1 #include "cache.h"
2 #include "strvec.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "dir.h"
6 #include "tree.h"
7 #include "tree-walk.h"
8 #include "cache-tree.h"
9 #include "unpack-trees.h"
10 #include "progress.h"
11 #include "refs.h"
12 #include "attr.h"
13 #include "split-index.h"
14 #include "submodule.h"
15 #include "submodule-config.h"
16 #include "fsmonitor.h"
17 #include "object-store.h"
18 #include "promisor-remote.h"
19 #include "entry.h"
20 #include "parallel-checkout.h"
21 
22 /*
23  * Error messages expected by scripts out of plumbing commands such as
24  * read-tree.  Non-scripted Porcelain is not required to use these messages
25  * and in fact are encouraged to reword them to better suit their particular
26  * situation better.  See how "git checkout" and "git merge" replaces
27  * them using setup_unpack_trees_porcelain(), for example.
28  */
29 static const char *unpack_plumbing_errors[NB_UNPACK_TREES_WARNING_TYPES] = {
30 	/* ERROR_WOULD_OVERWRITE */
31 	"Entry '%s' would be overwritten by merge. Cannot merge.",
32 
33 	/* ERROR_NOT_UPTODATE_FILE */
34 	"Entry '%s' not uptodate. Cannot merge.",
35 
36 	/* ERROR_NOT_UPTODATE_DIR */
37 	"Updating '%s' would lose untracked files in it",
38 
39 	/* ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN */
40 	"Untracked working tree file '%s' would be overwritten by merge.",
41 
42 	/* ERROR_WOULD_LOSE_UNTRACKED_REMOVED */
43 	"Untracked working tree file '%s' would be removed by merge.",
44 
45 	/* ERROR_BIND_OVERLAP */
46 	"Entry '%s' overlaps with '%s'.  Cannot bind.",
47 
48 	/* ERROR_WOULD_LOSE_SUBMODULE */
49 	"Submodule '%s' cannot checkout new HEAD.",
50 
51 	/* NB_UNPACK_TREES_ERROR_TYPES; just a meta value */
52 	"",
53 
54 	/* WARNING_SPARSE_NOT_UPTODATE_FILE */
55 	"Path '%s' not uptodate; will not remove from working tree.",
56 
57 	/* WARNING_SPARSE_UNMERGED_FILE */
58 	"Path '%s' unmerged; will not remove from working tree.",
59 
60 	/* WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN */
61 	"Path '%s' already present; will not overwrite with sparse update.",
62 };
63 
64 #define ERRORMSG(o,type) \
65 	( ((o) && (o)->msgs[(type)]) \
66 	  ? ((o)->msgs[(type)])      \
67 	  : (unpack_plumbing_errors[(type)]) )
68 
super_prefixed(const char * path)69 static const char *super_prefixed(const char *path)
70 {
71 	/*
72 	 * It is necessary and sufficient to have two static buffers
73 	 * here, as the return value of this function is fed to
74 	 * error() using the unpack_*_errors[] templates we see above.
75 	 */
76 	static struct strbuf buf[2] = {STRBUF_INIT, STRBUF_INIT};
77 	static int super_prefix_len = -1;
78 	static unsigned idx = ARRAY_SIZE(buf) - 1;
79 
80 	if (super_prefix_len < 0) {
81 		const char *super_prefix = get_super_prefix();
82 		if (!super_prefix) {
83 			super_prefix_len = 0;
84 		} else {
85 			int i;
86 			for (i = 0; i < ARRAY_SIZE(buf); i++)
87 				strbuf_addstr(&buf[i], super_prefix);
88 			super_prefix_len = buf[0].len;
89 		}
90 	}
91 
92 	if (!super_prefix_len)
93 		return path;
94 
95 	if (++idx >= ARRAY_SIZE(buf))
96 		idx = 0;
97 
98 	strbuf_setlen(&buf[idx], super_prefix_len);
99 	strbuf_addstr(&buf[idx], path);
100 
101 	return buf[idx].buf;
102 }
103 
setup_unpack_trees_porcelain(struct unpack_trees_options * opts,const char * cmd)104 void setup_unpack_trees_porcelain(struct unpack_trees_options *opts,
105 				  const char *cmd)
106 {
107 	int i;
108 	const char **msgs = opts->msgs;
109 	const char *msg;
110 
111 	strvec_init(&opts->msgs_to_free);
112 
113 	if (!strcmp(cmd, "checkout"))
114 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
115 		      ? _("Your local changes to the following files would be overwritten by checkout:\n%%s"
116 			  "Please commit your changes or stash them before you switch branches.")
117 		      : _("Your local changes to the following files would be overwritten by checkout:\n%%s");
118 	else if (!strcmp(cmd, "merge"))
119 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
120 		      ? _("Your local changes to the following files would be overwritten by merge:\n%%s"
121 			  "Please commit your changes or stash them before you merge.")
122 		      : _("Your local changes to the following files would be overwritten by merge:\n%%s");
123 	else
124 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
125 		      ? _("Your local changes to the following files would be overwritten by %s:\n%%s"
126 			  "Please commit your changes or stash them before you %s.")
127 		      : _("Your local changes to the following files would be overwritten by %s:\n%%s");
128 	msgs[ERROR_WOULD_OVERWRITE] = msgs[ERROR_NOT_UPTODATE_FILE] =
129 		strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
130 
131 	msgs[ERROR_NOT_UPTODATE_DIR] =
132 		_("Updating the following directories would lose untracked files in them:\n%s");
133 
134 	if (!strcmp(cmd, "checkout"))
135 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
136 		      ? _("The following untracked working tree files would be removed by checkout:\n%%s"
137 			  "Please move or remove them before you switch branches.")
138 		      : _("The following untracked working tree files would be removed by checkout:\n%%s");
139 	else if (!strcmp(cmd, "merge"))
140 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
141 		      ? _("The following untracked working tree files would be removed by merge:\n%%s"
142 			  "Please move or remove them before you merge.")
143 		      : _("The following untracked working tree files would be removed by merge:\n%%s");
144 	else
145 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
146 		      ? _("The following untracked working tree files would be removed by %s:\n%%s"
147 			  "Please move or remove them before you %s.")
148 		      : _("The following untracked working tree files would be removed by %s:\n%%s");
149 	msgs[ERROR_WOULD_LOSE_UNTRACKED_REMOVED] =
150 		strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
151 
152 	if (!strcmp(cmd, "checkout"))
153 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
154 		      ? _("The following untracked working tree files would be overwritten by checkout:\n%%s"
155 			  "Please move or remove them before you switch branches.")
156 		      : _("The following untracked working tree files would be overwritten by checkout:\n%%s");
157 	else if (!strcmp(cmd, "merge"))
158 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
159 		      ? _("The following untracked working tree files would be overwritten by merge:\n%%s"
160 			  "Please move or remove them before you merge.")
161 		      : _("The following untracked working tree files would be overwritten by merge:\n%%s");
162 	else
163 		msg = advice_enabled(ADVICE_COMMIT_BEFORE_MERGE)
164 		      ? _("The following untracked working tree files would be overwritten by %s:\n%%s"
165 			  "Please move or remove them before you %s.")
166 		      : _("The following untracked working tree files would be overwritten by %s:\n%%s");
167 	msgs[ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN] =
168 		strvec_pushf(&opts->msgs_to_free, msg, cmd, cmd);
169 
170 	/*
171 	 * Special case: ERROR_BIND_OVERLAP refers to a pair of paths, we
172 	 * cannot easily display it as a list.
173 	 */
174 	msgs[ERROR_BIND_OVERLAP] = _("Entry '%s' overlaps with '%s'.  Cannot bind.");
175 
176 	msgs[ERROR_WOULD_LOSE_SUBMODULE] =
177 		_("Cannot update submodule:\n%s");
178 
179 	msgs[WARNING_SPARSE_NOT_UPTODATE_FILE] =
180 		_("The following paths are not up to date and were left despite sparse patterns:\n%s");
181 	msgs[WARNING_SPARSE_UNMERGED_FILE] =
182 		_("The following paths are unmerged and were left despite sparse patterns:\n%s");
183 	msgs[WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN] =
184 		_("The following paths were already present and thus not updated despite sparse patterns:\n%s");
185 
186 	opts->show_all_errors = 1;
187 	/* rejected paths may not have a static buffer */
188 	for (i = 0; i < ARRAY_SIZE(opts->unpack_rejects); i++)
189 		opts->unpack_rejects[i].strdup_strings = 1;
190 }
191 
clear_unpack_trees_porcelain(struct unpack_trees_options * opts)192 void clear_unpack_trees_porcelain(struct unpack_trees_options *opts)
193 {
194 	strvec_clear(&opts->msgs_to_free);
195 	memset(opts->msgs, 0, sizeof(opts->msgs));
196 }
197 
do_add_entry(struct unpack_trees_options * o,struct cache_entry * ce,unsigned int set,unsigned int clear)198 static int do_add_entry(struct unpack_trees_options *o, struct cache_entry *ce,
199 			 unsigned int set, unsigned int clear)
200 {
201 	clear |= CE_HASHED;
202 
203 	if (set & CE_REMOVE)
204 		set |= CE_WT_REMOVE;
205 
206 	ce->ce_flags = (ce->ce_flags & ~clear) | set;
207 	return add_index_entry(&o->result, ce,
208 			       ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
209 }
210 
add_entry(struct unpack_trees_options * o,const struct cache_entry * ce,unsigned int set,unsigned int clear)211 static void add_entry(struct unpack_trees_options *o,
212 		      const struct cache_entry *ce,
213 		      unsigned int set, unsigned int clear)
214 {
215 	do_add_entry(o, dup_cache_entry(ce, &o->result), set, clear);
216 }
217 
218 /*
219  * add error messages on path <path>
220  * corresponding to the type <e> with the message <msg>
221  * indicating if it should be display in porcelain or not
222  */
add_rejected_path(struct unpack_trees_options * o,enum unpack_trees_error_types e,const char * path)223 static int add_rejected_path(struct unpack_trees_options *o,
224 			     enum unpack_trees_error_types e,
225 			     const char *path)
226 {
227 	if (o->quiet)
228 		return -1;
229 
230 	if (!o->show_all_errors)
231 		return error(ERRORMSG(o, e), super_prefixed(path));
232 
233 	/*
234 	 * Otherwise, insert in a list for future display by
235 	 * display_(error|warning)_msgs()
236 	 */
237 	string_list_append(&o->unpack_rejects[e], path);
238 	return -1;
239 }
240 
241 /*
242  * display all the error messages stored in a nice way
243  */
display_error_msgs(struct unpack_trees_options * o)244 static void display_error_msgs(struct unpack_trees_options *o)
245 {
246 	int e;
247 	unsigned error_displayed = 0;
248 	for (e = 0; e < NB_UNPACK_TREES_ERROR_TYPES; e++) {
249 		struct string_list *rejects = &o->unpack_rejects[e];
250 
251 		if (rejects->nr > 0) {
252 			int i;
253 			struct strbuf path = STRBUF_INIT;
254 
255 			error_displayed = 1;
256 			for (i = 0; i < rejects->nr; i++)
257 				strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
258 			error(ERRORMSG(o, e), super_prefixed(path.buf));
259 			strbuf_release(&path);
260 		}
261 		string_list_clear(rejects, 0);
262 	}
263 	if (error_displayed)
264 		fprintf(stderr, _("Aborting\n"));
265 }
266 
267 /*
268  * display all the warning messages stored in a nice way
269  */
display_warning_msgs(struct unpack_trees_options * o)270 static void display_warning_msgs(struct unpack_trees_options *o)
271 {
272 	int e;
273 	unsigned warning_displayed = 0;
274 	for (e = NB_UNPACK_TREES_ERROR_TYPES + 1;
275 	     e < NB_UNPACK_TREES_WARNING_TYPES; e++) {
276 		struct string_list *rejects = &o->unpack_rejects[e];
277 
278 		if (rejects->nr > 0) {
279 			int i;
280 			struct strbuf path = STRBUF_INIT;
281 
282 			warning_displayed = 1;
283 			for (i = 0; i < rejects->nr; i++)
284 				strbuf_addf(&path, "\t%s\n", rejects->items[i].string);
285 			warning(ERRORMSG(o, e), super_prefixed(path.buf));
286 			strbuf_release(&path);
287 		}
288 		string_list_clear(rejects, 0);
289 	}
290 	if (warning_displayed)
291 		fprintf(stderr, _("After fixing the above paths, you may want to run `git sparse-checkout reapply`.\n"));
292 }
check_submodule_move_head(const struct cache_entry * ce,const char * old_id,const char * new_id,struct unpack_trees_options * o)293 static int check_submodule_move_head(const struct cache_entry *ce,
294 				     const char *old_id,
295 				     const char *new_id,
296 				     struct unpack_trees_options *o)
297 {
298 	unsigned flags = SUBMODULE_MOVE_HEAD_DRY_RUN;
299 	const struct submodule *sub = submodule_from_ce(ce);
300 
301 	if (!sub)
302 		return 0;
303 
304 	if (o->reset)
305 		flags |= SUBMODULE_MOVE_HEAD_FORCE;
306 
307 	if (submodule_move_head(ce->name, old_id, new_id, flags))
308 		return add_rejected_path(o, ERROR_WOULD_LOSE_SUBMODULE, ce->name);
309 	return 0;
310 }
311 
312 /*
313  * Perform the loading of the repository's gitmodules file.  This function is
314  * used by 'check_update()' to perform loading of the gitmodules file in two
315  * different situations:
316  * (1) before removing entries from the working tree if the gitmodules file has
317  *     been marked for removal.  This situation is specified by 'state' == NULL.
