 static void add_same_unmerged(const struct cache_entry *ce,
674                               struct unpack_trees_options *o)
675 {
676         struct index_state *index = o->src_index;
677         int len = ce_namelen(ce);
678         int pos = index_name_pos(index, ce->name, len);
679
680         if (0 <= pos)
681                 die("programming error in a caller of mark_ce_used_same_name");
682         for (pos = -pos - 1; pos < index->cache_nr; pos++) {
683                 struct cache_entry *next = index->cache[pos];
684                 if (len != ce_namelen(next) ||
685                     memcmp(ce->name, next->name, len))
686                         break;
687                 add_entry(o, next, 0, 0);
688                 mark_ce_used(next, o);
689         }
690 }
691
692 static int unpack_index_entry(struct cache_entry *ce,
693                               struct unpack_trees_options *o)
694 {
695         const struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
696         int ret;
697
698         src[0] = ce;
699
700         mark_ce_used(ce, o);
701         if (ce_stage(ce)) {
702                 if (o->skip_unmerged) {
703                         add_entry(o, ce, 0, 0);
704                         return 0;
705                 }
706         }
707         ret = call_unpack_fn(src, o);
708         if (ce_stage(ce))
709                 mark_ce_used_same_name(ce, o);
710         return ret;
711 }
712
713 static int find_cache_pos(struct traverse_info *, const char *p, size_t len);
714
715 static void restore_cache_bottom(struct traverse_info *info, int bottom)
716 {
717         struct unpack_trees_options *o = info->data;
718
719         if (o->diff_index_cached)
720                 return;
721         o->cache_bottom = bottom;
722 }
723
724 static int switch_cache_bottom(struct traverse_info *info)
725 {
726         struct unpack_trees_options *o = info->data;
727         int ret, pos;
728
729         if (o->diff_index_cached)
730                 return 0;
731         ret = o->cache_bottom;
732         pos = find_cache_pos(info->prev, info->name, info->namelen);
733
734         if (pos < -1)
735                 o->cache_bottom = -2 - pos;
736         else if (pos < 0)
737                 o->cache_bottom = o->src_index->cache_nr;
738         return ret;
739 }
740
741 static inline int are_same_oid(struct name_entry *name_j, struct name_entry *name_k)
742 {
743         return !is_null_oid(&name_j->oid) && !is_null_oid(&name_k->oid) && oideq(&name_j->oid, &name_k->oid);
744 }
745
746 static int all_trees_same_as_cache_tree(int n, unsigned long dirmask,
747                                         struct name_entry *names,
748                                         struct traverse_info *info)
749 {
750         struct unpack_trees_options *o = info->data;
751         int i;
752
753         if (!o->merge || dirmask != ((1 << n) - 1))
754                 return 0;
755
756         for (i = 1; i < n; i++)
757                 if (!are_same_oid(names, names + i))
758                         return 0;
759
760         return cache_tree_matches_traversal(o->src_index->cache_tree, names, info);
761 }
762
763 static int index_pos_by_traverse_info(struct name_entry *names,
764                                       struct traverse_info *info)
765 {
766         struct unpack_trees_options *o = info->data;
767         struct strbuf name = STRBUF_INIT;
768         int pos;
769
770         strbuf_make_traverse_path(&name, info, names->path, names->pathlen);
771         strbuf_addch(&name, '/');
772         pos = index_name_pos(o->src_index, name.buf, name.len);
773         if (pos >= 0) {
774                 if (!o->src_index->sparse_index ||
775                     !(o->src_index->cache[pos]->ce_flags & CE_SKIP_WORKTREE))
776                         BUG("This is a directory and should not exist in index");
777         } else {
778                 pos = -pos - 1;
779         }
780         if (pos >= o->src_index->cache_nr ||
781             !starts_with(o->src_index->cache[pos]->name, name.buf) ||
782             (pos > 0 && starts_with(o->src_index->cache[pos-1]->name, name.buf)))
783                 BUG("pos %d doesn't point to the first entry of %s in index",
784                     pos, name.buf);
785         strbuf_release(&name);
786         return pos;
787 }
788
789 /*
790  * Fast path if we detect that all trees are the same as cache-tree at this
791  * path. We'll walk these trees in an iterative loop using cache-tree/index
792  * instead of ODB since we already know what these trees contain.
793  */
794 static int traverse_by_cache_tree(int pos, int nr_entries, int nr_names,
795                                   struct traverse_info *info)
796 {
797         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
798         struct unpack_trees_options *o = info->data;
799         struct cache_entry *tree_ce = NULL;
800         int ce_len = 0;
801         int i, d;
802
803         if (!o->merge)
804                 BUG("We need cache-tree to do this optimization");
805
806         /*
807          * Do what unpack_callback() and unpack_nondirectories() normally
808          * do. But we walk all paths in an iterative loop instead.
809          *
810          * D/F conflicts and higher stage entries are not a concern
811          * because cache-tree would be invalidated and we would never
812          * get here in the first place.
813          */
814         for (i = 0; i < nr_entries; i++) {
815                 int new_ce_len, len, rc;
816
817                 src[0] = o->src_index->cache[pos + i];
818
819                 if (S_ISSPARSEDIR(src[0]->ce_mode))
820                         continue;
821
822                 len = ce_namelen(src[0]);
823                 new_ce_len = cache_entry_size(len);
824
825                 if (new_ce_len > ce_len) {
826                         new_ce_len <<= 1;
827                         tree_ce = xrealloc(tree_ce, new_ce_len);
828                         memset(tree_ce, 0, new_ce_len);
829                         ce_len = new_ce_len;
830
831                         tree_ce->ce_flags = create_ce_flags(0);
832
833                         for (d = 1; d <= nr_names; d++)
834                                 src[d] = tree_ce;
835                 }
836
837                 tree_ce->ce_mode = src[0]->ce_mode;
838                 tree_ce->ce_namelen = len;
839                 oidcpy(&tree_ce->oid, &src[0]->oid);
840                 memcpy(tree_ce->name, src[0]->name, len + 1);
841
842                 rc = call_unpack_fn((const struct cache_entry * const *)src, o);
843                 if (rc < 0) {
844                         free(tree_ce);
845                         return rc;
846                 }
847
848                 mark_ce_used(src[0], o);
849         }
850         free(tree_ce);
851         if (o->debug_unpack)
852                 printf("Unpacked %d entries from %s to %s using cache-tree\n",
853                        nr_entries,
854                        o->src_index->cache[pos]->name,
855                        o->src_index->cache[pos + nr_entries - 1]->name);
856         return 0;
857 }
858
859 static int traverse_trees_recursive(int n, unsigned long dirmask,
860                                     unsigned long df_conflicts,
861                                     struct name_entry *names,
862                                     struct traverse_info *info)
863 {
864         struct unpack_trees_options *o = info->data;
865         int i, ret, bottom;
866         int nr_buf = 0;
867         struct tree_desc t[MAX_UNPACK_TREES];
868         void *buf[MAX_UNPACK_TREES];
869         struct traverse_info newinfo;
870         struct name_entry *p;
871         int nr_entries;
872
873         nr_entries = all_trees_same_as_cache_tree(n, dirmask, names, info);
874         if (nr_entries > 0) {
875                 int pos = index_pos_by_traverse_info(names, info);
876
877                 if (!o->merge || df_conflicts)
878                         BUG("Wrong condition to get here buddy");
879
880                 /*
881                  * All entries up to 'pos' must have been processed
882                  * (i.e. marked CE_UNPACKED) at this point. But to be safe,
883                  * save and restore cache_bottom anyway to not miss
884                  * unprocessed entries before 'pos'.
885                  */
886                 bottom = o->cache_bottom;
887                 ret = traverse_by_cache_tree(pos, nr_entries, n, info);
888                 o->cache_bottom = bottom;
889                 return ret;
890         }
891
892         p = names;
893         while (!p->mode)
894                 p++;
895
896         newinfo = *info;
897         newinfo.prev = info;
898         newinfo.pathspec = info->pathspec;
899         newinfo.name = p->path;
900         newinfo.namelen = p->pathlen;
901         newinfo.mode = p->mode;
902         newinfo.pathlen = st_add3(newinfo.pathlen, tree_entry_len(p), 1);
903         newinfo.df_conflicts |= df_conflicts;
904
905         /*
906          * Fetch the tree from the ODB for each peer directory in the
907          * n commits.
908          *
909          * For 2- and 3-way traversals, we try to avoid hitting the
910          * ODB twice for the same OID.  This should yield a nice speed
911          * up in checkouts and merges when the commits are similar.
912          *
913          * We don't bother doing the full O(n^2) search for larger n,
914          * because wider traversals don't happen that often and we
915          * avoid the search setup.
916          *
917          * When 2 peer OIDs are the same, we just copy the tree
918          * descriptor data.  This implicitly borrows the buffer
919          * data from the earlier cell.
