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