318  * (2) before checking out entries to the working tree if the gitmodules file
319  *     has been marked for update.  This situation is specified by 'state' != NULL.
320  */
load_gitmodules_file(struct index_state * index,struct checkout * state)321 static void load_gitmodules_file(struct index_state *index,
322 				 struct checkout *state)
323 {
324 	int pos = index_name_pos(index, GITMODULES_FILE, strlen(GITMODULES_FILE));
325 
326 	if (pos >= 0) {
327 		struct cache_entry *ce = index->cache[pos];
328 		if (!state && ce->ce_flags & CE_WT_REMOVE) {
329 			repo_read_gitmodules(the_repository, 0);
330 		} else if (state && (ce->ce_flags & CE_UPDATE)) {
331 			submodule_free(the_repository);
332 			checkout_entry(ce, state, NULL, NULL);
333 			repo_read_gitmodules(the_repository, 0);
334 		}
335 	}
336 }
337 
get_progress(struct unpack_trees_options * o,struct index_state * index)338 static struct progress *get_progress(struct unpack_trees_options *o,
339 				     struct index_state *index)
340 {
341 	unsigned cnt = 0, total = 0;
342 
343 	if (!o->update || !o->verbose_update)
344 		return NULL;
345 
346 	for (; cnt < index->cache_nr; cnt++) {
347 		const struct cache_entry *ce = index->cache[cnt];
348 		if (ce->ce_flags & (CE_UPDATE | CE_WT_REMOVE))
349 			total++;
350 	}
351 
352 	return start_delayed_progress(_("Updating files"), total);
353 }
354 
setup_collided_checkout_detection(struct checkout * state,struct index_state * index)355 static void setup_collided_checkout_detection(struct checkout *state,
356 					      struct index_state *index)
357 {
358 	int i;
359 
360 	state->clone = 1;
361 	for (i = 0; i < index->cache_nr; i++)
362 		index->cache[i]->ce_flags &= ~CE_MATCHED;
363 }
364 
report_collided_checkout(struct index_state * index)365 static void report_collided_checkout(struct index_state *index)
366 {
367 	struct string_list list = STRING_LIST_INIT_NODUP;
368 	int i;
369 
370 	for (i = 0; i < index->cache_nr; i++) {
371 		struct cache_entry *ce = index->cache[i];
372 
373 		if (!(ce->ce_flags & CE_MATCHED))
374 			continue;
375 
376 		string_list_append(&list, ce->name);
377 		ce->ce_flags &= ~CE_MATCHED;
378 	}
379 
380 	list.cmp = fspathcmp;
381 	string_list_sort(&list);
382 
383 	if (list.nr) {
384 		warning(_("the following paths have collided (e.g. case-sensitive paths\n"
385 			  "on a case-insensitive filesystem) and only one from the same\n"
386 			  "colliding group is in the working tree:\n"));
387 
388 		for (i = 0; i < list.nr; i++)
389 			fprintf(stderr, "  '%s'\n", list.items[i].string);
390 	}
391 
392 	string_list_clear(&list, 0);
393 }
394 
must_checkout(const struct cache_entry * ce)395 static int must_checkout(const struct cache_entry *ce)
396 {
397 	return ce->ce_flags & CE_UPDATE;
398 }
399 
check_updates(struct unpack_trees_options * o,struct index_state * index)400 static int check_updates(struct unpack_trees_options *o,
401 			 struct index_state *index)
402 {
403 	unsigned cnt = 0;
404 	int errs = 0;
405 	struct progress *progress;
406 	struct checkout state = CHECKOUT_INIT;
407 	int i, pc_workers, pc_threshold;
408 
409 	trace_performance_enter();
410 	state.force = 1;
411 	state.quiet = 1;
412 	state.refresh_cache = 1;
413 	state.istate = index;
414 	clone_checkout_metadata(&state.meta, &o->meta, NULL);
415 
416 	if (!o->update || o->dry_run) {
417 		remove_marked_cache_entries(index, 0);
418 		trace_performance_leave("check_updates");
419 		return 0;
420 	}
421 
422 	if (o->clone)
423 		setup_collided_checkout_detection(&state, index);
424 
425 	progress = get_progress(o, index);
426 
427 	/* Start with clean cache to avoid using any possibly outdated info. */
428 	invalidate_lstat_cache();
429 
430 	git_attr_set_direction(GIT_ATTR_CHECKOUT);
431 
432 	if (should_update_submodules())
433 		load_gitmodules_file(index, NULL);
434 
435 	for (i = 0; i < index->cache_nr; i++) {
436 		const struct cache_entry *ce = index->cache[i];
437 
438 		if (ce->ce_flags & CE_WT_REMOVE) {
439 			display_progress(progress, ++cnt);
440 			unlink_entry(ce);
441 		}
442 	}
443 
444 	remove_marked_cache_entries(index, 0);
445 	remove_scheduled_dirs();
446 
447 	if (should_update_submodules())
448 		load_gitmodules_file(index, &state);
449 
450 	if (has_promisor_remote())
451 		/*
452 		 * Prefetch the objects that are to be checked out in the loop
453 		 * below.
454 		 */
455 		prefetch_cache_entries(index, must_checkout);
456 
457 	get_parallel_checkout_configs(&pc_workers, &pc_threshold);
458 
459 	enable_delayed_checkout(&state);
460 	if (pc_workers > 1)
461 		init_parallel_checkout();
462 	for (i = 0; i < index->cache_nr; i++) {
463 		struct cache_entry *ce = index->cache[i];
464 
465 		if (must_checkout(ce)) {
466 			size_t last_pc_queue_size = pc_queue_size();
467 
468 			if (ce->ce_flags & CE_WT_REMOVE)
469 				BUG("both update and delete flags are set on %s",
470 				    ce->name);
471 			ce->ce_flags &= ~CE_UPDATE;
472 			errs |= checkout_entry(ce, &state, NULL, NULL);
473 
474 			if (last_pc_queue_size == pc_queue_size())
475 				display_progress(progress, ++cnt);
476 		}
477 	}
478 	if (pc_workers > 1)
479 		errs |= run_parallel_checkout(&state, pc_workers, pc_threshold,
480 					      progress, &cnt);
481 	stop_progress(&progress);
482 	errs |= finish_delayed_checkout(&state, NULL, o->verbose_update);
483 	git_attr_set_direction(GIT_ATTR_CHECKIN);
484 
485 	if (o->clone)
486 		report_collided_checkout(index);
487 
488 	trace_performance_leave("check_updates");
489 	return errs != 0;
490 }
491 
492 static int verify_uptodate_sparse(const struct cache_entry *ce,
493 				  struct unpack_trees_options *o);
494 static int verify_absent_sparse(const struct cache_entry *ce,
495 				enum unpack_trees_error_types,
496 				struct unpack_trees_options *o);
497 
apply_sparse_checkout(struct index_state * istate,struct cache_entry * ce,struct unpack_trees_options * o)498 static int apply_sparse_checkout(struct index_state *istate,
499 				 struct cache_entry *ce,
500 				 struct unpack_trees_options *o)
501 {
502 	int was_skip_worktree = ce_skip_worktree(ce);
503 
504 	if (ce->ce_flags & CE_NEW_SKIP_WORKTREE)
505 		ce->ce_flags |= CE_SKIP_WORKTREE;
506 	else
507 		ce->ce_flags &= ~CE_SKIP_WORKTREE;
508 	if (was_skip_worktree != ce_skip_worktree(ce)) {
509 		ce->ce_flags |= CE_UPDATE_IN_BASE;
510 		mark_fsmonitor_invalid(istate, ce);
511 		istate->cache_changed |= CE_ENTRY_CHANGED;
512 	}
513 
514 	/*
515 	 * if (!was_skip_worktree && !ce_skip_worktree()) {
516 	 *	This is perfectly normal. Move on;
517 	 * }
518 	 */
519 
520 	/*
521 	 * Merge strategies may set CE_UPDATE|CE_REMOVE outside checkout
522 	 * area as a result of ce_skip_worktree() shortcuts in
523 	 * verify_absent() and verify_uptodate().
524 	 * Make sure they don't modify worktree if they are already
525 	 * outside checkout area
526 	 */
527 	if (was_skip_worktree && ce_skip_worktree(ce)) {
528 		ce->ce_flags &= ~CE_UPDATE;
529 
530 		/*
531 		 * By default, when CE_REMOVE is on, CE_WT_REMOVE is also
532 		 * on to get that file removed from both index and worktree.
533 		 * If that file is already outside worktree area, don't
534 		 * bother remove it.
535 		 */
536 		if (ce->ce_flags & CE_REMOVE)
537 			ce->ce_flags &= ~CE_WT_REMOVE;
538 	}
539 
540 	if (!was_skip_worktree && ce_skip_worktree(ce)) {
541 		/*
542 		 * If CE_UPDATE is set, verify_uptodate() must be called already
543 		 * also stat info may have lost after merged_entry() so calling
544 		 * verify_uptodate() again may fail
545 		 */
546 		if (!(ce->ce_flags & CE_UPDATE) &&
547 		    verify_uptodate_sparse(ce, o)) {
548 			ce->ce_flags &= ~CE_SKIP_WORKTREE;
549 			return -1;
550 		}
551 		ce->ce_flags |= CE_WT_REMOVE;
552 		ce->ce_flags &= ~CE_UPDATE;
553 	}
554 	if (was_skip_worktree && !ce_skip_worktree(ce)) {
555 		if (verify_absent_sparse(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
556 			return -1;
557 		ce->ce_flags |= CE_UPDATE;
558 	}
559 	return 0;
560 }
561 
warn_conflicted_path(struct index_state * istate,int i,struct unpack_trees_options * o)562 static int warn_conflicted_path(struct index_state *istate,
563 				int i,
564 				struct unpack_trees_options *o)
565 {
566 	char *conflicting_path = istate->cache[i]->name;
567 	int count = 0;
568 
569 	add_rejected_path(o, WARNING_SPARSE_UNMERGED_FILE, conflicting_path);
570 
571 	/* Find out how many higher stage entries are at same path */
572 	while ((++count) + i < istate->cache_nr &&
573 	       !strcmp(conflicting_path, istate->cache[count + i]->name))
574 		; /* do nothing */
575 
576 	return count;
577 }
578 
call_unpack_fn(const struct cache_entry * const * src,struct unpack_trees_options * o)579 static inline int call_unpack_fn(const struct cache_entry * const *src,
580 				 struct unpack_trees_options *o)
581 {
582 	int ret = o->fn(src, o);
583 	if (ret > 0)
584 		ret = 0;
585 	return ret;
586 }
587 
mark_ce_used(struct cache_entry * ce,struct unpack_trees_options * o)588 static void mark_ce_used(struct cache_entry *ce, struct unpack_trees_options *o)
589 {
590 	ce->ce_flags |= CE_UNPACKED;
591 
592 	/*
593 	 * If this is a sparse directory, don't advance cache_bottom.
594 	 * That will be advanced later using the cache-tree data.
595 	 */
596 	if (S_ISSPARSEDIR(ce->ce_mode))
597 		return;
598 
599 	if (o->cache_bottom < o->src_index->cache_nr &&
600 	    o->src_index->cache[o->cache_bottom] == ce) {
601 		int bottom = o->cache_bottom;
602 		while (bottom < o->src_index->cache_nr &&
603 		       o->src_index->cache[bottom]->ce_flags & CE_UNPACKED)
604 			bottom++;
605 		o->cache_bottom = bottom;
606 	}
607 }
608 
mark_all_ce_unused(struct index_state * index)609 static void mark_all_ce_unused(struct index_state *index)
610 {
611 	int i;
612 	for (i = 0; i < index->cache_nr; i++)
613 		index->cache[i]->ce_flags &= ~(CE_UNPACKED | CE_ADDED | CE_NEW_SKIP_WORKTREE);
614 }
615 
locate_in_src_index(const struct cache_entry * ce,struct unpack_trees_options * o)616 static int locate_in_src_index(const struct cache_entry *ce,
617 			       struct unpack_trees_options *o)
618 {
619 	struct index_state *index = o->src_index;
620 	int len = ce_namelen(ce);
621 	int pos = index_name_pos(index, ce->name, len);
622 	if (pos < 0)
623 		pos = -1 - pos;
624 	return pos;
625 }
626 
627 /*
628  * We call unpack_index_entry() with an unmerged cache entry
629  * only in diff-index, and it wants a single callback.  Skip
630  * the other unmerged entry with the same name.