920          */
921         for (i = 0; i < n; i++, dirmask >>= 1) {
922                 if (i > 0 && are_same_oid(&names[i], &names[i - 1]))
923                         t[i] = t[i - 1];
924                 else if (i > 1 && are_same_oid(&names[i], &names[i - 2]))
925                         t[i] = t[i - 2];
926                 else {
927                         const struct object_id *oid = NULL;
928                         if (dirmask & 1)
929                                 oid = &names[i].oid;
930                         buf[nr_buf++] = fill_tree_descriptor(the_repository, t + i, oid);
931                 }
932         }
933
934         bottom = switch_cache_bottom(&newinfo);
935         ret = traverse_trees(o->src_index, n, t, &newinfo);
936         restore_cache_bottom(&newinfo, bottom);
937
938         for (i = 0; i < nr_buf; i++)
939                 free(buf[i]);
940
941         return ret;
942 }
943
944 /*
945  * Compare the traverse-path to the cache entry without actually
946  * having to generate the textual representation of the traverse
947  * path.
948  *
949  * NOTE! This *only* compares up to the size of the traverse path
950  * itself - the caller needs to do the final check for the cache
951  * entry having more data at the end!
952  */
953 static int do_compare_entry_piecewise(const struct cache_entry *ce,
954                                       const struct traverse_info *info,
955                                       const char *name, size_t namelen,
956                                       unsigned mode)
957 {
958         int pathlen, ce_len;
959         const char *ce_name;
960
961         if (info->prev) {
962                 int cmp = do_compare_entry_piecewise(ce, info->prev,
963                                                      info->name, info->namelen,
964                                                      info->mode);
965                 if (cmp)
966                         return cmp;
967         }
968         pathlen = info->pathlen;
969         ce_len = ce_namelen(ce);
970
971         /* If ce_len < pathlen then we must have previously hit "name == directory" entry */
972         if (ce_len < pathlen)
973                 return -1;
974
975         ce_len -= pathlen;
976         ce_name = ce->name + pathlen;
977
978         return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
979 }
980
981 static int do_compare_entry(const struct cache_entry *ce,
982                             const struct traverse_info *info,
983                             const char *name, size_t namelen,
984                             unsigned mode)
985 {
986         int pathlen, ce_len;
987         const char *ce_name;
988         int cmp;
989         unsigned ce_mode;
990
991         /*
992          * If we have not precomputed the traverse path, it is quicker
993          * to avoid doing so.  But if we have precomputed it,
994          * it is quicker to use the precomputed version.
995          */
996         if (!info->traverse_path)
997                 return do_compare_entry_piecewise(ce, info, name, namelen, mode);
998
999         cmp = strncmp(ce->name, info->traverse_path, info->pathlen);
1000         if (cmp)
1001                 return cmp;
1002
1003         pathlen = info->pathlen;
1004         ce_len = ce_namelen(ce);
1005
1006         if (ce_len < pathlen)
1007                 return -1;
1008
1009         ce_len -= pathlen;
1010         ce_name = ce->name + pathlen;
1011
1012         ce_mode = S_ISSPARSEDIR(ce->ce_mode) ? S_IFDIR : S_IFREG;
1013         return df_name_compare(ce_name, ce_len, ce_mode, name, namelen, mode);
1014 }
1015
1016 static int compare_entry(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
1017 {
1018         int cmp = do_compare_entry(ce, info, n->path, n->pathlen, n->mode);
1019         if (cmp)
1020                 return cmp;
1021
1022         /*
1023          * At this point, we know that we have a prefix match. If ce
1024          * is a sparse directory, then allow an exact match. This only
1025          * works when the input name is a directory, since ce->name
1026          * ends in a directory separator.
1027          */
1028         if (S_ISSPARSEDIR(ce->ce_mode))
1029                 return 0;
1030
1031         /*
1032          * Even if the beginning compared identically, the ce should
1033          * compare as bigger than a directory leading up to it!
1034          */
1035         return ce_namelen(ce) > traverse_path_len(info, tree_entry_len(n));
1036 }
1037
1038 static int ce_in_traverse_path(const struct cache_entry *ce,
1039                                const struct traverse_info *info)
1040 {
1041         if (!info->prev)
1042                 return 1;
1043         if (do_compare_entry(ce, info->prev,
1044                              info->name, info->namelen, info->mode))
1045                 return 0;
1046         /*
1047          * If ce (blob) is the same name as the path (which is a tree
1048          * we will be descending into), it won't be inside it.
1049          */
1050         return (info->pathlen < ce_namelen(ce));
1051 }
1052
1053 static struct cache_entry *create_ce_entry(const struct traverse_info *info,
1054         const struct name_entry *n,
1055         int stage,
1056         struct index_state *istate,
1057         int is_transient)
1058 {
1059         size_t len = traverse_path_len(info, tree_entry_len(n));
1060         struct cache_entry *ce =
1061                 is_transient ?
1062                 make_empty_transient_cache_entry(len, NULL) :
1063                 make_empty_cache_entry(istate, len);
1064
1065         ce->ce_mode = create_ce_mode(n->mode);
1066         ce->ce_flags = create_ce_flags(stage);
1067         ce->ce_namelen = len;
1068         oidcpy(&ce->oid, &n->oid);
1069         /* len+1 because the cache_entry allocates space for NUL */
1070         make_traverse_path(ce->name, len + 1, info, n->path, n->pathlen);
1071
1072         return ce;
1073 }
1074
1075 /*
1076  * Note that traverse_by_cache_tree() duplicates some logic in this function
1077  * without actually calling it. If you change the logic here you may need to
1078  * check and change there as well.
1079  */
1080 static int unpack_nondirectories(int n, unsigned long mask,
1081                                  unsigned long dirmask,
1082                                  struct cache_entry **src,
1083                                  const struct name_entry *names,
1084                                  const struct traverse_info *info)
1085 {
1086         int i;
1087         struct unpack_trees_options *o = info->data;
1088         unsigned long conflicts = info->df_conflicts | dirmask;
1089
1090         /* Do we have *only* directories? Nothing to do */
1091         if (mask == dirmask && !src[0])
1092                 return 0;
1093
1094         if (src[0] && S_ISSPARSEDIR(src[0]->ce_mode))
1095                 return 0;
1096
1097         /*
1098          * Ok, we've filled in up to any potential index entry in src[0],
1099          * now do the rest.
1100          */
1101         for (i = 0; i < n; i++) {
1102                 int stage;
1103                 unsigned int bit = 1ul << i;
1104                 if (conflicts & bit) {
1105                         src[i + o->merge] = o->df_conflict_entry;
1106                         continue;
1107                 }
1108                 if (!(mask & bit))
1109                         continue;
1110                 if (!o->merge)
1111                         stage = 0;
1112                 else if (i + 1 < o->head_idx)
1113                         stage = 1;
1114                 else if (i + 1 > o->head_idx)
1115                         stage = 3;
1116                 else
1117                         stage = 2;
1118
1119                 /*
1120                  * If the merge bit is set, then the cache entries are
1121                  * discarded in the following block.  In this case,
1122                  * construct "transient" cache_entries, as they are
1123                  * not stored in the index.  otherwise construct the
1124                  * cache entry from the index aware logic.
1125                  */
1126                 src[i + o->merge] = create_ce_entry(info, names + i, stage, &o->result, o->merge);
1127         }
1128
1129         if (o->merge) {
1130                 int rc = call_unpack_fn((const struct cache_entry * const *)src,
1131                                         o);
1132                 for (i = 0; i < n; i++) {
1133                         struct cache_entry *ce = src[i + o->merge];
1134                         if (ce != o->df_conflict_entry)
1135                                 discard_cache_entry(ce);
1136                 }
1137                 return rc;
1138         }
1139
1140         for (i = 0; i < n; i++)
1141                 if (src[i] && src[i] != o->df_conflict_entry)
1142                         if (do_add_entry(o, src[i], 0, 0))
1143                                 return -1;
1144
1145         return 0;
1146 }
1147
1148 static int unpack_failed(struct unpack_trees_options *o, const char *message)
1149 {
1150         discard_index(&o->result);
1151         if (!o->quiet && !o->exiting_early) {
1152                 if (message)
1153                         return error("%s", message);
1154                 return -1;
1155         }
1156         return -1;
1157 }
1158
1159 /*
1160  * The tree traversal is looking at name p.  If we have a matching entry,
1161  * return it.  If name p is a directory in the index, do not return
1162  * anything, as we will want to match it when the traversal descends into
1163  * the directory.
