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