631  */
mark_ce_used_same_name(struct cache_entry * ce,struct unpack_trees_options * o)632 static void mark_ce_used_same_name(struct cache_entry *ce,
633 				   struct unpack_trees_options *o)
634 {
635 	struct index_state *index = o->src_index;
636 	int len = ce_namelen(ce);
637 	int pos;
638 
639 	for (pos = locate_in_src_index(ce, o); pos < index->cache_nr; pos++) {
640 		struct cache_entry *next = index->cache[pos];
641 		if (len != ce_namelen(next) ||
642 		    memcmp(ce->name, next->name, len))
643 			break;
644 		mark_ce_used(next, o);
645 	}
646 }
647 
next_cache_entry(struct unpack_trees_options * o)648 static struct cache_entry *next_cache_entry(struct unpack_trees_options *o)
649 {
650 	const struct index_state *index = o->src_index;
651 	int pos = o->cache_bottom;
652 
653 	while (pos < index->cache_nr) {
654 		struct cache_entry *ce = index->cache[pos];
655 		if (!(ce->ce_flags & CE_UNPACKED))
656 			return ce;
657 		pos++;
658 	}
659 	return NULL;
660 }
661 
add_same_unmerged(const struct cache_entry * ce,struct unpack_trees_options * o)662 static void add_same_unmerged(const struct cache_entry *ce,
663 			      struct unpack_trees_options *o)
664 {
665 	struct index_state *index = o->src_index;
666 	int len = ce_namelen(ce);
667 	int pos = index_name_pos(index, ce->name, len);
668 
669 	if (0 <= pos)
670 		die("programming error in a caller of mark_ce_used_same_name");
671 	for (pos = -pos - 1; pos < index->cache_nr; pos++) {
672 		struct cache_entry *next = index->cache[pos];
673 		if (len != ce_namelen(next) ||
674 		    memcmp(ce->name, next->name, len))
675 			break;
676 		add_entry(o, next, 0, 0);
677 		mark_ce_used(next, o);
678 	}
679 }
680 
unpack_index_entry(struct cache_entry * ce,struct unpack_trees_options * o)681 static int unpack_index_entry(struct cache_entry *ce,
682 			      struct unpack_trees_options *o)
683 {
684 	const struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
685 	int ret;
686 
687 	src[0] = ce;
688 
689 	mark_ce_used(ce, o);
690 	if (ce_stage(ce)) {
691 		if (o->skip_unmerged) {
692 			add_entry(o, ce, 0, 0);
693 			return 0;
694 		}
695 	}
696 	ret = call_unpack_fn(src, o);
697 	if (ce_stage(ce))
698 		mark_ce_used_same_name(ce, o);
699 	return ret;
700 }
701 
702 static int find_cache_pos(struct traverse_info *, const char *p, size_t len);
703 
restore_cache_bottom(struct traverse_info * info,int bottom)704 static void restore_cache_bottom(struct traverse_info *info, int bottom)
705 {
706 	struct unpack_trees_options *o = info->data;
707 
708 	if (o->diff_index_cached)
709 		return;
710 	o->cache_bottom = bottom;
711 }
712 
switch_cache_bottom(struct traverse_info * info)713 static int switch_cache_bottom(struct traverse_info *info)
714 {
715 	struct unpack_trees_options *o = info->data;
716 	int ret, pos;
717 
718 	if (o->diff_index_cached)
719 		return 0;
720 	ret = o->cache_bottom;
721 	pos = find_cache_pos(info->prev, info->name, info->namelen);
722 
723 	if (pos < -1)
724 		o->cache_bottom = -2 - pos;
725 	else if (pos < 0)
726 		o->cache_bottom = o->src_index->cache_nr;
727 	return ret;
728 }
729 
are_same_oid(struct name_entry * name_j,struct name_entry * name_k)730 static inline int are_same_oid(struct name_entry *name_j, struct name_entry *name_k)
731 {
732 	return !is_null_oid(&name_j->oid) && !is_null_oid(&name_k->oid) && oideq(&name_j->oid, &name_k->oid);
733 }
734 
all_trees_same_as_cache_tree(int n,unsigned long dirmask,struct name_entry * names,struct traverse_info * info)735 static int all_trees_same_as_cache_tree(int n, unsigned long dirmask,
736 					struct name_entry *names,
737 					struct traverse_info *info)
738 {
739 	struct unpack_trees_options *o = info->data;
740 	int i;
741 
742 	if (!o->merge || dirmask != ((1 << n) - 1))
743 		return 0;
744 
745 	for (i = 1; i < n; i++)
746 		if (!are_same_oid(names, names + i))
747 			return 0;
748 
749 	return cache_tree_matches_traversal(o->src_index->cache_tree, names, info);
750 }
751 
index_pos_by_traverse_info(struct name_entry * names,struct traverse_info * info)752 static int index_pos_by_traverse_info(struct name_entry *names,
753 				      struct traverse_info *info)
754 {
755 	struct unpack_trees_options *o = info->data;
756 	struct strbuf name = STRBUF_INIT;
757 	int pos;
758 
759 	strbuf_make_traverse_path(&name, info, names->path, names->pathlen);
760 	strbuf_addch(&name, '/');
761 	pos = index_name_pos(o->src_index, name.buf, name.len);
762 	if (pos >= 0) {
763 		if (!o->src_index->sparse_index ||
764 		    !(o->src_index->cache[pos]->ce_flags & CE_SKIP_WORKTREE))
765 			BUG("This is a directory and should not exist in index");
766 	} else {
767 		pos = -pos - 1;
768 	}
769 	if (pos >= o->src_index->cache_nr ||
770 	    !starts_with(o->src_index->cache[pos]->name, name.buf) ||
771 	    (pos > 0 && starts_with(o->src_index->cache[pos-1]->name, name.buf)))
772 		BUG("pos %d doesn't point to the first entry of %s in index",
773 		    pos, name.buf);
774 	strbuf_release(&name);
775 	return pos;
776 }
777 
778 /*
779  * Fast path if we detect that all trees are the same as cache-tree at this
780  * path. We'll walk these trees in an iterative loop using cache-tree/index
781  * instead of ODB since we already know what these trees contain.
782  */
traverse_by_cache_tree(int pos,int nr_entries,int nr_names,struct traverse_info * info)783 static int traverse_by_cache_tree(int pos, int nr_entries, int nr_names,
784 				  struct traverse_info *info)
785 {
786 	struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
787 	struct unpack_trees_options *o = info->data;
788 	struct cache_entry *tree_ce = NULL;
789 	int ce_len = 0;
790 	int i, d;
791 
792 	if (!o->merge)
793 		BUG("We need cache-tree to do this optimization");
794 
795 	/*
796 	 * Do what unpack_callback() and unpack_single_entry() normally
797 	 * do. But we walk all paths in an iterative loop instead.
798 	 *
799 	 * D/F conflicts and higher stage entries are not a concern
800 	 * because cache-tree would be invalidated and we would never
801 	 * get here in the first place.
802 	 */
803 	for (i = 0; i < nr_entries; i++) {
804 		int new_ce_len, len, rc;
805 
806 		src[0] = o->src_index->cache[pos + i];
807 
808 		len = ce_namelen(src[0]);
809 		new_ce_len = cache_entry_size(len);
810 
811 		if (new_ce_len > ce_len) {
812 			new_ce_len <<= 1;
813 			tree_ce = xrealloc(tree_ce, new_ce_len);
814 			memset(tree_ce, 0, new_ce_len);
815 			ce_len = new_ce_len;
816 
817 			tree_ce->ce_flags = create_ce_flags(0);
818 
819 			for (d = 1; d <= nr_names; d++)
820 				src[d] = tree_ce;
821 		}
822 
823 		tree_ce->ce_mode = src[0]->ce_mode;
824 		tree_ce->ce_namelen = len;
825 		oidcpy(&tree_ce->oid, &src[0]->oid);
826 		memcpy(tree_ce->name, src[0]->name, len + 1);
827 
828 		rc = call_unpack_fn((const struct cache_entry * const *)src, o);
829 		if (rc < 0) {
830 			free(tree_ce);
831 			return rc;
832 		}
833 
834 		mark_ce_used(src[0], o);
835 	}
836 	free(tree_ce);
837 	if (o->debug_unpack)
838 		printf("Unpacked %d entries from %s to %s using cache-tree\n",
839 		       nr_entries,
840 		       o->src_index->cache[pos]->name,
841 		       o->src_index->cache[pos + nr_entries - 1]->name);
842 	return 0;
843 }
844 
traverse_trees_recursive(int n,unsigned long dirmask,unsigned long df_conflicts,struct name_entry * names,struct traverse_info * info)845 static int traverse_trees_recursive(int n, unsigned long dirmask,
846 				    unsigned long df_conflicts,
847 				    struct name_entry *names,
848 				    struct traverse_info *info)
849 {
850 	struct unpack_trees_options *o = info->data;
851 	int i, ret, bottom;
852 	int nr_buf = 0;
853 	struct tree_desc t[MAX_UNPACK_TREES];
854 	void *buf[MAX_UNPACK_TREES];
855 	struct traverse_info newinfo;
856 	struct name_entry *p;
857 	int nr_entries;
858 
859 	nr_entries = all_trees_same_as_cache_tree(n, dirmask, names, info);
860 	if (nr_entries > 0) {
861 		int pos = index_pos_by_traverse_info(names, info);
862 
863 		if (!o->merge || df_conflicts)
864 			BUG("Wrong condition to get here buddy");
865 
866 		/*
867 		 * All entries up to 'pos' must have been processed
868 		 * (i.e. marked CE_UNPACKED) at this point. But to be safe,
869 		 * save and restore cache_bottom anyway to not miss
870 		 * unprocessed entries before 'pos'.
871 		 */
872 		bottom = o->cache_bottom;
873 		ret = traverse_by_cache_tree(pos, nr_entries, n, info);
874 		o->cache_bottom = bottom;
875 		return ret;
876 	}
877 
878 	p = names;
879 	while (!p->mode)
880 		p++;
881 
882 	newinfo = *info;
883 	newinfo.prev = info;
884 	newinfo.pathspec = info->pathspec;
885 	newinfo.name = p->path;
886 	newinfo.namelen = p->pathlen;
887 	newinfo.mode = p->mode;
888 	newinfo.pathlen = st_add3(newinfo.pathlen, tree_entry_len(p), 1);
889 	newinfo.df_conflicts |= df_conflicts;
890 
891 	/*
892 	 * Fetch the tree from the ODB for each peer directory in the
893 	 * n commits.
894 	 *
895 	 * For 2- and 3-way traversals, we try to avoid hitting the
896 	 * ODB twice for the same OID.  This should yield a nice speed
897 	 * up in checkouts and merges when the commits are similar.
898 	 *
899 	 * We don't bother doing the full O(n^2) search for larger n,
900 	 * because wider traversals don't happen that often and we
901 	 * avoid the search setup.
902 	 *
903 	 * When 2 peer OIDs are the same, we just copy the tree
904 	 * descriptor data.  This implicitly borrows the buffer
905 	 * data from the earlier cell.
906 	 */
907 	for (i = 0; i < n; i++, dirmask >>= 1) {
908 		if (i > 0 && are_same_oid(&names[i], &names[i - 1]))
909 			t[i] = t[i - 1];
910 		else if (i > 1 && are_same_oid(&names[i], &names[i - 2]))
911 			t[i] = t[i - 2];
912 		else {
913 			const struct object_id *oid = NULL;
914 			if (dirmask & 1)
915 				oid = &names[i].oid;
916 			buf[nr_buf++] = fill_tree_descriptor(the_repository, t + i, oid);
917 		}
918 	}
919 
920 	bottom = switch_cache_bottom(&newinfo);
921 	ret = traverse_trees(o->src_index, n, t, &newinfo);
922 	restore_cache_bottom(&newinfo, bottom);
923 
924 	for (i = 0; i < nr_buf; i++)
925 		free(buf[i]);
926 
927 	return ret;
928 }
929 
930 /*
931  * Compare the traverse-path to the cache entry without actually
932  * having to generate the textual representation of the traverse
933  * path.
934  *
935  * NOTE! This *only* compares up to the size of the traverse path
936  * itself - the caller needs to do the final check for the cache
937  * entry having more data at the end!