1164  */
1165 static int find_cache_pos(struct traverse_info *info,
1166                           const char *p, size_t p_len)
1167 {
1168         int pos;
1169         struct unpack_trees_options *o = info->data;
1170         struct index_state *index = o->src_index;
1171         int pfxlen = info->pathlen;
1172
1173         for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
1174                 const struct cache_entry *ce = index->cache[pos];
1175                 const char *ce_name, *ce_slash;
1176                 int cmp, ce_len;
1177
1178                 if (ce->ce_flags & CE_UNPACKED) {
1179                         /*
1180                          * cache_bottom entry is already unpacked, so
1181                          * we can never match it; don't check it
1182                          * again.
1183                          */
1184                         if (pos == o->cache_bottom)
1185                                 ++o->cache_bottom;
1186                         continue;
1187                 }
1188                 if (!ce_in_traverse_path(ce, info)) {
1189                         /*
1190                          * Check if we can skip future cache checks
1191                          * (because we're already past all possible
1192                          * entries in the traverse path).
1193                          */
1194                         if (info->traverse_path) {
1195                                 if (strncmp(ce->name, info->traverse_path,
1196                                             info->pathlen) > 0)
1197                                         break;
1198                         }
1199                         continue;
1200                 }
1201                 ce_name = ce->name + pfxlen;
1202                 ce_slash = strchr(ce_name, '/');
1203                 if (ce_slash)
1204                         ce_len = ce_slash - ce_name;
1205                 else
1206                         ce_len = ce_namelen(ce) - pfxlen;
1207                 cmp = name_compare(p, p_len, ce_name, ce_len);
1208                 /*
1209                  * Exact match; if we have a directory we need to
1210                  * delay returning it.
1211                  */
1212                 if (!cmp)
1213                         return ce_slash ? -2 - pos : pos;
1214                 if (0 < cmp)
1215                         continue; /* keep looking */
1216                 /*
1217                  * ce_name sorts after p->path; could it be that we
1218                  * have files under p->path directory in the index?
1219                  * E.g.  ce_name == "t-i", and p->path == "t"; we may
1220                  * have "t/a" in the index.
1221                  */
1222                 if (p_len < ce_len && !memcmp(ce_name, p, p_len) &&
1223                     ce_name[p_len] < '/')
1224                         continue; /* keep looking */
1225                 break;
1226         }
1227         return -1;
1228 }
1229
1230 static struct cache_entry *find_cache_entry(struct traverse_info *info,
1231                                             const struct name_entry *p)
1232 {
1233         int pos = find_cache_pos(info, p->path, p->pathlen);
1234         struct unpack_trees_options *o = info->data;
1235
1236         if (0 <= pos)
1237                 return o->src_index->cache[pos];
1238         else
1239                 return NULL;
1240 }
1241
1242 static void debug_path(struct traverse_info *info)
1243 {
1244         if (info->prev) {
1245                 debug_path(info->prev);
1246                 if (*info->prev->name)
1247                         putchar('/');
1248         }
1249         printf("%s", info->name);
1250 }
1251
1252 static void debug_name_entry(int i, struct name_entry *n)
1253 {
1254         printf("ent#%d %06o %s\n", i,
1255                n->path ? n->mode : 0,
1256                n->path ? n->path : "(missing)");
1257 }
1258
1259 static void debug_unpack_callback(int n,
1260                                   unsigned long mask,
1261                                   unsigned long dirmask,
1262                                   struct name_entry *names,
1263                                   struct traverse_info *info)
1264 {
1265         int i;
1266         printf("* unpack mask %lu, dirmask %lu, cnt %d ",
1267                mask, dirmask, n);
1268         debug_path(info);
1269         putchar('\n');
1270         for (i = 0; i < n; i++)
1271                 debug_name_entry(i, names + i);
1272 }
1273
1274 /*
1275  * Note that traverse_by_cache_tree() duplicates some logic in this function
1276  * without actually calling it. If you change the logic here you may need to
1277  * check and change there as well.
1278  */
1279 static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
1280 {
1281         struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
1282         struct unpack_trees_options *o = info->data;
1283         const struct name_entry *p = names;
1284
1285         /* Find first entry with a real name (we could use "mask" too) */
1286         while (!p->mode)
1287                 p++;
1288
1289         if (o->debug_unpack)
1290                 debug_unpack_callback(n, mask, dirmask, names, info);
1291
1292         /* Are we supposed to look at the index too? */
1293         if (o->merge) {
1294                 while (1) {
1295                         int cmp;
1296                         struct cache_entry *ce;
1297
1298                         if (o->diff_index_cached)
1299                                 ce = next_cache_entry(o);
1300                         else
1301                                 ce = find_cache_entry(info, p);
1302
1303                         if (!ce)
1304                                 break;
1305                         cmp = compare_entry(ce, info, p);
1306                         if (cmp < 0) {
1307                                 if (unpack_index_entry(ce, o) < 0)
1308                                         return unpack_failed(o, NULL);
1309                                 continue;
1310                         }
1311                         if (!cmp) {
1312                                 if (ce_stage(ce)) {
1313                                         /*
1314                                          * If we skip unmerged index
1315                                          * entries, we'll skip this
1316                                          * entry *and* the tree
1317                                          * entries associated with it!
1318                                          */
1319                                         if (o->skip_unmerged) {
1320                                                 add_same_unmerged(ce, o);
1321                                                 return mask;
1322                                         }
1323                                 }
1324                                 src[0] = ce;
1325                         }
1326                         break;
1327                 }
1328         }
1329
1330         if (unpack_nondirectories(n, mask, dirmask, src, names, info) < 0)
1331                 return -1;
1332
1333         if (o->merge && src[0]) {
1334                 if (ce_stage(src[0]))
1335                         mark_ce_used_same_name(src[0], o);
1336                 else
1337                         mark_ce_used(src[0], o);
1338         }
1339
1340         /* Now handle any directories.. */
1341         if (dirmask) {
1342                 /* special case: "diff-index --cached" looking at a tree */
1343                 if (o->diff_index_cached &&
1344                     n == 1 && dirmask == 1 && S_ISDIR(names->mode)) {
1345                         int matches;
1346                         matches = cache_tree_matches_traversal(o->src_index->cache_tree,
1347                                                                names, info);
1348                         /*
1349                          * Everything under the name matches; skip the
1350                          * entire hierarchy.  diff_index_cached codepath
1351                          * special cases D/F conflicts in such a way that
1352                          * it does not do any look-ahead, so this is safe.
1353                          */
1354                         if (matches) {
1355                                 o->cache_bottom += matches;
1356                                 return mask;
1357                         }
1358                 }
1359
1360                 if (traverse_trees_recursive(n, dirmask, mask & ~dirmask,
1361                                              names, info) < 0)
1362                         return -1;
1363                 return mask;
1364         }
1365
1366         return mask;
1367 }
1368
1369 static int clear_ce_flags_1(struct index_state *istate,
1370                             struct cache_entry **cache, int nr,
1371                             struct strbuf *prefix,
1372                             int select_mask, int clear_mask,
1373                             struct pattern_list *pl,
1374                             enum pattern_match_result default_match,
1375                             int progress_nr);
1376
1377 /* Whole directory matching */
1378 static int clear_ce_flags_dir(struct index_state *istate,
1379                               struct cache_entry **cache, int nr,
1380                               struct strbuf *prefix,
1381                               char *basename,
1382                               int select_mask, int clear_mask,
1383                               struct pattern_list *pl,
1384                               enum pattern_match_result default_match,
1385                               int progress_nr)
1386 {
1387         struct cache_entry **cache_end;
1388         int dtype = DT_DIR;
1389         int rc;
1390         enum pattern_match_result ret, orig_ret;
1391         orig_ret = path_matches_pattern_list(prefix->buf, prefix->len,
1392                                              basename, &dtype, pl, istate);
1393
1394         strbuf_addch(prefix, '/');
1395
1396         /* If undecided, use matching result of parent dir in defval */
1397         if (orig_ret == UNDECIDED)
1398                 ret = default_match;
1399         else
1400                 ret = orig_ret;
1401
1402         for (cache_end = cache; cache_end != cache + nr; cache_end++) {
1403                 struct cache_entry *ce = *cache_end;
1404                 if (strncmp(ce->name, prefix->buf, prefix->len))
1405                         break;
1406         }
1407
1408         if (pl->use_cone_patterns && orig_ret == MATCHED_RECURSIVE) {
1409                 struct cache_entry **ce = cache;
1410                 rc = cache_end - cache;
1411
1412                 while (ce < cache_end) {
1413                         (*ce)->ce_flags &= ~clear_mask;
1414                         ce++;
1415                 }
1416         } else if (pl->use_cone_patterns && orig_ret == NOT_MATCHED) {
1417                 rc = cache_end - cache;
1418         } else {
1419                 rc = clear_ce_flags_1(istate, cache, cache_end - cache,
1420                                       prefix,
1421                                       select_mask, clear_mask,
1422                                       pl, ret,
1423                                       progress_nr);
1424         }
1425
1426         strbuf_setlen(prefix, prefix->len - 1);
1427         return rc;
1428 }
1429
1430 /*
1431  * Traverse the index, find every entry that matches according to
1432  * o->pl. Do "ce_flags &= ~clear_mask" on those entries. Return the
1433  * number of traversed entries.
1434  *
1435  * If select_mask is non-zero, only entries whose ce_flags has on of
1436  * those bits enabled are traversed.
1437  *
1438  * cache        : pointer to an index entry
1439  * prefix_len   : an offset to its path
1440  *
1441  * The current path ("prefix") including the trailing '/' is
1442  *   cache[0]->name[0..(prefix_len-1)]
1443  * Top level path has prefix_len zero.
1444  */
1445 static int clear_ce_flags_1(struct index_state *istate,
1446                             struct cache_entry **cache, int nr,
1447                             struct strbuf *prefix,
1448                             int select_mask, int clear_mask,
1449                             struct pattern_list *pl,
1450                             enum pattern_match_result default_match,
1451                             int progress_nr)
1452 {
1453         struct cache_entry **cache_end = nr ? cache + nr : cache;
1454
1455         /*
1456          * Process all entries that have the given prefix and meet
1457          * select_mask condition
1458          */
1459         while(cache != cache_end) {
1460                 struct cache_entry *ce = *cache;
1461                 const char *name, *slash;
1462                 int len, dtype;
1463                 enum pattern_match_result ret;
1464
1465                 display_progress(istate->progress, progress_nr);
1466
1467                 if (select_mask && !(ce->ce_flags & select_mask)) {
1468                         cache++;
1469                         progress_nr++;
1470                         continue;
1471                 }
1472
1473                 if (prefix->len && strncmp(ce->name, prefix->buf, prefix->len))
1474                         break;
1475
1476                 name = ce->name + prefix->len;
1477                 slash = strchr(name, '/');
1478
1479                 /* If it's a directory, try whole directory match first */
1480                 if (slash) {
1481                         int processed;
1482
1483                         len = slash - name;
1484                         strbuf_add(prefix, name, len);
1485
1486                         processed = clear_ce_flags_dir(istate, cache, cache_end - cache,
1487                                                        prefix,
1488                                                        prefix->buf + prefix->len - len,
1489                                                        select_mask, clear_mask,
1490                                                        pl, default_match,
1491                                                        progress_nr);
1492
1493                         /* clear_c_f_dir eats a whole dir already? */
1494                         if (processed) {
1495                                 cache += processed;
1496                                 progress_nr += processed;
1497                                 strbuf_setlen(prefix, prefix->len - len);
1498                                 continue;
1499                         }
1500
1501                         strbuf_addch(prefix, '/');
1502                         processed = clear_ce_flags_1(istate, cache, cache_end - cache,
1503                                                      prefix,
1504                                                      select_mask, clear_mask, pl,
1505                                                      default_match, progress_nr);
1506
1507                         cache += processed;
1508                         progress_nr += processed;
1509
1510                         strbuf_setlen(prefix, prefix->len - len - 1);
1511                         continue;
1512                 }
1513
1514                 /* Non-directory */
1515                 dtype = ce_to_dtype(ce);
1516                 ret = path_matches_pattern_list(ce->name,
1517                                                 ce_namelen(ce),
1518                                                 name, &dtype, pl, istate);
1519                 if (ret == UNDECIDED)
1520                         ret = default_match;
1521                 if (ret == MATCHED || ret == MATCHED_RECURSIVE)
1522                         ce->ce_flags &= ~clear_mask;
1523                 cache++;
1524                 progress_nr++;
1525         }
1526
1527         display_progress(istate->progress, progress_nr);
1528         return nr - (cache_end - cache);
1529 }
1530
1531 static int clear_ce_flags(struct index_state *istate,
1532                           int select_mask, int clear_mask,
1533                           struct pattern_list *pl,
1534                           int show_progress)
1535 {
1536         static struct strbuf prefix = STRBUF_INIT;
1537         char label[100];
1538         int rval;
1539
1540         strbuf_reset(&prefix);
1541         if (show_progress)
1542                 istate->progress = start_delayed_progress(
1543                                         _("Updating index flags"),
1544                                         istate->cache_nr);
1545
1546         xsnprintf(label, sizeof(label), "clear_ce_flags(0x%08lx,0x%08lx)",
1547                   (unsigned long)select_mask, (unsigned long)clear_mask);
1548         trace2_region_enter("unpack_trees", label, the_repository);
1549         rval = clear_ce_flags_1(istate,
1550                                 istate->cache,
1551                                 istate->cache_nr,
1552                                 &prefix,
1553                                 select_mask, clear_mask,
1554                                 pl, 0, 0);
1555         trace2_region_leave("unpack_trees", label, the_repository);
1556
1557         stop_progress(&istate->progress);
1558         return rval;
1559 }
1560
1561 /*
1562  * Set/Clear CE_NEW_SKIP_WORKTREE according to $GIT_DIR/info/sparse-checkout
1563  */
1564 static void mark_new_skip_worktree(struct pattern_list *pl,
1565                                    struct index_state *istate,
1566                                    int select_flag, int skip_wt_flag,
1567                                    int show_progress)
1568 {
1569         int i;
1570
1571         /*
1572          * 1. Pretend the narrowest worktree: only unmerged entries
1573          * are checked out
1574          */
1575         for (i = 0; i < istate->cache_nr; i++) {
1576                 struct cache_entry *ce = istate->cache[i];
1577
1578                 if (select_flag && !(ce->ce_flags & select_flag))
1579                         continue;
1580
1581                 if (!ce_stage(ce) && !(ce->ce_flags & CE_CONFLICTED))
1582                         ce->ce_flags |= skip_wt_flag;
1583                 else
1584                         ce->ce_flags &= ~skip_wt_flag;
1585         }
1586
1587         /*
1588          * 2. Widen worktree according to sparse-checkout file.
1589          * Matched entries will have skip_wt_flag cleared (i.e. "in")
1590          */
1591         clear_ce_flags(istate, select_flag, skip_wt_flag, pl, show_progress);
1592 }
1593
1594 static void populate_from_existing_patterns(struct unpack_trees_options *o,
1595                                             struct pattern_list *pl)
1596 {
1597         if (get_sparse_checkout_patterns(pl) < 0)
1598                 o->skip_sparse_checkout = 1;
1599         else
1600                 o->pl = pl;
1601 }
1602
1603
1604 static int verify_absent(const struct cache_entry *,
1605                          enum unpack_trees_error_types,
1606                          struct unpack_trees_options *);
1607 /*
1608  * N-way merge "len" trees.  Returns 0 on success, -1 on failure to manipulate the
1609  * resulting index, -2 on failure to reflect the changes to the work tree.
1610  *
1611  * CE_ADDED, CE_UNPACKED and CE_NEW_SKIP_WORKTREE are used internally
1612  */
1613 int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options *o)
1614 {
1615         struct repository *repo = the_repository;
1616         int i, ret;
1617         static struct cache_entry *dfc;
1618         struct pattern_list pl;
1619         int free_pattern_list = 0;
1620
1621         if (len > MAX_UNPACK_TREES)
1622                 die("unpack_trees takes at most %d trees", MAX_UNPACK_TREES);
1623
1624         trace_performance_enter();
1625         trace2_region_enter("unpack_trees", "unpack_trees", the_repository);
1626
1627         prepare_repo_settings(repo);
1628         if (repo->settings.command_requires_full_index) {
1629                 ensure_full_index(o->src_index);
1630                 ensure_full_index(o->dst_index);
1631         }
1632
1633         if (!core_apply_sparse_checkout || !o->update)
1634                 o->skip_sparse_checkout = 1;
1635         if (!o->skip_sparse_checkout && !o->pl) {
1636                 memset(&pl, 0, sizeof(pl));
1637                 free_pattern_list = 1;
1638                 populate_from_existing_patterns(o, &pl);
1639         }
1640
1641         memset(&o->result, 0, sizeof(o->result));
1642         o->result.initialized = 1;
1643         o->result.timestamp.sec = o->src_index->timestamp.sec;
1644         o->result.timestamp.nsec = o->src_index->timestamp.nsec;
1645         o->result.version = o->src_index->version;
1646         if (!o->src_index->split_index) {
1647                 o->result.split_index = NULL;
1648         } else if (o->src_index == o->dst_index) {
1649                 /*
1650                  * o->dst_index (and thus o->src_index) will be discarded
1651                  * and overwritten with o->result at the end of this function,
1652                  * so just use src_index's split_index to avoid having to
1653                  * create a new one.