938  */
do_compare_entry_piecewise(const struct cache_entry * ce,const struct traverse_info * info,const char * name,size_t namelen,unsigned mode)939 static int do_compare_entry_piecewise(const struct cache_entry *ce,
940 				      const struct traverse_info *info,
941 				      const char *name, size_t namelen,
942 				      unsigned mode)
943 {
944 	int pathlen, ce_len;
945 	const char *ce_name;
946 
947 	if (info->prev) {
948 		int cmp = do_compare_entry_piecewise(ce, info->prev,
949 						     info->name, info->namelen,
950 						     info->mode);
951 		if (cmp)
952 			return cmp;
953 	}
954 	pathlen = info->pathlen;
955 	ce_len = ce_namelen(ce);
956 
957 	/* If ce_len < pathlen then we must have previously hit "name == directory" entry */
958 	if (ce_len < pathlen)
959 		return -1;
960 
961 	ce_len -= pathlen;
962 	ce_name = ce->name + pathlen;
963 
964 	return df_name_compare(ce_name, ce_len, S_IFREG, name, namelen, mode);
965 }
966 
do_compare_entry(const struct cache_entry * ce,const struct traverse_info * info,const char * name,size_t namelen,unsigned mode)967 static int do_compare_entry(const struct cache_entry *ce,
968 			    const struct traverse_info *info,
969 			    const char *name, size_t namelen,
970 			    unsigned mode)
971 {
972 	int pathlen, ce_len;
973 	const char *ce_name;
974 	int cmp;
975 	unsigned ce_mode;
976 
977 	/*
978 	 * If we have not precomputed the traverse path, it is quicker
979 	 * to avoid doing so.  But if we have precomputed it,
980 	 * it is quicker to use the precomputed version.
981 	 */
982 	if (!info->traverse_path)
983 		return do_compare_entry_piecewise(ce, info, name, namelen, mode);
984 
985 	cmp = strncmp(ce->name, info->traverse_path, info->pathlen);
986 	if (cmp)
987 		return cmp;
988 
989 	pathlen = info->pathlen;
990 	ce_len = ce_namelen(ce);
991 
992 	if (ce_len < pathlen)
993 		return -1;
994 
995 	ce_len -= pathlen;
996 	ce_name = ce->name + pathlen;
997 
998 	ce_mode = S_ISSPARSEDIR(ce->ce_mode) ? S_IFDIR : S_IFREG;
999 	return df_name_compare(ce_name, ce_len, ce_mode, name, namelen, mode);
1000 }
1001 
compare_entry(const struct cache_entry * ce,const struct traverse_info * info,const struct name_entry * n)1002 static int compare_entry(const struct cache_entry *ce, const struct traverse_info *info, const struct name_entry *n)
1003 {
1004 	int cmp = do_compare_entry(ce, info, n->path, n->pathlen, n->mode);
1005 	if (cmp)
1006 		return cmp;
1007 
1008 	/*
1009 	 * At this point, we know that we have a prefix match. If ce
1010 	 * is a sparse directory, then allow an exact match. This only
1011 	 * works when the input name is a directory, since ce->name
1012 	 * ends in a directory separator.
1013 	 */
1014 	if (S_ISSPARSEDIR(ce->ce_mode) &&
1015 	    ce->ce_namelen == traverse_path_len(info, tree_entry_len(n)) + 1)
1016 		return 0;
1017 
1018 	/*
1019 	 * Even if the beginning compared identically, the ce should
1020 	 * compare as bigger than a directory leading up to it!
1021 	 */
1022 	return ce_namelen(ce) > traverse_path_len(info, tree_entry_len(n));
1023 }
1024 
ce_in_traverse_path(const struct cache_entry * ce,const struct traverse_info * info)1025 static int ce_in_traverse_path(const struct cache_entry *ce,
1026 			       const struct traverse_info *info)
1027 {
1028 	if (!info->prev)
1029 		return 1;
1030 	if (do_compare_entry(ce, info->prev,
1031 			     info->name, info->namelen, info->mode))
1032 		return 0;
1033 	/*
1034 	 * If ce (blob) is the same name as the path (which is a tree
1035 	 * we will be descending into), it won't be inside it.
1036 	 */
1037 	return (info->pathlen < ce_namelen(ce));
1038 }
1039 
create_ce_entry(const struct traverse_info * info,const struct name_entry * n,int stage,struct index_state * istate,int is_transient,int is_sparse_directory)1040 static struct cache_entry *create_ce_entry(const struct traverse_info *info,
1041 	const struct name_entry *n,
1042 	int stage,
1043 	struct index_state *istate,
1044 	int is_transient,
1045 	int is_sparse_directory)
1046 {
1047 	size_t len = traverse_path_len(info, tree_entry_len(n));
1048 	size_t alloc_len = is_sparse_directory ? len + 1 : len;
1049 	struct cache_entry *ce =
1050 		is_transient ?
1051 		make_empty_transient_cache_entry(alloc_len, NULL) :
1052 		make_empty_cache_entry(istate, alloc_len);
1053 
1054 	ce->ce_mode = create_ce_mode(n->mode);
1055 	ce->ce_flags = create_ce_flags(stage);
1056 	ce->ce_namelen = len;
1057 	oidcpy(&ce->oid, &n->oid);
1058 	/* len+1 because the cache_entry allocates space for NUL */
1059 	make_traverse_path(ce->name, len + 1, info, n->path, n->pathlen);
1060 
1061 	if (is_sparse_directory) {
1062 		ce->name[len] = '/';
1063 		ce->name[len + 1] = '\0';
1064 		ce->ce_namelen++;
1065 		ce->ce_flags |= CE_SKIP_WORKTREE;
1066 	}
1067 
1068 	return ce;
1069 }
1070 
1071 /*
1072  * Note that traverse_by_cache_tree() duplicates some logic in this function
1073  * without actually calling it. If you change the logic here you may need to
1074  * check and change there as well.
1075  */
unpack_single_entry(int n,unsigned long mask,unsigned long dirmask,struct cache_entry ** src,const struct name_entry * names,const struct traverse_info * info)1076 static int unpack_single_entry(int n, unsigned long mask,
1077 			       unsigned long dirmask,
1078 			       struct cache_entry **src,
1079 			       const struct name_entry *names,
1080 			       const struct traverse_info *info)
1081 {
1082 	int i;
1083 	struct unpack_trees_options *o = info->data;
1084 	unsigned long conflicts = info->df_conflicts | dirmask;
1085 
1086 	if (mask == dirmask && !src[0])
1087 		return 0;
1088 
1089 	/*
1090 	 * When we have a sparse directory entry for src[0],
1091 	 * then this isn't necessarily a directory-file conflict.
1092 	 */
1093 	if (mask == dirmask && src[0] &&
1094 	    S_ISSPARSEDIR(src[0]->ce_mode))
1095 		conflicts = 0;
1096 
1097 	/*
1098 	 * Ok, we've filled in up to any potential index entry in src[0],
1099 	 * now do the rest.
1100 	 */
1101 	for (i = 0; i < n; i++) {
1102 		int stage;
1103 		unsigned int bit = 1ul << i;
1104 		if (conflicts & bit) {
1105 			src[i + o->merge] = o->df_conflict_entry;
1106 			continue;
1107 		}
1108 		if (!(mask & bit))
1109 			continue;
1110 		if (!o->merge)
1111 			stage = 0;
1112 		else if (i + 1 < o->head_idx)
1113 			stage = 1;
1114 		else if (i + 1 > o->head_idx)
1115 			stage = 3;
1116 		else
1117 			stage = 2;
1118 
1119 		/*
1120 		 * If the merge bit is set, then the cache entries are
1121 		 * discarded in the following block.  In this case,
1122 		 * construct "transient" cache_entries, as they are
1123 		 * not stored in the index.  otherwise construct the
1124 		 * cache entry from the index aware logic.
1125 		 */
1126 		src[i + o->merge] = create_ce_entry(info, names + i, stage,
1127 						    &o->result, o->merge,
1128 						    bit & dirmask);
1129 	}
1130 
1131 	if (o->merge) {
1132 		int rc = call_unpack_fn((const struct cache_entry * const *)src,
1133 					o);
1134 		for (i = 0; i < n; i++) {
1135 			struct cache_entry *ce = src[i + o->merge];
1136 			if (ce != o->df_conflict_entry)
1137 				discard_cache_entry(ce);
1138 		}
1139 		return rc;
1140 	}
1141 
1142 	for (i = 0; i < n; i++)
1143 		if (src[i] && src[i] != o->df_conflict_entry)
1144 			if (do_add_entry(o, src[i], 0, 0))
1145 				return -1;
1146 
1147 	return 0;
1148 }
1149 
unpack_failed(struct unpack_trees_options * o,const char * message)1150 static int unpack_failed(struct unpack_trees_options *o, const char *message)
1151 {
1152 	discard_index(&o->result);
1153 	if (!o->quiet && !o->exiting_early) {
1154 		if (message)
1155 			return error("%s", message);
1156 		return -1;
1157 	}
1158 	return -1;
1159 }
1160 
1161 /*
1162  * The tree traversal is looking at name p.  If we have a matching entry,
1163  * return it.  If name p is a directory in the index, do not return
1164  * anything, as we will want to match it when the traversal descends into
1165  * the directory.
1166  */
find_cache_pos(struct traverse_info * info,const char * p,size_t p_len)1167 static int find_cache_pos(struct traverse_info *info,
1168 			  const char *p, size_t p_len)
1169 {
1170 	int pos;
1171 	struct unpack_trees_options *o = info->data;
1172 	struct index_state *index = o->src_index;
1173 	int pfxlen = info->pathlen;
1174 
1175 	for (pos = o->cache_bottom; pos < index->cache_nr; pos++) {
1176 		const struct cache_entry *ce = index->cache[pos];
1177 		const char *ce_name, *ce_slash;
1178 		int cmp, ce_len;
1179 
1180 		if (ce->ce_flags & CE_UNPACKED) {
1181 			/*
1182 			 * cache_bottom entry is already unpacked, so
1183 			 * we can never match it; don't check it
1184 			 * again.
1185 			 */
1186 			if (pos == o->cache_bottom)
1187 				++o->cache_bottom;
1188 			continue;
1189 		}
1190 		if (!ce_in_traverse_path(ce, info)) {
1191 			/*
1192 			 * Check if we can skip future cache checks
1193 			 * (because we're already past all possible
1194 			 * entries in the traverse path).
1195 			 */
1196 			if (info->traverse_path) {
1197 				if (strncmp(ce->name, info->traverse_path,
1198 					    info->pathlen) > 0)
1199 					break;
1200 			}
1201 			continue;
1202 		}
1203 		ce_name = ce->name + pfxlen;
1204 		ce_slash = strchr(ce_name, '/');
1205 		if (ce_slash)
1206 			ce_len = ce_slash - ce_name;
1207 		else
1208 			ce_len = ce_namelen(ce) - pfxlen;
1209 		cmp = name_compare(p, p_len, ce_name, ce_len);
1210 		/*
1211 		 * Exact match; if we have a directory we need to
1212 		 * delay returning it.
1213 		 */
1214 		if (!cmp)
1215 			return ce_slash ? -2 - pos : pos;
1216 		if (0 < cmp)
1217 			continue; /* keep looking */
1218 		/*
1219 		 * ce_name sorts after p->path; could it be that we
1220 		 * have files under p->path directory in the index?
1221 		 * E.g.  ce_name == "t-i", and p->path == "t"; we may
1222 		 * have "t/a" in the index.
1223 		 */
1224 		if (p_len < ce_len && !memcmp(ce_name, p, p_len) &&
1225 		    ce_name[p_len] < '/')
1226 			continue; /* keep looking */
1227 		break;
1228 	}
1229 	return -1;
1230 }
1231 
1232 /*
1233  * Given a sparse directory entry 'ce', compare ce->name to
1234  * info->name + '/' + p->path + '/' if info->name is non-empty.
1235  * Compare ce->name to p->path + '/' otherwise. Note that
1236  * ce->name must end in a trailing '/' because it is a sparse
1237  * directory entry.
1238  */
sparse_dir_matches_path(const struct cache_entry * ce,struct traverse_info * info,const struct name_entry * p)1239 static int sparse_dir_matches_path(const struct cache_entry *ce,
1240 				   struct traverse_info *info,
1241 				   const struct name_entry *p)
1242 {
1243 	assert(S_ISSPARSEDIR(ce->ce_mode));
1244 	assert(ce->name[ce->ce_namelen - 1] == '/');
1245 
1246 	if (info->namelen)
1247 		return ce->ce_namelen == info->namelen + p->pathlen + 2 &&
1248 		       ce->name[info->namelen] == '/' &&
1249 		       !strncmp(ce->name, info->name, info->namelen) &&
1250 		       !strncmp(ce->name + info->namelen + 1, p->path, p->pathlen);
1251 	return ce->ce_namelen == p->pathlen + 1 &&
1252 	       !strncmp(ce->name, p->path, p->pathlen);
1253 }
1254 
find_cache_entry(struct traverse_info * info,const struct name_entry * p)1255 static struct cache_entry *find_cache_entry(struct traverse_info *info,
1256 					    const struct name_entry *p)
1257 {
1258 	const char *path;
1259 	int pos = find_cache_pos(info, p->path, p->pathlen);
1260 	struct unpack_trees_options *o = info->data;
1261 
1262 	if (0 <= pos)
1263 		return o->src_index->cache[pos];
1264 
1265 	/*
1266 	 * Check for a sparse-directory entry named "path/".