1654                  */
1655                 o->result.split_index = o->src_index->split_index;
1656                 o->result.split_index->refcount++;
1657         } else {
1658                 o->result.split_index = init_split_index(&o->result);
1659         }
1660         oidcpy(&o->result.oid, &o->src_index->oid);
1661         o->merge_size = len;
1662         mark_all_ce_unused(o->src_index);
1663
1664         o->result.fsmonitor_last_update =
1665                 xstrdup_or_null(o->src_index->fsmonitor_last_update);
1666
1667         /*
1668          * Sparse checkout loop #1: set NEW_SKIP_WORKTREE on existing entries
1669          */
1670         if (!o->skip_sparse_checkout)
1671                 mark_new_skip_worktree(o->pl, o->src_index, 0,
1672                                        CE_NEW_SKIP_WORKTREE, o->verbose_update);
1673
1674         if (!dfc)
1675                 dfc = xcalloc(1, cache_entry_size(0));
1676         o->df_conflict_entry = dfc;
1677
1678         if (len) {
1679                 const char *prefix = o->prefix ? o->prefix : "";
1680                 struct traverse_info info;
1681
1682                 setup_traverse_info(&info, prefix);
1683                 info.fn = unpack_callback;
1684                 info.data = o;
1685                 info.show_all_errors = o->show_all_errors;
1686                 info.pathspec = o->pathspec;
1687
1688                 if (o->prefix) {
1689                         /*
1690                          * Unpack existing index entries that sort before the
1691                          * prefix the tree is spliced into.  Note that o->merge
1692                          * is always true in this case.
1693                          */
1694                         while (1) {
1695                                 struct cache_entry *ce = next_cache_entry(o);
1696                                 if (!ce)
1697                                         break;
1698                                 if (ce_in_traverse_path(ce, &info))
1699                                         break;
1700                                 if (unpack_index_entry(ce, o) < 0)
1701                                         goto return_failed;
1702                         }
1703                 }
1704
1705                 trace_performance_enter();
1706                 trace2_region_enter("unpack_trees", "traverse_trees", the_repository);
1707                 ret = traverse_trees(o->src_index, len, t, &info);
1708                 trace2_region_leave("unpack_trees", "traverse_trees", the_repository);
1709                 trace_performance_leave("traverse_trees");
1710                 if (ret < 0)
1711                         goto return_failed;
1712         }
1713
1714         /* Any left-over entries in the index? */
1715         if (o->merge) {
1716                 while (1) {
1717                         struct cache_entry *ce = next_cache_entry(o);
1718                         if (!ce)
1719                                 break;
1720                         if (unpack_index_entry(ce, o) < 0)
1721                                 goto return_failed;
1722                 }
1723         }
1724         mark_all_ce_unused(o->src_index);
1725
1726         if (o->trivial_merges_only && o->nontrivial_merge) {
1727                 ret = unpack_failed(o, "Merge requires file-level merging");
1728                 goto done;
1729         }
1730
1731         if (!o->skip_sparse_checkout) {
1732                 /*
1733                  * Sparse checkout loop #2: set NEW_SKIP_WORKTREE on entries not in loop #1
1734                  * If they will have NEW_SKIP_WORKTREE, also set CE_SKIP_WORKTREE
1735                  * so apply_sparse_checkout() won't attempt to remove it from worktree
1736                  */
1737                 mark_new_skip_worktree(o->pl, &o->result,
1738                                        CE_ADDED, CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE,
1739                                        o->verbose_update);
1740
1741                 ret = 0;
1742                 for (i = 0; i < o->result.cache_nr; i++) {
1743                         struct cache_entry *ce = o->result.cache[i];
1744
1745                         /*
1746                          * Entries marked with CE_ADDED in merged_entry() do not have
1747                          * verify_absent() check (the check is effectively disabled
1748                          * because CE_NEW_SKIP_WORKTREE is set unconditionally).
1749                          *
1750                          * Do the real check now because we have had
1751                          * correct CE_NEW_SKIP_WORKTREE
1752                          */
1753                         if (ce->ce_flags & CE_ADDED &&
1754                             verify_absent(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
1755                                 ret = 1;
1756
1757                         if (apply_sparse_checkout(&o->result, ce, o))
1758                                 ret = 1;
1759                 }
1760                 if (ret == 1) {
1761                         /*
1762                          * Inability to sparsify or de-sparsify individual
1763                          * paths is not an error, but just a warning.
1764                          */
1765                         if (o->show_all_errors)
1766                                 display_warning_msgs(o);
1767                         ret = 0;
1768                 }
1769         }
1770
1771         ret = check_updates(o, &o->result) ? (-2) : 0;
1772         if (o->dst_index) {
1773                 move_index_extensions(&o->result, o->src_index);
1774                 if (!ret) {
1775                         if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
1776                                 cache_tree_verify(the_repository, &o->result);
1777                         if (!cache_tree_fully_valid(o->result.cache_tree))
1778                                 cache_tree_update(&o->result,
1779                                                   WRITE_TREE_SILENT |
1780                                                   WRITE_TREE_REPAIR);
1781                 }
1782
1783                 o->result.updated_workdir = 1;
1784                 discard_index(o->dst_index);
1785                 *o->dst_index = o->result;
1786         } else {
1787                 discard_index(&o->result);
1788         }
1789         o->src_index = NULL;
1790
1791 done:
1792         if (free_pattern_list)
1793                 clear_pattern_list(&pl);
1794         trace2_region_leave("unpack_trees", "unpack_trees", the_repository);
1795         trace_performance_leave("unpack_trees");
1796         return ret;
1797
1798 return_failed:
1799         if (o->show_all_errors)
1800                 display_error_msgs(o);
1801         mark_all_ce_unused(o->src_index);
1802         ret = unpack_failed(o, NULL);
1803         if (o->exiting_early)
1804                 ret = 0;
1805         goto done;
1806 }
1807
1808 /*
1809  * Update SKIP_WORKTREE bits according to sparsity patterns, and update
1810  * working directory to match.
1811  *
1812  * CE_NEW_SKIP_WORKTREE is used internally.
1813  */
1814 enum update_sparsity_result update_sparsity(struct unpack_trees_options *o)
1815 {
1816         enum update_sparsity_result ret = UPDATE_SPARSITY_SUCCESS;
1817         struct pattern_list pl;
1818         int i;
1819         unsigned old_show_all_errors;
1820         int free_pattern_list = 0;
1821
1822         old_show_all_errors = o->show_all_errors;
1823         o->show_all_errors = 1;
1824
1825         /* Sanity checks */
1826         if (!o->update || o->index_only || o->skip_sparse_checkout)
1827                 BUG("update_sparsity() is for reflecting sparsity patterns in working directory");
1828         if (o->src_index != o->dst_index || o->fn)
1829                 BUG("update_sparsity() called wrong");
1830
1831         trace_performance_enter();
1832
1833         /* If we weren't given patterns, use the recorded ones */
1834         if (!o->pl) {
1835                 memset(&pl, 0, sizeof(pl));
1836                 free_pattern_list = 1;
1837                 populate_from_existing_patterns(o, &pl);
1838                 if (o->skip_sparse_checkout)
1839                         goto skip_sparse_checkout;
1840         }
1841
1842         /* Set NEW_SKIP_WORKTREE on existing entries. */
1843         mark_all_ce_unused(o->src_index);
1844         mark_new_skip_worktree(o->pl, o->src_index, 0,
1845                                CE_NEW_SKIP_WORKTREE, o->verbose_update);
1846
1847         /* Then loop over entries and update/remove as needed */
1848         ret = UPDATE_SPARSITY_SUCCESS;
1849         for (i = 0; i < o->src_index->cache_nr; i++) {
1850                 struct cache_entry *ce = o->src_index->cache[i];
1851
1852
1853                 if (ce_stage(ce)) {
1854                         /* -1 because for loop will increment by 1 */
1855                         i += warn_conflicted_path(o->src_index, i, o) - 1;
1856                         ret = UPDATE_SPARSITY_WARNINGS;
1857                         continue;
1858                 }
1859
1860                 if (apply_sparse_checkout(o->src_index, ce, o))
1861                         ret = UPDATE_SPARSITY_WARNINGS;
1862         }
1863
1864 skip_sparse_checkout:
1865         if (check_updates(o, o->src_index))
1866                 ret = UPDATE_SPARSITY_WORKTREE_UPDATE_FAILURES;
1867
1868         display_warning_msgs(o);
1869         o->show_all_errors = old_show_all_errors;
1870         if (free_pattern_list)
1871                 clear_pattern_list(&pl);
1872         trace_performance_leave("update_sparsity");
1873         return ret;
1874 }
1875
1876 /* Here come the merge functions */
1877
1878 static int reject_merge(const struct cache_entry *ce,
1879                         struct unpack_trees_options *o)
1880 {
1881         return add_rejected_path(o, ERROR_WOULD_OVERWRITE, ce->name);
1882 }
1883
1884 static int same(const struct cache_entry *a, const struct cache_entry *b)
1885 {
1886         if (!!a != !!b)
1887                 return 0;
1888         if (!a && !b)
1889                 return 1;
1890         if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
1891                 return 0;
1892         return a->ce_mode == b->ce_mode &&
1893                oideq(&a->oid, &b->oid);
1894 }
1895
1896
1897 /*
1898  * When a CE gets turned into an unmerged entry, we
1899  * want it to be up-to-date
1900  */
1901 static int verify_uptodate_1(const struct cache_entry *ce,
1902                              struct unpack_trees_options *o,
1903                              enum unpack_trees_error_types error_type)
1904 {
1905         struct stat st;
1906
1907         if (o->index_only)
1908                 return 0;
1909
1910         /*
1911          * CE_VALID and CE_SKIP_WORKTREE cheat, we better check again
1912          * if this entry is truly up-to-date because this file may be
1913          * overwritten.
1914          */
1915         if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
1916                 ; /* keep checking */
1917         else if (o->reset || ce_uptodate(ce))
1918                 return 0;
1919
1920         if (!lstat(ce->name, &st)) {
1921                 int flags = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE;
1922                 unsigned changed = ie_match_stat(o->src_index, ce, &st, flags);
1923
1924                 if (submodule_from_ce(ce)) {
1925                         int r = check_submodule_move_head(ce,
1926                                 "HEAD", oid_to_hex(&ce->oid), o);
1927                         if (r)
1928                                 return add_rejected_path(o, error_type, ce->name);
1929                         return 0;
1930                 }
1931
1932                 if (!changed)
1933                         return 0;
1934                 /*
1935                  * Historic default policy was to allow submodule to be out
1936                  * of sync wrt the superproject index. If the submodule was
1937                  * not considered interesting above, we don't care here.
1938                  */
1939                 if (S_ISGITLINK(ce->ce_mode))
1940                         return 0;
1941
1942                 errno = 0;
1943         }
1944         if (errno == ENOENT)
1945                 return 0;
1946         return add_rejected_path(o, error_type, ce->name);
1947 }
1948
1949 int verify_uptodate(const struct cache_entry *ce,
1950                     struct unpack_trees_options *o)
1951 {
1952         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
1953                 return 0;
1954         return verify_uptodate_1(ce, o, ERROR_NOT_UPTODATE_FILE);
1955 }
1956
1957 static int verify_uptodate_sparse(const struct cache_entry *ce,
1958                                   struct unpack_trees_options *o)
1959 {
1960         return verify_uptodate_1(ce, o, WARNING_SPARSE_NOT_UPTODATE_FILE);
1961 }
1962
1963 /*
1964  * TODO: We should actually invalidate o->result, not src_index [1].
1965  * But since cache tree and untracked cache both are not copied to
1966  * o->result until unpacking is complete, we invalidate them on
1967  * src_index instead with the assumption that they will be copied to
1968  * dst_index at the end.
1969  *
1970  * [1] src_index->cache_tree is also used in unpack_callback() so if
1971  * we invalidate o->result, we need to update it to use
1972  * o->result.cache_tree as well.
1973  */
1974 static void invalidate_ce_path(const struct cache_entry *ce,
1975                                struct unpack_trees_options *o)
1976 {
1977         if (!ce)
1978                 return;
1979         cache_tree_invalidate_path(o->src_index, ce->name);
1980         untracked_cache_invalidate_path(o->src_index, ce->name, 1);
1981 }
1982
1983 /*
1984  * Check that checking out ce->sha1 in subdir ce->name is not
1985  * going to overwrite any working files.
1986  */
1987 static int verify_clean_submodule(const char *old_sha1,
1988                                   const struct cache_entry *ce,
1989                                   struct unpack_trees_options *o)
1990 {
1991         if (!submodule_from_ce(ce))
1992                 return 0;
1993
1994         return check_submodule_move_head(ce, old_sha1,
1995                                          oid_to_hex(&ce->oid), o);
1996 }
1997
1998 static int verify_clean_subdirectory(const struct cache_entry *ce,
1999                                      struct unpack_trees_options *o)
2000 {
2001         /*
2002          * we are about to extract "ce->name"; we would not want to lose
2003          * anything in the existing directory there.
2004          */
2005         int namelen;
2006         int i;
2007         struct dir_struct d;
2008         char *pathbuf;
2009         int cnt = 0;
2010
2011         if (S_ISGITLINK(ce->ce_mode)) {
2012                 struct object_id oid;
2013                 int sub_head = resolve_gitlink_ref(ce->name, "HEAD", &oid);
2014                 /*
2015                  * If we are not going to update the submodule, then
2016                  * we don't care.
2017                  */
2018                 if (!sub_head && oideq(&oid, &ce->oid))
2019                         return 0;
2020                 return verify_clean_submodule(sub_head ? NULL : oid_to_hex(&oid),
2021                                               ce, o);
2022         }
2023
2024         /*
2025          * First let's make sure we do not have a local modification
2026          * in that directory.
2027          */
2028         namelen = ce_namelen(ce);
2029         for (i = locate_in_src_index(ce, o);
2030              i < o->src_index->cache_nr;
2031              i++) {
2032                 struct cache_entry *ce2 = o->src_index->cache[i];
2033                 int len = ce_namelen(ce2);
2034                 if (len < namelen ||
2035                     strncmp(ce->name, ce2->name, namelen) ||
2036                     ce2->name[namelen] != '/')
2037                         break;
2038                 /*
2039                  * ce2->name is an entry in the subdirectory to be
2040                  * removed.
2041                  */
2042                 if (!ce_stage(ce2)) {
2043                         if (verify_uptodate(ce2, o))
2044                                 return -1;
2045                         add_entry(o, ce2, CE_REMOVE, 0);
2046                         invalidate_ce_path(ce, o);
2047                         mark_ce_used(ce2, o);
2048                 }
2049                 cnt++;
2050         }
2051
2052         /*
2053          * Then we need to make sure that we do not lose a locally
2054          * present file that is not ignored.
2055          */
2056         pathbuf = xstrfmt("%.*s/", namelen, ce->name);
2057
2058         memset(&d, 0, sizeof(d));
2059         if (o->dir)
2060                 d.exclude_per_dir = o->dir->exclude_per_dir;
2061         i = read_directory(&d, o->src_index, pathbuf, namelen+1, NULL);
2062         if (i)
2063                 return add_rejected_path(o, ERROR_NOT_UPTODATE_DIR, ce->name);
2064         free(pathbuf);
2065         return cnt;
2066 }
2067
2068 /*
2069  * This gets called when there was no index entry for the tree entry 'dst',
2070  * but we found a file in the working tree that 'lstat()' said was fine,
2071  * and we're on a case-insensitive filesystem.
2072  *
2073  * See if we can find a case-insensitive match in the index that also
2074  * matches the stat information, and assume it's that other file!
2075  */
2076 static int icase_exists(struct unpack_trees_options *o, const char *name, int len, struct stat *st)
2077 {
2078         const struct cache_entry *src;
2079
2080         src = index_file_exists(o->src_index, name, len, 1);
2081         return src && !ie_match_stat(o->src_index, src, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2082 }
2083
2084 static int check_ok_to_remove(const char *name, int len, int dtype,
2085                               const struct cache_entry *ce, struct stat *st,
2086                               enum unpack_trees_error_types error_type,
2087                               struct unpack_trees_options *o)
2088 {
2089         const struct cache_entry *result;
2090
2091         /*
2092          * It may be that the 'lstat()' succeeded even though
2093          * target 'ce' was absent, because there is an old
2094          * entry that is different only in case..
2095          *
2096          * Ignore that lstat() if it matches.
2097          */
2098         if (ignore_case && icase_exists(o, name, len, st))
2099                 return 0;
2100
2101         if (o->dir &&
2102             is_excluded(o->dir, o->src_index, name, &dtype))
2103                 /*
2104                  * ce->name is explicitly excluded, so it is Ok to
2105                  * overwrite it.
2106                  */
2107                 return 0;
2108         if (S_ISDIR(st->st_mode)) {
2109                 /*
2110                  * We are checking out path "foo" and
2111                  * found "foo/." in the working tree.
2112                  * This is tricky -- if we have modified
2113                  * files that are in "foo/" we would lose
2114                  * them.
2115                  */
2116                 if (verify_clean_subdirectory(ce, o) < 0)
2117                         return -1;
2118                 return 0;
2119         }
2120
2121         /*
2122          * The previous round may already have decided to
2123          * delete this path, which is in a subdirectory that
2124          * is being replaced with a blob.
2125          */
2126         result = index_file_exists(&o->result, name, len, 0);
2127         if (result) {
2128                 if (result->ce_flags & CE_REMOVE)
2129                         return 0;
2130         }
2131
2132         return add_rejected_path(o, error_type, name);
2133 }
2134
2135 /*
2136  * We do not want to remove or overwrite a working tree file that
2137  * is not tracked, unless it is ignored.