1267 	 * Due to the input p->path not having a trailing
1268 	 * slash, the negative 'pos' value overshoots the
1269 	 * expected position, hence "-2" instead of "-1".
1270 	 */
1271 	pos = -pos - 2;
1272 
1273 	if (pos < 0 || pos >= o->src_index->cache_nr)
1274 		return NULL;
1275 
1276 	/*
1277 	 * Due to lexicographic sorting and sparse directory
1278 	 * entries ending with a trailing slash, our path as a
1279 	 * sparse directory (e.g "subdir/") and	our path as a
1280 	 * file (e.g. "subdir") might be separated by other
1281 	 * paths (e.g. "subdir-").
1282 	 */
1283 	while (pos >= 0) {
1284 		struct cache_entry *ce = o->src_index->cache[pos];
1285 
1286 		if (!skip_prefix(ce->name, info->traverse_path, &path) ||
1287 		    strncmp(path, p->path, p->pathlen) ||
1288 		    path[p->pathlen] != '/')
1289 			return NULL;
1290 
1291 		if (S_ISSPARSEDIR(ce->ce_mode) &&
1292 		    sparse_dir_matches_path(ce, info, p))
1293 			return ce;
1294 
1295 		pos--;
1296 	}
1297 
1298 	return NULL;
1299 }
1300 
debug_path(struct traverse_info * info)1301 static void debug_path(struct traverse_info *info)
1302 {
1303 	if (info->prev) {
1304 		debug_path(info->prev);
1305 		if (*info->prev->name)
1306 			putchar('/');
1307 	}
1308 	printf("%s", info->name);
1309 }
1310 
debug_name_entry(int i,struct name_entry * n)1311 static void debug_name_entry(int i, struct name_entry *n)
1312 {
1313 	printf("ent#%d %06o %s\n", i,
1314 	       n->path ? n->mode : 0,
1315 	       n->path ? n->path : "(missing)");
1316 }
1317 
debug_unpack_callback(int n,unsigned long mask,unsigned long dirmask,struct name_entry * names,struct traverse_info * info)1318 static void debug_unpack_callback(int n,
1319 				  unsigned long mask,
1320 				  unsigned long dirmask,
1321 				  struct name_entry *names,
1322 				  struct traverse_info *info)
1323 {
1324 	int i;
1325 	printf("* unpack mask %lu, dirmask %lu, cnt %d ",
1326 	       mask, dirmask, n);
1327 	debug_path(info);
1328 	putchar('\n');
1329 	for (i = 0; i < n; i++)
1330 		debug_name_entry(i, names + i);
1331 }
1332 
1333 /*
1334  * Returns true if and only if the given cache_entry is a
1335  * sparse-directory entry that matches the given name_entry
1336  * from the tree walk at the given traverse_info.
1337  */
is_sparse_directory_entry(struct cache_entry * ce,struct name_entry * name,struct traverse_info * info)1338 static int is_sparse_directory_entry(struct cache_entry *ce,
1339 				     struct name_entry *name,
1340 				     struct traverse_info *info)
1341 {
1342 	if (!ce || !name || !S_ISSPARSEDIR(ce->ce_mode))
1343 		return 0;
1344 
1345 	return sparse_dir_matches_path(ce, info, name);
1346 }
1347 
1348 /*
1349  * Note that traverse_by_cache_tree() duplicates some logic in this function
1350  * without actually calling it. If you change the logic here you may need to
1351  * check and change there as well.
1352  */
unpack_callback(int n,unsigned long mask,unsigned long dirmask,struct name_entry * names,struct traverse_info * info)1353 static int unpack_callback(int n, unsigned long mask, unsigned long dirmask, struct name_entry *names, struct traverse_info *info)
1354 {
1355 	struct cache_entry *src[MAX_UNPACK_TREES + 1] = { NULL, };
1356 	struct unpack_trees_options *o = info->data;
1357 	const struct name_entry *p = names;
1358 
1359 	/* Find first entry with a real name (we could use "mask" too) */
1360 	while (!p->mode)
1361 		p++;
1362 
1363 	if (o->debug_unpack)
1364 		debug_unpack_callback(n, mask, dirmask, names, info);
1365 
1366 	/* Are we supposed to look at the index too? */
1367 	if (o->merge) {
1368 		while (1) {
1369 			int cmp;
1370 			struct cache_entry *ce;
1371 
1372 			if (o->diff_index_cached)
1373 				ce = next_cache_entry(o);
1374 			else
1375 				ce = find_cache_entry(info, p);
1376 
1377 			if (!ce)
1378 				break;
1379 			cmp = compare_entry(ce, info, p);
1380 			if (cmp < 0) {
1381 				if (unpack_index_entry(ce, o) < 0)
1382 					return unpack_failed(o, NULL);
1383 				continue;
1384 			}
1385 			if (!cmp) {
1386 				if (ce_stage(ce)) {
1387 					/*
1388 					 * If we skip unmerged index
1389 					 * entries, we'll skip this
1390 					 * entry *and* the tree
1391 					 * entries associated with it!
1392 					 */
1393 					if (o->skip_unmerged) {
1394 						add_same_unmerged(ce, o);
1395 						return mask;
1396 					}
1397 				}
1398 				src[0] = ce;
1399 			}
1400 			break;
1401 		}
1402 	}
1403 
1404 	if (unpack_single_entry(n, mask, dirmask, src, names, info) < 0)
1405 		return -1;
1406 
1407 	if (o->merge && src[0]) {
1408 		if (ce_stage(src[0]))
1409 			mark_ce_used_same_name(src[0], o);
1410 		else
1411 			mark_ce_used(src[0], o);
1412 	}
1413 
1414 	/* Now handle any directories.. */
1415 	if (dirmask) {
1416 		/* special case: "diff-index --cached" looking at a tree */
1417 		if (o->diff_index_cached &&
1418 		    n == 1 && dirmask == 1 && S_ISDIR(names->mode)) {
1419 			int matches;
1420 			matches = cache_tree_matches_traversal(o->src_index->cache_tree,
1421 							       names, info);
1422 			/*
1423 			 * Everything under the name matches; skip the
1424 			 * entire hierarchy.  diff_index_cached codepath
1425 			 * special cases D/F conflicts in such a way that
1426 			 * it does not do any look-ahead, so this is safe.
1427 			 */
1428 			if (matches) {
1429 				o->cache_bottom += matches;
1430 				return mask;
1431 			}
1432 		}
1433 
1434 		if (!is_sparse_directory_entry(src[0], names, info) &&
1435 		    traverse_trees_recursive(n, dirmask, mask & ~dirmask,
1436 						    names, info) < 0) {
1437 			return -1;
1438 		}
1439 
1440 		return mask;
1441 	}
1442 
1443 	return mask;
1444 }
1445 
1446 static int clear_ce_flags_1(struct index_state *istate,
1447 			    struct cache_entry **cache, int nr,
1448 			    struct strbuf *prefix,
1449 			    int select_mask, int clear_mask,
1450 			    struct pattern_list *pl,
1451 			    enum pattern_match_result default_match,
1452 			    int progress_nr);
1453 
1454 /* Whole directory matching */
clear_ce_flags_dir(struct index_state * istate,struct cache_entry ** cache,int nr,struct strbuf * prefix,char * basename,int select_mask,int clear_mask,struct pattern_list * pl,enum pattern_match_result default_match,int progress_nr)1455 static int clear_ce_flags_dir(struct index_state *istate,
1456 			      struct cache_entry **cache, int nr,
1457 			      struct strbuf *prefix,
1458 			      char *basename,
1459 			      int select_mask, int clear_mask,
1460 			      struct pattern_list *pl,
1461 			      enum pattern_match_result default_match,
1462 			      int progress_nr)
1463 {
1464 	struct cache_entry **cache_end;
1465 	int dtype = DT_DIR;
1466 	int rc;
1467 	enum pattern_match_result ret, orig_ret;
1468 	orig_ret = path_matches_pattern_list(prefix->buf, prefix->len,
1469 					     basename, &dtype, pl, istate);
1470 
1471 	strbuf_addch(prefix, '/');
1472 
1473 	/* If undecided, use matching result of parent dir in defval */
1474 	if (orig_ret == UNDECIDED)
1475 		ret = default_match;
1476 	else
1477 		ret = orig_ret;
1478 
1479 	for (cache_end = cache; cache_end != cache + nr; cache_end++) {
1480 		struct cache_entry *ce = *cache_end;
1481 		if (strncmp(ce->name, prefix->buf, prefix->len))
1482 			break;
1483 	}
1484 
1485 	if (pl->use_cone_patterns && orig_ret == MATCHED_RECURSIVE) {
1486 		struct cache_entry **ce = cache;
1487 		rc = cache_end - cache;
1488 
1489 		while (ce < cache_end) {
1490 			(*ce)->ce_flags &= ~clear_mask;
1491 			ce++;
1492 		}
1493 	} else if (pl->use_cone_patterns && orig_ret == NOT_MATCHED) {
1494 		rc = cache_end - cache;
1495 	} else {
1496 		rc = clear_ce_flags_1(istate, cache, cache_end - cache,
1497 				      prefix,
1498 				      select_mask, clear_mask,
1499 				      pl, ret,
1500 				      progress_nr);
1501 	}
1502 
1503 	strbuf_setlen(prefix, prefix->len - 1);
1504 	return rc;
1505 }
1506 
1507 /*
1508  * Traverse the index, find every entry that matches according to
1509  * o->pl. Do "ce_flags &= ~clear_mask" on those entries. Return the
1510  * number of traversed entries.
1511  *
1512  * If select_mask is non-zero, only entries whose ce_flags has on of
1513  * those bits enabled are traversed.
1514  *
1515  * cache	: pointer to an index entry
1516  * prefix_len	: an offset to its path
1517  *
1518  * The current path ("prefix") including the trailing '/' is
1519  *   cache[0]->name[0..(prefix_len-1)]
1520  * Top level path has prefix_len zero.
1521  */
clear_ce_flags_1(struct index_state * istate,struct cache_entry ** cache,int nr,struct strbuf * prefix,int select_mask,int clear_mask,struct pattern_list * pl,enum pattern_match_result default_match,int progress_nr)1522 static int clear_ce_flags_1(struct index_state *istate,
1523 			    struct cache_entry **cache, int nr,
1524 			    struct strbuf *prefix,
1525 			    int select_mask, int clear_mask,
1526 			    struct pattern_list *pl,
1527 			    enum pattern_match_result default_match,
1528 			    int progress_nr)
1529 {
1530 	struct cache_entry **cache_end = nr ? cache + nr : cache;
1531 
1532 	/*
1533 	 * Process all entries that have the given prefix and meet
1534 	 * select_mask condition
1535 	 */
1536 	while(cache != cache_end) {
1537 		struct cache_entry *ce = *cache;
1538 		const char *name, *slash;
1539 		int len, dtype;
1540 		enum pattern_match_result ret;
1541 
1542 		display_progress(istate->progress, progress_nr);
1543 
1544 		if (select_mask && !(ce->ce_flags & select_mask)) {
1545 			cache++;
1546 			progress_nr++;
1547 			continue;
1548 		}
1549 
1550 		if (prefix->len && strncmp(ce->name, prefix->buf, prefix->len))
1551 			break;
1552 
1553 		name = ce->name + prefix->len;
1554 		slash = strchr(name, '/');
1555 
1556 		/* If it's a directory, try whole directory match first */
1557 		if (slash) {
1558 			int processed;
1559 
1560 			len = slash - name;
1561 			strbuf_add(prefix, name, len);
1562 
1563 			processed = clear_ce_flags_dir(istate, cache, cache_end - cache,
1564 						       prefix,
1565 						       prefix->buf + prefix->len - len,
1566 						       select_mask, clear_mask,
1567 						       pl, default_match,
1568 						       progress_nr);
1569 
1570 			/* clear_c_f_dir eats a whole dir already? */
1571 			if (processed) {
1572 				cache += processed;
1573 				progress_nr += processed;
1574 				strbuf_setlen(prefix, prefix->len - len);
1575 				continue;
1576 			}
1577 
1578 			strbuf_addch(prefix, '/');
1579 			processed = clear_ce_flags_1(istate, cache, cache_end - cache,
1580 						     prefix,
1581 						     select_mask, clear_mask, pl,
1582 						     default_match, progress_nr);
1583 
1584 			cache += processed;
1585 			progress_nr += processed;
1586 
1587 			strbuf_setlen(prefix, prefix->len - len - 1);
1588 			continue;
1589 		}
1590 
1591 		/* Non-directory */
1592 		dtype = ce_to_dtype(ce);
1593 		ret = path_matches_pattern_list(ce->name,
1594 						ce_namelen(ce),
1595 						name, &dtype, pl, istate);
1596 		if (ret == UNDECIDED)
1597 			ret = default_match;
1598 		if (ret == MATCHED || ret == MATCHED_RECURSIVE)
1599 			ce->ce_flags &= ~clear_mask;
1600 		cache++;
1601 		progress_nr++;
1602 	}
1603 
1604 	display_progress(istate->progress, progress_nr);
1605 	return nr - (cache_end - cache);
1606 }
1607 
clear_ce_flags(struct index_state * istate,int select_mask,int clear_mask,struct pattern_list * pl,int show_progress)1608 static int clear_ce_flags(struct index_state *istate,
1609 			  int select_mask, int clear_mask,
1610 			  struct pattern_list *pl,
1611 			  int show_progress)
1612 {
1613 	static struct strbuf prefix = STRBUF_INIT;
1614 	char label[100];
1615 	int rval;
1616 
1617 	strbuf_reset(&prefix);
1618 	if (show_progress)
1619 		istate->progress = start_delayed_progress(
1620 					_("Updating index flags"),
1621 					istate->cache_nr);
1622 
1623 	xsnprintf(label, sizeof(label), "clear_ce_flags(0x%08lx,0x%08lx)",
1624 		  (unsigned long)select_mask, (unsigned long)clear_mask);
1625 	trace2_region_enter("unpack_trees", label, the_repository);
1626 	rval = clear_ce_flags_1(istate,
1627 				istate->cache,
1628 				istate->cache_nr,
1629 				&prefix,
1630 				select_mask, clear_mask,
1631 				pl, 0, 0);
1632 	trace2_region_leave("unpack_trees", label, the_repository);
1633 
1634 	stop_progress(&istate->progress);
1635 	return rval;
1636 }
1637 
1638 /*
1639  * Set/Clear CE_NEW_SKIP_WORKTREE according to $GIT_DIR/info/sparse-checkout
1640  */
mark_new_skip_worktree(struct pattern_list * pl,struct index_state * istate,int select_flag,int skip_wt_flag,int show_progress)1641 static void mark_new_skip_worktree(struct pattern_list *pl,
1642 				   struct index_state *istate,
1643 				   int select_flag, int skip_wt_flag,
1644 				   int show_progress)
1645 {
1646 	int i;
1647 
1648 	/*
1649 	 * 1. Pretend the narrowest worktree: only unmerged entries
1650 	 * are checked out
1651 	 */
1652 	for (i = 0; i < istate->cache_nr; i++) {
1653 		struct cache_entry *ce = istate->cache[i];
1654 
1655 		if (select_flag && !(ce->ce_flags & select_flag))
1656 			continue;
1657 
1658 		if (!ce_stage(ce) && !(ce->ce_flags & CE_CONFLICTED))
1659 			ce->ce_flags |= skip_wt_flag;
1660 		else
1661 			ce->ce_flags &= ~skip_wt_flag;
1662 	}
1663 
1664 	/*
1665 	 * 2. Widen worktree according to sparse-checkout file.
1666 	 * Matched entries will have skip_wt_flag cleared (i.e. "in")
1667 	 */
1668 	clear_ce_flags(istate, select_flag, skip_wt_flag, pl, show_progress);
1669 }
1670 
populate_from_existing_patterns(struct unpack_trees_options * o,struct pattern_list * pl)1671 static void populate_from_existing_patterns(struct unpack_trees_options *o,
1672 					    struct pattern_list *pl)
1673 {
1674 	if (get_sparse_checkout_patterns(pl) < 0)
1675 		o->skip_sparse_checkout = 1;
1676 	else
1677 		o->pl = pl;
1678 }
1679 
1680 
1681 static int verify_absent(const struct cache_entry *,
1682 			 enum unpack_trees_error_types,
1683 			 struct unpack_trees_options *);
1684 /*
1685  * N-way merge "len" trees.  Returns 0 on success, -1 on failure to manipulate the
1686  * resulting index, -2 on failure to reflect the changes to the work tree.
1687  *
1688  * CE_ADDED, CE_UNPACKED and CE_NEW_SKIP_WORKTREE are used internally
1689  */
unpack_trees(unsigned len,struct tree_desc * t,struct unpack_trees_options * o)1690 int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options *o)
1691 {
1692 	struct repository *repo = the_repository;
1693 	int i, ret;
1694 	static struct cache_entry *dfc;
1695 	struct pattern_list pl;
1696 	int free_pattern_list = 0;
1697 	struct dir_struct dir = DIR_INIT;
1698 
1699 	if (o->reset == UNPACK_RESET_INVALID)
1700 		BUG("o->reset had a value of 1; should be UNPACK_TREES_*_UNTRACKED");
1701 
1702 	if (len > MAX_UNPACK_TREES)
1703 		die("unpack_trees takes at most %d trees", MAX_UNPACK_TREES);
1704 	if (o->dir)
1705 		BUG("o->dir is for internal use only");
1706 
1707 	trace_performance_enter();
1708 	trace2_region_enter("unpack_trees", "unpack_trees", the_repository);
1709 
1710 	prepare_repo_settings(repo);
1711 	if (repo->settings.command_requires_full_index) {
1712 		ensure_full_index(o->src_index);
1713 		ensure_full_index(o->dst_index);
1714 	}
1715 
1716 	if (o->reset == UNPACK_RESET_OVERWRITE_UNTRACKED &&
1717 	    o->preserve_ignored)
1718 		BUG("UNPACK_RESET_OVERWRITE_UNTRACKED incompatible with preserved ignored files");
1719 
1720 	if (!o->preserve_ignored) {
1721 		o->dir = &dir;
1722 		o->dir->flags |= DIR_SHOW_IGNORED;
1723 		setup_standard_excludes(o->dir);
1724 	}
1725 
1726 	if (!core_apply_sparse_checkout || !o->update)
1727 		o->skip_sparse_checkout = 1;
1728 	if (!o->skip_sparse_checkout && !o->pl) {
1729 		memset(&pl, 0, sizeof(pl));
1730 		free_pattern_list = 1;
1731 		populate_from_existing_patterns(o, &pl);
1732 	}
1733 
1734 	memset(&o->result, 0, sizeof(o->result));
1735 	o->result.initialized = 1;
1736 	o->result.timestamp.sec = o->src_index->timestamp.sec;
1737 	o->result.timestamp.nsec = o->src_index->timestamp.nsec;
1738 	o->result.version = o->src_index->version;
1739 	if (!o->src_index->split_index) {
1740 		o->result.split_index = NULL;
1741 	} else if (o->src_index == o->dst_index) {
1742 		/*
1743 		 * o->dst_index (and thus o->src_index) will be discarded
1744 		 * and overwritten with o->result at the end of this function,
1745 		 * so just use src_index's split_index to avoid having to
1746 		 * create a new one.
1747 		 */
1748 		o->result.split_index = o->src_index->split_index;
1749 		o->result.split_index->refcount++;
1750 	} else {
1751 		o->result.split_index = init_split_index(&o->result);
1752 	}
1753 	oidcpy(&o->result.oid, &o->src_index->oid);
1754 	o->merge_size = len;
1755 	mark_all_ce_unused(o->src_index);
1756 
1757 	o->result.fsmonitor_last_update =
1758 		xstrdup_or_null(o->src_index->fsmonitor_last_update);
1759 
1760 	/*
1761 	 * Sparse checkout loop #1: set NEW_SKIP_WORKTREE on existing entries
1762 	 */
1763 	if (!o->skip_sparse_checkout)
1764 		mark_new_skip_worktree(o->pl, o->src_index, 0,
1765 				       CE_NEW_SKIP_WORKTREE, o->verbose_update);
1766 
1767 	if (!dfc)
1768 		dfc = xcalloc(1, cache_entry_size(0));
1769 	o->df_conflict_entry = dfc;
1770 
1771 	if (len) {
1772 		const char *prefix = o->prefix ? o->prefix : "";
1773 		struct traverse_info info;
1774 
1775 		setup_traverse_info(&info, prefix);
1776 		info.fn = unpack_callback;
1777 		info.data = o;
1778 		info.show_all_errors = o->show_all_errors;
1779 		info.pathspec = o->pathspec;
1780 
1781 		if (o->prefix) {
1782 			/*
1783 			 * Unpack existing index entries that sort before the
1784 			 * prefix the tree is spliced into.  Note that o->merge
1785 			 * is always true in this case.
1786 			 */
1787 			while (1) {
1788 				struct cache_entry *ce = next_cache_entry(o);
1789 				if (!ce)
1790 					break;
1791 				if (ce_in_traverse_path(ce, &info))
1792 					break;
1793 				if (unpack_index_entry(ce, o) < 0)
1794 					goto return_failed;
1795 			}
1796 		}
1797 
1798 		trace_performance_enter();
1799 		trace2_region_enter("unpack_trees", "traverse_trees", the_repository);
1800 		ret = traverse_trees(o->src_index, len, t, &info);
1801 		trace2_region_leave("unpack_trees", "traverse_trees", the_repository);
1802 		trace_performance_leave("traverse_trees");
1803 		if (ret < 0)
1804 			goto return_failed;
1805 	}
1806 
1807 	/* Any left-over entries in the index? */
1808 	if (o->merge) {
1809 		while (1) {
1810 			struct cache_entry *ce = next_cache_entry(o);
1811 			if (!ce)
1812 				break;
1813 			if (unpack_index_entry(ce, o) < 0)
1814 				goto return_failed;
1815 		}
1816 	}
1817 	mark_all_ce_unused(o->src_index);
1818 
1819 	if (o->trivial_merges_only && o->nontrivial_merge) {
1820 		ret = unpack_failed(o, "Merge requires file-level merging");
1821 		goto done;
1822 	}
1823 
1824 	if (!o->skip_sparse_checkout) {
1825 		/*
1826 		 * Sparse checkout loop #2: set NEW_SKIP_WORKTREE on entries not in loop #1
1827 		 * If they will have NEW_SKIP_WORKTREE, also set CE_SKIP_WORKTREE
1828 		 * so apply_sparse_checkout() won't attempt to remove it from worktree
1829 		 */
1830 		mark_new_skip_worktree(o->pl, &o->result,
1831 				       CE_ADDED, CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE,
1832 				       o->verbose_update);
1833 
1834 		ret = 0;
1835 		for (i = 0; i < o->result.cache_nr; i++) {
1836 			struct cache_entry *ce = o->result.cache[i];
1837 
1838 			/*
1839 			 * Entries marked with CE_ADDED in merged_entry() do not have
1840 			 * verify_absent() check (the check is effectively disabled
1841 			 * because CE_NEW_SKIP_WORKTREE is set unconditionally).
1842 			 *
1843 			 * Do the real check now because we have had
1844 			 * correct CE_NEW_SKIP_WORKTREE
1845 			 */
1846 			if (ce->ce_flags & CE_ADDED &&
1847 			    verify_absent(ce, WARNING_SPARSE_ORPHANED_NOT_OVERWRITTEN, o))
1848 				ret = 1;
1849 
1850 			if (apply_sparse_checkout(&o->result, ce, o))
1851 				ret = 1;
1852 		}
1853 		if (ret == 1) {
1854 			/*
1855 			 * Inability to sparsify or de-sparsify individual
1856 			 * paths is not an error, but just a warning.
1857 			 */
1858 			if (o->show_all_errors)
1859 				display_warning_msgs(o);
1860 			ret = 0;
1861 		}
1862 	}
1863 
1864 	ret = check_updates(o, &o->result) ? (-2) : 0;
1865 	if (o->dst_index) {
1866 		move_index_extensions(&o->result, o->src_index);
1867 		if (!ret) {
1868 			if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
1869 				cache_tree_verify(the_repository, &o->result);
1870 			if (!cache_tree_fully_valid(o->result.cache_tree))
1871 				cache_tree_update(&o->result,
1872 						  WRITE_TREE_SILENT |
1873 						  WRITE_TREE_REPAIR);
1874 		}
1875 
1876 		o->result.updated_workdir = 1;
1877 		discard_index(o->dst_index);
1878 		*o->dst_index = o->result;
1879 	} else {
1880 		discard_index(&o->result);
1881 	}
1882 	o->src_index = NULL;
1883 
1884 done:
1885 	if (free_pattern_list)
1886 		clear_pattern_list(&pl);
1887 	if (o->dir) {
1888 		dir_clear(o->dir);
1889 		o->dir = NULL;
1890 	}
1891 	trace2_region_leave("unpack_trees", "unpack_trees", the_repository);
1892 	trace_performance_leave("unpack_trees");
1893 	return ret;
1894 
1895 return_failed:
1896 	if (o->show_all_errors)
1897 		display_error_msgs(o);
1898 	mark_all_ce_unused(o->src_index);
1899 	ret = unpack_failed(o, NULL);
1900 	if (o->exiting_early)
1901 		ret = 0;
1902 	goto done;
1903 }
1904 
1905 /*
1906  * Update SKIP_WORKTREE bits according to sparsity patterns, and update
1907  * working directory to match.