2138  */
2139 static int verify_absent_1(const struct cache_entry *ce,
2140                            enum unpack_trees_error_types error_type,
2141                            struct unpack_trees_options *o)
2142 {
2143         int len;
2144         struct stat st;
2145
2146         if (o->index_only || o->reset || !o->update)
2147                 return 0;
2148
2149         len = check_leading_path(ce->name, ce_namelen(ce), 0);
2150         if (!len)
2151                 return 0;
2152         else if (len > 0) {
2153                 char *path;
2154                 int ret;
2155
2156                 path = xmemdupz(ce->name, len);
2157                 if (lstat(path, &st))
2158                         ret = error_errno("cannot stat '%s'", path);
2159                 else {
2160                         if (submodule_from_ce(ce))
2161                                 ret = check_submodule_move_head(ce,
2162                                                                 oid_to_hex(&ce->oid),
2163                                                                 NULL, o);
2164                         else
2165                                 ret = check_ok_to_remove(path, len, DT_UNKNOWN, NULL,
2166                                                          &st, error_type, o);
2167                 }
2168                 free(path);
2169                 return ret;
2170         } else if (lstat(ce->name, &st)) {
2171                 if (errno != ENOENT)
2172                         return error_errno("cannot stat '%s'", ce->name);
2173                 return 0;
2174         } else {
2175                 if (submodule_from_ce(ce))
2176                         return check_submodule_move_head(ce, oid_to_hex(&ce->oid),
2177                                                          NULL, o);
2178
2179                 return check_ok_to_remove(ce->name, ce_namelen(ce),
2180                                           ce_to_dtype(ce), ce, &st,
2181                                           error_type, o);
2182         }
2183 }
2184
2185 static int verify_absent(const struct cache_entry *ce,
2186                          enum unpack_trees_error_types error_type,
2187                          struct unpack_trees_options *o)
2188 {
2189         if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
2190                 return 0;
2191         return verify_absent_1(ce, error_type, o);
2192 }
2193
2194 static int verify_absent_sparse(const struct cache_entry *ce,
2195                                 enum unpack_trees_error_types error_type,
2196                                 struct unpack_trees_options *o)
2197 {
2198         return verify_absent_1(ce, error_type, o);
2199 }
2200
2201 static int merged_entry(const struct cache_entry *ce,
2202                         const struct cache_entry *old,
2203                         struct unpack_trees_options *o)
2204 {
2205         int update = CE_UPDATE;
2206         struct cache_entry *merge = dup_cache_entry(ce, &o->result);
2207
2208         if (!old) {
2209                 /*
2210                  * New index entries. In sparse checkout, the following
2211                  * verify_absent() will be delayed until after
2212                  * traverse_trees() finishes in unpack_trees(), then:
2213                  *
2214                  *  - CE_NEW_SKIP_WORKTREE will be computed correctly
2215                  *  - verify_absent() be called again, this time with
2216                  *    correct CE_NEW_SKIP_WORKTREE
2217                  *
2218                  * verify_absent() call here does nothing in sparse
2219                  * checkout (i.e. o->skip_sparse_checkout == 0)
2220                  */
2221                 update |= CE_ADDED;
2222                 merge->ce_flags |= CE_NEW_SKIP_WORKTREE;
2223
2224                 if (verify_absent(merge,
2225                                   ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2226                         discard_cache_entry(merge);
2227                         return -1;
2228                 }
2229                 invalidate_ce_path(merge, o);
2230
2231                 if (submodule_from_ce(ce) && file_exists(ce->name)) {
2232                         int ret = check_submodule_move_head(ce, NULL,
2233                                                             oid_to_hex(&ce->oid),
2234                                                             o);
2235                         if (ret)
2236                                 return ret;
2237                 }
2238
2239         } else if (!(old->ce_flags & CE_CONFLICTED)) {
2240                 /*
2241                  * See if we can re-use the old CE directly?
2242                  * That way we get the uptodate stat info.
2243                  *
2244                  * This also removes the UPDATE flag on a match; otherwise
2245                  * we will end up overwriting local changes in the work tree.
2246                  */
2247                 if (same(old, merge)) {
2248                         copy_cache_entry(merge, old);
2249                         update = 0;
2250                 } else {
2251                         if (verify_uptodate(old, o)) {
2252                                 discard_cache_entry(merge);
2253                                 return -1;
2254                         }
2255                         /* Migrate old flags over */
2256                         update |= old->ce_flags & (CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
2257                         invalidate_ce_path(old, o);
2258                 }
2259
2260                 if (submodule_from_ce(ce) && file_exists(ce->name)) {
2261                         int ret = check_submodule_move_head(ce, oid_to_hex(&old->oid),
2262                                                             oid_to_hex(&ce->oid),
2263                                                             o);
2264                         if (ret)
2265                                 return ret;
2266                 }
2267         } else {
2268                 /*
2269                  * Previously unmerged entry left as an existence
2270                  * marker by read_index_unmerged();
2271                  */
2272                 invalidate_ce_path(old, o);
2273         }
2274
2275         if (do_add_entry(o, merge, update, CE_STAGEMASK) < 0)
2276                 return -1;
2277         return 1;
2278 }
2279
2280 static int deleted_entry(const struct cache_entry *ce,
2281                          const struct cache_entry *old,
2282                          struct unpack_trees_options *o)
2283 {
2284         /* Did it exist in the index? */
2285         if (!old) {
2286                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2287                         return -1;
2288                 return 0;
2289         }
2290         if (!(old->ce_flags & CE_CONFLICTED) && verify_uptodate(old, o))
2291                 return -1;
2292         add_entry(o, ce, CE_REMOVE, 0);
2293         invalidate_ce_path(ce, o);
2294         return 1;
2295 }
2296
2297 static int keep_entry(const struct cache_entry *ce,
2298                       struct unpack_trees_options *o)
2299 {
2300         add_entry(o, ce, 0, 0);
2301         if (ce_stage(ce))
2302                 invalidate_ce_path(ce, o);
2303         return 1;
2304 }
2305
2306 #if DBRT_DEBUG
2307 static void show_stage_entry(FILE *o,
2308                              const char *label, const struct cache_entry *ce)
2309 {
2310         if (!ce)
2311                 fprintf(o, "%s (missing)\n", label);
2312         else
2313                 fprintf(o, "%s%06o %s %d\t%s\n",
2314                         label,
2315                         ce->ce_mode,
2316                         oid_to_hex(&ce->oid),
2317                         ce_stage(ce),
2318                         ce->name);
2319 }
2320 #endif
2321
2322 int threeway_merge(const struct cache_entry * const *stages,
2323                    struct unpack_trees_options *o)
2324 {
2325         const struct cache_entry *index;
2326         const struct cache_entry *head;
2327         const struct cache_entry *remote = stages[o->head_idx + 1];
2328         int count;
2329         int head_match = 0;
2330         int remote_match = 0;
2331
2332         int df_conflict_head = 0;
2333         int df_conflict_remote = 0;
2334
2335         int any_anc_missing = 0;
2336         int no_anc_exists = 1;
2337         int i;
2338
2339         for (i = 1; i < o->head_idx; i++) {
2340                 if (!stages[i] || stages[i] == o->df_conflict_entry)
2341                         any_anc_missing = 1;
2342                 else
2343                         no_anc_exists = 0;
2344         }
2345
2346         index = stages[0];
2347         head = stages[o->head_idx];
2348
2349         if (head == o->df_conflict_entry) {
2350                 df_conflict_head = 1;
2351                 head = NULL;
2352         }
2353
2354         if (remote == o->df_conflict_entry) {
2355                 df_conflict_remote = 1;
2356                 remote = NULL;
2357         }
2358
2359         /*
2360          * First, if there's a #16 situation, note that to prevent #13
2361          * and #14.
2362          */
2363         if (!same(remote, head)) {
2364                 for (i = 1; i < o->head_idx; i++) {
2365                         if (same(stages[i], head)) {
2366                                 head_match = i;
2367                         }
2368                         if (same(stages[i], remote)) {
2369                                 remote_match = i;
2370                         }
2371                 }
2372         }
2373
2374         /*
2375          * We start with cases where the index is allowed to match
2376          * something other than the head: #14(ALT) and #2ALT, where it
2377          * is permitted to match the result instead.
2378          */
2379         /* #14, #14ALT, #2ALT */
2380         if (remote && !df_conflict_head && head_match && !remote_match) {
2381                 if (index && !same(index, remote) && !same(index, head))
2382                         return reject_merge(index, o);
2383                 return merged_entry(remote, index, o);
2384         }
2385         /*
2386          * If we have an entry in the index cache, then we want to
2387          * make sure that it matches head.
2388          */
2389         if (index && !same(index, head))
2390                 return reject_merge(index, o);
2391
2392         if (head) {
2393                 /* #5ALT, #15 */
2394                 if (same(head, remote))
2395                         return merged_entry(head, index, o);
2396                 /* #13, #3ALT */
2397                 if (!df_conflict_remote && remote_match && !head_match)
2398                         return merged_entry(head, index, o);
2399         }
2400
2401         /* #1 */
2402         if (!head && !remote && any_anc_missing)
2403                 return 0;
2404
2405         /*
2406          * Under the "aggressive" rule, we resolve mostly trivial
2407          * cases that we historically had git-merge-one-file resolve.
2408          */
2409         if (o->aggressive) {
2410                 int head_deleted = !head;
2411                 int remote_deleted = !remote;
2412                 const struct cache_entry *ce = NULL;
2413
2414                 if (index)
2415                         ce = index;
2416                 else if (head)
2417                         ce = head;
2418                 else if (remote)
2419                         ce = remote;
2420                 else {
2421                         for (i = 1; i < o->head_idx; i++) {
2422                                 if (stages[i] && stages[i] != o->df_conflict_entry) {
2423                                         ce = stages[i];
2424                                         break;
2425                                 }
2426                         }
2427                 }
2428
2429                 /*
2430                  * Deleted in both.
2431                  * Deleted in one and unchanged in the other.
2432                  */
2433                 if ((head_deleted && remote_deleted) ||
2434                     (head_deleted && remote && remote_match) ||
2435                     (remote_deleted && head && head_match)) {
2436                         if (index)
2437                                 return deleted_entry(index, index, o);
2438                         if (ce && !head_deleted) {
2439                                 if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2440                                         return -1;
2441                         }
2442                         return 0;
2443                 }
2444                 /*
2445                  * Added in both, identically.
2446                  */
2447                 if (no_anc_exists && head && remote && same(head, remote))
2448                         return merged_entry(head, index, o);
2449
2450         }
2451
2452         /* Below are "no merge" cases, which require that the index be
2453          * up-to-date to avoid the files getting overwritten with
2454          * conflict resolution files.
2455          */
2456         if (index) {
2457                 if (verify_uptodate(index, o))
2458                         return -1;
2459         }
2460
2461         o->nontrivial_merge = 1;
2462
2463         /* #2, #3, #4, #6, #7, #9, #10, #11. */
2464         count = 0;
2465         if (!head_match || !remote_match) {
2466                 for (i = 1; i < o->head_idx; i++) {
2467                         if (stages[i] && stages[i] != o->df_conflict_entry) {
2468                                 keep_entry(stages[i], o);
2469                                 count++;
2470                                 break;
2471                         }
2472                 }
2473         }
2474 #if DBRT_DEBUG
2475         else {
2476                 fprintf(stderr, "read-tree: warning #16 detected\n");
2477                 show_stage_entry(stderr, "head   ", stages[head_match]);
2478                 show_stage_entry(stderr, "remote ", stages[remote_match]);
2479         }
2480 #endif
2481         if (head) { count += keep_entry(head, o); }
2482         if (remote) { count += keep_entry(remote, o); }
2483         return count;
2484 }
2485
2486 /*
2487  * Two-way merge.
2488  *
2489  * The rule is to "carry forward" what is in the index without losing
2490  * information across a "fast-forward", favoring a successful merge
2491  * over a merge failure when it makes sense.  For details of the
2492  * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
2493  *
2494  */
2495 int twoway_merge(const struct cache_entry * const *src,
2496                  struct unpack_trees_options *o)
2497 {
2498         const struct cache_entry *current = src[0];
2499         const struct cache_entry *oldtree = src[1];
2500         const struct cache_entry *newtree = src[2];
2501
2502         if (o->merge_size != 2)
2503                 return error("Cannot do a twoway merge of %d trees",
2504                              o->merge_size);
2505
2506         if (oldtree == o->df_conflict_entry)
2507                 oldtree = NULL;
2508         if (newtree == o->df_conflict_entry)
2509                 newtree = NULL;
2510
2511         if (current) {
2512                 if (current->ce_flags & CE_CONFLICTED) {
2513                         if (same(oldtree, newtree) || o->reset) {
2514                                 if (!newtree)
2515                                         return deleted_entry(current, current, o);
2516                                 else
2517                                         return merged_entry(newtree, current, o);
2518                         }
2519                         return reject_merge(current, o);
2520                 } else if ((!oldtree && !newtree) || /* 4 and 5 */
2521                          (!oldtree && newtree &&
2522                           same(current, newtree)) || /* 6 and 7 */
2523                          (oldtree && newtree &&
2524                           same(oldtree, newtree)) || /* 14 and 15 */
2525                          (oldtree && newtree &&
2526                           !same(oldtree, newtree) && /* 18 and 19 */
2527                           same(current, newtree))) {
2528                         return keep_entry(current, o);
2529                 } else if (oldtree && !newtree && same(current, oldtree)) {
2530                         /* 10 or 11 */
2531                         return deleted_entry(oldtree, current, o);
2532                 } else if (oldtree && newtree &&
2533                          same(current, oldtree) && !same(current, newtree)) {
2534                         /* 20 or 21 */
2535                         return merged_entry(newtree, current, o);
2536                 } else
2537                         return reject_merge(current, o);
2538         }
2539         else if (newtree) {
2540                 if (oldtree && !o->initial_checkout) {
2541                         /*
2542                          * deletion of the path was staged;
2543                          */
2544                         if (same(oldtree, newtree))
2545                                 return 1;
2546                         return reject_merge(oldtree, o);
2547                 }
2548                 return merged_entry(newtree, current, o);
2549         }
2550         return deleted_entry(oldtree, current, o);
2551 }
2552
2553 /*
2554  * Bind merge.
2555  *
2556  * Keep the index entries at stage0, collapse stage1 but make sure
2557  * stage0 does not have anything there.
2558  */
2559 int bind_merge(const struct cache_entry * const *src,
2560                struct unpack_trees_options *o)
2561 {
2562         const struct cache_entry *old = src[0];
2563         const struct cache_entry *a = src[1];
2564
2565         if (o->merge_size != 1)
2566                 return error("Cannot do a bind merge of %d trees",
2567                              o->merge_size);
2568         if (a && old)
2569                 return o->quiet ? -1 :
2570                         error(ERRORMSG(o, ERROR_BIND_OVERLAP),
2571                               super_prefixed(a->name),
2572                               super_prefixed(old->name));
2573         if (!a)
2574                 return keep_entry(old, o);
2575         else
2576                 return merged_entry(a, NULL, o);
2577 }
2578
2579 /*
2580  * One-way merge.
2581  *
2582  * The rule is:
2583  * - take the stat information from stage0, take the data from stage1
2584  */
2585 int oneway_merge(const struct cache_entry * const *src,
2586                  struct unpack_trees_options *o)
2587 {
2588         const struct cache_entry *old = src[0];
2589         const struct cache_entry *a = src[1];
2590
2591         if (o->merge_size != 1)
2592                 return error("Cannot do a oneway merge of %d trees",
2593                              o->merge_size);
2594
2595         if (!a || a == o->df_conflict_entry)
2596                 return deleted_entry(old, old, o);
2597
2598         if (old && same(old, a)) {
2599                 int update = 0;
2600                 if (o->reset && o->update && !ce_uptodate(old) && !ce_skip_worktree(old) &&
2601                         !(old->ce_flags & CE_FSMONITOR_VALID)) {
2602                         struct stat st;
2603                         if (lstat(old->name, &st) ||
2604                             ie_match_stat(o->src_index, old, &st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE))
2605                                 update |= CE_UPDATE;
2606                 }
2607                 if (o->update && S_ISGITLINK(old->ce_mode) &&
2608                     should_update_submodules() && !verify_uptodate(old, o))
2609                         update |= CE_UPDATE;
2610                 add_entry(o, old, update, CE_STAGEMASK);
2611                 return 0;
2612         }
2613         return merged_entry(a, old, o);
2614 }
2615
2616 /*
2617  * Merge worktree and untracked entries in a stash entry.
2618  *
2619  * Ignore all index entries. Collapse remaining trees but make sure that they
2620  * don't have any conflicting files.
2621  */
2622 int stash_worktree_untracked_merge(const struct cache_entry * const *src,
2623                                    struct unpack_trees_options *o)
2624 {
2625         const struct cache_entry *worktree = src[1];
2626         const struct cache_entry *untracked = src[2];
2627
2628         if (o->merge_size != 2)
2629                 BUG("invalid merge_size: %d", o->merge_size);
2630
2631         if (worktree && untracked)
2632                 return error(_("worktree and untracked commit have duplicate entries: %s"),
2633                              super_prefixed(worktree->name));
2634
2635         return merged_entry(worktree ? worktree : untracked, NULL, o);
2636 }