1908  *
1909  * CE_NEW_SKIP_WORKTREE is used internally.
1910  */
update_sparsity(struct unpack_trees_options * o)1911 enum update_sparsity_result update_sparsity(struct unpack_trees_options *o)
1912 {
1913 	enum update_sparsity_result ret = UPDATE_SPARSITY_SUCCESS;
1914 	struct pattern_list pl;
1915 	int i;
1916 	unsigned old_show_all_errors;
1917 	int free_pattern_list = 0;
1918 
1919 	old_show_all_errors = o->show_all_errors;
1920 	o->show_all_errors = 1;
1921 
1922 	/* Sanity checks */
1923 	if (!o->update || o->index_only || o->skip_sparse_checkout)
1924 		BUG("update_sparsity() is for reflecting sparsity patterns in working directory");
1925 	if (o->src_index != o->dst_index || o->fn)
1926 		BUG("update_sparsity() called wrong");
1927 
1928 	trace_performance_enter();
1929 
1930 	/* If we weren't given patterns, use the recorded ones */
1931 	if (!o->pl) {
1932 		memset(&pl, 0, sizeof(pl));
1933 		free_pattern_list = 1;
1934 		populate_from_existing_patterns(o, &pl);
1935 		if (o->skip_sparse_checkout)
1936 			goto skip_sparse_checkout;
1937 	}
1938 
1939 	/* Set NEW_SKIP_WORKTREE on existing entries. */
1940 	mark_all_ce_unused(o->src_index);
1941 	mark_new_skip_worktree(o->pl, o->src_index, 0,
1942 			       CE_NEW_SKIP_WORKTREE, o->verbose_update);
1943 
1944 	/* Then loop over entries and update/remove as needed */
1945 	ret = UPDATE_SPARSITY_SUCCESS;
1946 	for (i = 0; i < o->src_index->cache_nr; i++) {
1947 		struct cache_entry *ce = o->src_index->cache[i];
1948 
1949 
1950 		if (ce_stage(ce)) {
1951 			/* -1 because for loop will increment by 1 */
1952 			i += warn_conflicted_path(o->src_index, i, o) - 1;
1953 			ret = UPDATE_SPARSITY_WARNINGS;
1954 			continue;
1955 		}
1956 
1957 		if (apply_sparse_checkout(o->src_index, ce, o))
1958 			ret = UPDATE_SPARSITY_WARNINGS;
1959 	}
1960 
1961 skip_sparse_checkout:
1962 	if (check_updates(o, o->src_index))
1963 		ret = UPDATE_SPARSITY_WORKTREE_UPDATE_FAILURES;
1964 
1965 	display_warning_msgs(o);
1966 	o->show_all_errors = old_show_all_errors;
1967 	if (free_pattern_list)
1968 		clear_pattern_list(&pl);
1969 	trace_performance_leave("update_sparsity");
1970 	return ret;
1971 }
1972 
1973 /* Here come the merge functions */
1974 
reject_merge(const struct cache_entry * ce,struct unpack_trees_options * o)1975 static int reject_merge(const struct cache_entry *ce,
1976 			struct unpack_trees_options *o)
1977 {
1978 	return add_rejected_path(o, ERROR_WOULD_OVERWRITE, ce->name);
1979 }
1980 
same(const struct cache_entry * a,const struct cache_entry * b)1981 static int same(const struct cache_entry *a, const struct cache_entry *b)
1982 {
1983 	if (!!a != !!b)
1984 		return 0;
1985 	if (!a && !b)
1986 		return 1;
1987 	if ((a->ce_flags | b->ce_flags) & CE_CONFLICTED)
1988 		return 0;
1989 	return a->ce_mode == b->ce_mode &&
1990 	       oideq(&a->oid, &b->oid);
1991 }
1992 
1993 
1994 /*
1995  * When a CE gets turned into an unmerged entry, we
1996  * want it to be up-to-date
1997  */
verify_uptodate_1(const struct cache_entry * ce,struct unpack_trees_options * o,enum unpack_trees_error_types error_type)1998 static int verify_uptodate_1(const struct cache_entry *ce,
1999 			     struct unpack_trees_options *o,
2000 			     enum unpack_trees_error_types error_type)
2001 {
2002 	struct stat st;
2003 
2004 	if (o->index_only)
2005 		return 0;
2006 
2007 	/*
2008 	 * CE_VALID and CE_SKIP_WORKTREE cheat, we better check again
2009 	 * if this entry is truly up-to-date because this file may be
2010 	 * overwritten.
2011 	 */
2012 	if ((ce->ce_flags & CE_VALID) || ce_skip_worktree(ce))
2013 		; /* keep checking */
2014 	else if (o->reset || ce_uptodate(ce))
2015 		return 0;
2016 
2017 	if (!lstat(ce->name, &st)) {
2018 		int flags = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE;
2019 		unsigned changed = ie_match_stat(o->src_index, ce, &st, flags);
2020 
2021 		if (submodule_from_ce(ce)) {
2022 			int r = check_submodule_move_head(ce,
2023 				"HEAD", oid_to_hex(&ce->oid), o);
2024 			if (r)
2025 				return add_rejected_path(o, error_type, ce->name);
2026 			return 0;
2027 		}
2028 
2029 		if (!changed)
2030 			return 0;
2031 		/*
2032 		 * Historic default policy was to allow submodule to be out
2033 		 * of sync wrt the superproject index. If the submodule was
2034 		 * not considered interesting above, we don't care here.
2035 		 */
2036 		if (S_ISGITLINK(ce->ce_mode))
2037 			return 0;
2038 
2039 		errno = 0;
2040 	}
2041 	if (errno == ENOENT)
2042 		return 0;
2043 	return add_rejected_path(o, error_type, ce->name);
2044 }
2045 
verify_uptodate(const struct cache_entry * ce,struct unpack_trees_options * o)2046 int verify_uptodate(const struct cache_entry *ce,
2047 		    struct unpack_trees_options *o)
2048 {
2049 	if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
2050 		return 0;
2051 	return verify_uptodate_1(ce, o, ERROR_NOT_UPTODATE_FILE);
2052 }
2053 
verify_uptodate_sparse(const struct cache_entry * ce,struct unpack_trees_options * o)2054 static int verify_uptodate_sparse(const struct cache_entry *ce,
2055 				  struct unpack_trees_options *o)
2056 {
2057 	return verify_uptodate_1(ce, o, WARNING_SPARSE_NOT_UPTODATE_FILE);
2058 }
2059 
2060 /*
2061  * TODO: We should actually invalidate o->result, not src_index [1].
2062  * But since cache tree and untracked cache both are not copied to
2063  * o->result until unpacking is complete, we invalidate them on
2064  * src_index instead with the assumption that they will be copied to
2065  * dst_index at the end.
2066  *
2067  * [1] src_index->cache_tree is also used in unpack_callback() so if
2068  * we invalidate o->result, we need to update it to use
2069  * o->result.cache_tree as well.
2070  */
invalidate_ce_path(const struct cache_entry * ce,struct unpack_trees_options * o)2071 static void invalidate_ce_path(const struct cache_entry *ce,
2072 			       struct unpack_trees_options *o)
2073 {
2074 	if (!ce)
2075 		return;
2076 	cache_tree_invalidate_path(o->src_index, ce->name);
2077 	untracked_cache_invalidate_path(o->src_index, ce->name, 1);
2078 }
2079 
2080 /*
2081  * Check that checking out ce->sha1 in subdir ce->name is not
2082  * going to overwrite any working files.
2083  */
verify_clean_submodule(const char * old_sha1,const struct cache_entry * ce,struct unpack_trees_options * o)2084 static int verify_clean_submodule(const char *old_sha1,
2085 				  const struct cache_entry *ce,
2086 				  struct unpack_trees_options *o)
2087 {
2088 	if (!submodule_from_ce(ce))
2089 		return 0;
2090 
2091 	return check_submodule_move_head(ce, old_sha1,
2092 					 oid_to_hex(&ce->oid), o);
2093 }
2094 
verify_clean_subdirectory(const struct cache_entry * ce,struct unpack_trees_options * o)2095 static int verify_clean_subdirectory(const struct cache_entry *ce,
2096 				     struct unpack_trees_options *o)
2097 {
2098 	/*
2099 	 * we are about to extract "ce->name"; we would not want to lose
2100 	 * anything in the existing directory there.
2101 	 */
2102 	int namelen;
2103 	int i;
2104 	struct dir_struct d;
2105 	char *pathbuf;
2106 	int cnt = 0;
2107 
2108 	if (S_ISGITLINK(ce->ce_mode)) {
2109 		struct object_id oid;
2110 		int sub_head = resolve_gitlink_ref(ce->name, "HEAD", &oid);
2111 		/*
2112 		 * If we are not going to update the submodule, then
2113 		 * we don't care.
2114 		 */
2115 		if (!sub_head && oideq(&oid, &ce->oid))
2116 			return 0;
2117 		return verify_clean_submodule(sub_head ? NULL : oid_to_hex(&oid),
2118 					      ce, o);
2119 	}
2120 
2121 	/*
2122 	 * First let's make sure we do not have a local modification
2123 	 * in that directory.
2124 	 */
2125 	namelen = ce_namelen(ce);
2126 	for (i = locate_in_src_index(ce, o);
2127 	     i < o->src_index->cache_nr;
2128 	     i++) {
2129 		struct cache_entry *ce2 = o->src_index->cache[i];
2130 		int len = ce_namelen(ce2);
2131 		if (len < namelen ||
2132 		    strncmp(ce->name, ce2->name, namelen) ||
2133 		    ce2->name[namelen] != '/')
2134 			break;
2135 		/*
2136 		 * ce2->name is an entry in the subdirectory to be
2137 		 * removed.
2138 		 */
2139 		if (!ce_stage(ce2)) {
2140 			if (verify_uptodate(ce2, o))
2141 				return -1;
2142 			add_entry(o, ce2, CE_REMOVE, 0);
2143 			invalidate_ce_path(ce, o);
2144 			mark_ce_used(ce2, o);
2145 		}
2146 		cnt++;
2147 	}
2148 
2149 	/*
2150 	 * Then we need to make sure that we do not lose a locally
2151 	 * present file that is not ignored.
2152 	 */
2153 	pathbuf = xstrfmt("%.*s/", namelen, ce->name);
2154 
2155 	memset(&d, 0, sizeof(d));
2156 	if (o->dir)
2157 		d.exclude_per_dir = o->dir->exclude_per_dir;
2158 	i = read_directory(&d, o->src_index, pathbuf, namelen+1, NULL);
2159 	dir_clear(&d);
2160 	free(pathbuf);
2161 	if (i)
2162 		return add_rejected_path(o, ERROR_NOT_UPTODATE_DIR, ce->name);
2163 	return cnt;
2164 }
2165 
2166 /*
2167  * This gets called when there was no index entry for the tree entry 'dst',
2168  * but we found a file in the working tree that 'lstat()' said was fine,
2169  * and we're on a case-insensitive filesystem.
2170  *
2171  * See if we can find a case-insensitive match in the index that also
2172  * matches the stat information, and assume it's that other file!
2173  */
icase_exists(struct unpack_trees_options * o,const char * name,int len,struct stat * st)2174 static int icase_exists(struct unpack_trees_options *o, const char *name, int len, struct stat *st)
2175 {
2176 	const struct cache_entry *src;
2177 
2178 	src = index_file_exists(o->src_index, name, len, 1);
2179 	return src && !ie_match_stat(o->src_index, src, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2180 }
2181 
2182 enum absent_checking_type {
2183 	COMPLETELY_ABSENT,
2184 	ABSENT_ANY_DIRECTORY
2185 };
2186 
check_ok_to_remove(const char * name,int len,int dtype,const struct cache_entry * ce,struct stat * st,enum unpack_trees_error_types error_type,enum absent_checking_type absent_type,struct unpack_trees_options * o)2187 static int check_ok_to_remove(const char *name, int len, int dtype,
2188 			      const struct cache_entry *ce, struct stat *st,
2189 			      enum unpack_trees_error_types error_type,
2190 			      enum absent_checking_type absent_type,
2191 			      struct unpack_trees_options *o)
2192 {
2193 	const struct cache_entry *result;
2194 
2195 	/*
2196 	 * It may be that the 'lstat()' succeeded even though
2197 	 * target 'ce' was absent, because there is an old
2198 	 * entry that is different only in case..
2199 	 *
2200 	 * Ignore that lstat() if it matches.
2201 	 */
2202 	if (ignore_case && icase_exists(o, name, len, st))
2203 		return 0;
2204 
2205 	if (o->dir &&
2206 	    is_excluded(o->dir, o->src_index, name, &dtype))
2207 		/*
2208 		 * ce->name is explicitly excluded, so it is Ok to
2209 		 * overwrite it.
2210 		 */
2211 		return 0;
2212 	if (S_ISDIR(st->st_mode)) {
2213 		/*
2214 		 * We are checking out path "foo" and
2215 		 * found "foo/." in the working tree.
2216 		 * This is tricky -- if we have modified
2217 		 * files that are in "foo/" we would lose
2218 		 * them.
2219 		 */
2220 		if (verify_clean_subdirectory(ce, o) < 0)
2221 			return -1;
2222 		return 0;
2223 	}
2224 
2225 	/* If we only care about directories, then we can remove */
2226 	if (absent_type == ABSENT_ANY_DIRECTORY)
2227 		return 0;
2228 
2229 	/*
2230 	 * The previous round may already have decided to
2231 	 * delete this path, which is in a subdirectory that
2232 	 * is being replaced with a blob.
2233 	 */
2234 	result = index_file_exists(&o->result, name, len, 0);
2235 	if (result) {
2236 		if (result->ce_flags & CE_REMOVE)
2237 			return 0;
2238 	}
2239 
2240 	return add_rejected_path(o, error_type, name);
2241 }
2242 
2243 /*
2244  * We do not want to remove or overwrite a working tree file that
2245  * is not tracked, unless it is ignored.
2246  */
verify_absent_1(const struct cache_entry * ce,enum unpack_trees_error_types error_type,enum absent_checking_type absent_type,struct unpack_trees_options * o)2247 static int verify_absent_1(const struct cache_entry *ce,
2248 			   enum unpack_trees_error_types error_type,
2249 			   enum absent_checking_type absent_type,
2250 			   struct unpack_trees_options *o)
2251 {
2252 	int len;
2253 	struct stat st;
2254 
2255 	if (o->index_only || !o->update ||
2256 	    o->reset == UNPACK_RESET_OVERWRITE_UNTRACKED)
2257 		return 0;
2258 
2259 	len = check_leading_path(ce->name, ce_namelen(ce), 0);
2260 	if (!len)
2261 		return 0;
2262 	else if (len > 0) {
2263 		char *path;
2264 		int ret;
2265 
2266 		path = xmemdupz(ce->name, len);
2267 		if (lstat(path, &st))
2268 			ret = error_errno("cannot stat '%s'", path);
2269 		else {
2270 			if (submodule_from_ce(ce))
2271 				ret = check_submodule_move_head(ce,
2272 								oid_to_hex(&ce->oid),
2273 								NULL, o);
2274 			else
2275 				ret = check_ok_to_remove(path, len, DT_UNKNOWN, NULL,
2276 							 &st, error_type,
2277 							 absent_type, o);
2278 		}
2279 		free(path);
2280 		return ret;
2281 	} else if (lstat(ce->name, &st)) {
2282 		if (errno != ENOENT)
2283 			return error_errno("cannot stat '%s'", ce->name);
2284 		return 0;
2285 	} else {
2286 		if (submodule_from_ce(ce))
2287 			return check_submodule_move_head(ce, oid_to_hex(&ce->oid),
2288 							 NULL, o);
2289 
2290 		return check_ok_to_remove(ce->name, ce_namelen(ce),
2291 					  ce_to_dtype(ce), ce, &st,
2292 					  error_type, absent_type, o);
2293 	}
2294 }
2295 
verify_absent(const struct cache_entry * ce,enum unpack_trees_error_types error_type,struct unpack_trees_options * o)2296 static int verify_absent(const struct cache_entry *ce,
2297 			 enum unpack_trees_error_types error_type,
2298 			 struct unpack_trees_options *o)
2299 {
2300 	if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
2301 		return 0;
2302 	return verify_absent_1(ce, error_type, COMPLETELY_ABSENT, o);
2303 }
2304 
verify_absent_if_directory(const struct cache_entry * ce,enum unpack_trees_error_types error_type,struct unpack_trees_options * o)2305 static int verify_absent_if_directory(const struct cache_entry *ce,
2306 				      enum unpack_trees_error_types error_type,
2307 				      struct unpack_trees_options *o)
2308 {
2309 	if (!o->skip_sparse_checkout && (ce->ce_flags & CE_NEW_SKIP_WORKTREE))
2310 		return 0;
2311 	return verify_absent_1(ce, error_type, ABSENT_ANY_DIRECTORY, o);
2312 }
2313 
verify_absent_sparse(const struct cache_entry * ce,enum unpack_trees_error_types error_type,struct unpack_trees_options * o)2314 static int verify_absent_sparse(const struct cache_entry *ce,
2315 				enum unpack_trees_error_types error_type,
2316 				struct unpack_trees_options *o)
2317 {
2318 	return verify_absent_1(ce, error_type, COMPLETELY_ABSENT, o);
2319 }
2320 
merged_entry(const struct cache_entry * ce,const struct cache_entry * old,struct unpack_trees_options * o)2321 static int merged_entry(const struct cache_entry *ce,
2322 			const struct cache_entry *old,
2323 			struct unpack_trees_options *o)
2324 {
2325 	int update = CE_UPDATE;
2326 	struct cache_entry *merge = dup_cache_entry(ce, &o->result);
2327 
2328 	if (!old) {
2329 		/*
2330 		 * New index entries. In sparse checkout, the following
2331 		 * verify_absent() will be delayed until after
2332 		 * traverse_trees() finishes in unpack_trees(), then:
2333 		 *
2334 		 *  - CE_NEW_SKIP_WORKTREE will be computed correctly
2335 		 *  - verify_absent() be called again, this time with
2336 		 *    correct CE_NEW_SKIP_WORKTREE
2337 		 *
2338 		 * verify_absent() call here does nothing in sparse
2339 		 * checkout (i.e. o->skip_sparse_checkout == 0)
2340 		 */
2341 		update |= CE_ADDED;
2342 		merge->ce_flags |= CE_NEW_SKIP_WORKTREE;
2343 
2344 		if (verify_absent(merge,
2345 				  ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2346 			discard_cache_entry(merge);
2347 			return -1;
2348 		}
2349 		invalidate_ce_path(merge, o);
2350 
2351 		if (submodule_from_ce(ce) && file_exists(ce->name)) {
2352 			int ret = check_submodule_move_head(ce, NULL,
2353 							    oid_to_hex(&ce->oid),
2354 							    o);
2355 			if (ret)
2356 				return ret;
2357 		}
2358 
2359 	} else if (!(old->ce_flags & CE_CONFLICTED)) {
2360 		/*
2361 		 * See if we can re-use the old CE directly?
2362 		 * That way we get the uptodate stat info.
2363 		 *
2364 		 * This also removes the UPDATE flag on a match; otherwise
2365 		 * we will end up overwriting local changes in the work tree.
2366 		 */
2367 		if (same(old, merge)) {
2368 			copy_cache_entry(merge, old);
2369 			update = 0;
2370 		} else {
2371 			if (verify_uptodate(old, o)) {
2372 				discard_cache_entry(merge);
2373 				return -1;
2374 			}
2375 			/* Migrate old flags over */
2376 			update |= old->ce_flags & (CE_SKIP_WORKTREE | CE_NEW_SKIP_WORKTREE);
2377 			invalidate_ce_path(old, o);
2378 		}
2379 
2380 		if (submodule_from_ce(ce) && file_exists(ce->name)) {
2381 			int ret = check_submodule_move_head(ce, oid_to_hex(&old->oid),
2382 							    oid_to_hex(&ce->oid),
2383 							    o);
2384 			if (ret)
2385 				return ret;
2386 		}
2387 	} else {
2388 		/*
2389 		 * Previously unmerged entry left as an existence
2390 		 * marker by read_index_unmerged();
2391 		 */
2392 		if (verify_absent_if_directory(merge,
2393 				  ERROR_WOULD_LOSE_UNTRACKED_OVERWRITTEN, o)) {
2394 			discard_cache_entry(merge);
2395 			return -1;
2396 		}
2397 
2398 		invalidate_ce_path(old, o);
2399 	}
2400 
2401 	if (do_add_entry(o, merge, update, CE_STAGEMASK) < 0)
2402 		return -1;
2403 	return 1;
2404 }
2405 
deleted_entry(const struct cache_entry * ce,const struct cache_entry * old,struct unpack_trees_options * o)2406 static int deleted_entry(const struct cache_entry *ce,
2407 			 const struct cache_entry *old,
2408 			 struct unpack_trees_options *o)
2409 {
2410 	/* Did it exist in the index? */
2411 	if (!old) {
2412 		if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2413 			return -1;
2414 		return 0;
2415 	} else if (verify_absent_if_directory(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o)) {
2416 		return -1;
2417 	}
2418 
2419 	if (!(old->ce_flags & CE_CONFLICTED) && verify_uptodate(old, o))
2420 		return -1;
2421 	add_entry(o, ce, CE_REMOVE, 0);
2422 	invalidate_ce_path(ce, o);
2423 	return 1;
2424 }
2425 
keep_entry(const struct cache_entry * ce,struct unpack_trees_options * o)2426 static int keep_entry(const struct cache_entry *ce,
2427 		      struct unpack_trees_options *o)
2428 {
2429 	add_entry(o, ce, 0, 0);
2430 	if (ce_stage(ce))
2431 		invalidate_ce_path(ce, o);
2432 	return 1;
2433 }
2434 
2435 #if DBRT_DEBUG
show_stage_entry(FILE * o,const char * label,const struct cache_entry * ce)2436 static void show_stage_entry(FILE *o,
2437 			     const char *label, const struct cache_entry *ce)
2438 {
2439 	if (!ce)
2440 		fprintf(o, "%s (missing)\n", label);
2441 	else
2442 		fprintf(o, "%s%06o %s %d\t%s\n",
2443 			label,
2444 			ce->ce_mode,
2445 			oid_to_hex(&ce->oid),
2446 			ce_stage(ce),
2447 			ce->name);
2448 }
2449 #endif
2450 
threeway_merge(const struct cache_entry * const * stages,struct unpack_trees_options * o)2451 int threeway_merge(const struct cache_entry * const *stages,
2452 		   struct unpack_trees_options *o)
2453 {
2454 	const struct cache_entry *index;
2455 	const struct cache_entry *head;
2456 	const struct cache_entry *remote = stages[o->head_idx + 1];
2457 	int count;
2458 	int head_match = 0;
2459 	int remote_match = 0;
2460 
2461 	int df_conflict_head = 0;
2462 	int df_conflict_remote = 0;
2463 
2464 	int any_anc_missing = 0;
2465 	int no_anc_exists = 1;
2466 	int i;
2467 
2468 	for (i = 1; i < o->head_idx; i++) {
2469 		if (!stages[i] || stages[i] == o->df_conflict_entry)
2470 			any_anc_missing = 1;
2471 		else
2472 			no_anc_exists = 0;
2473 	}
2474 
2475 	index = stages[0];
2476 	head = stages[o->head_idx];
2477 
2478 	if (head == o->df_conflict_entry) {
2479 		df_conflict_head = 1;
2480 		head = NULL;
2481 	}
2482 
2483 	if (remote == o->df_conflict_entry) {
2484 		df_conflict_remote = 1;
2485 		remote = NULL;
2486 	}
2487 
2488 	/*
2489 	 * First, if there's a #16 situation, note that to prevent #13
2490 	 * and #14.
2491 	 */
2492 	if (!same(remote, head)) {
2493 		for (i = 1; i < o->head_idx; i++) {
2494 			if (same(stages[i], head)) {
2495 				head_match = i;
2496 			}
2497 			if (same(stages[i], remote)) {
2498 				remote_match = i;
2499 			}
2500 		}
2501 	}
2502 
2503 	/*
2504 	 * We start with cases where the index is allowed to match
2505 	 * something other than the head: #14(ALT) and #2ALT, where it
2506 	 * is permitted to match the result instead.
2507 	 */
2508 	/* #14, #14ALT, #2ALT */
2509 	if (remote && !df_conflict_head && head_match && !remote_match) {
2510 		if (index && !same(index, remote) && !same(index, head))
2511 			return reject_merge(index, o);
2512 		return merged_entry(remote, index, o);
2513 	}
2514 	/*
2515 	 * If we have an entry in the index cache, then we want to
2516 	 * make sure that it matches head.
2517 	 */
2518 	if (index && !same(index, head))
2519 		return reject_merge(index, o);
2520 
2521 	if (head) {
2522 		/* #5ALT, #15 */
2523 		if (same(head, remote))
2524 			return merged_entry(head, index, o);
2525 		/* #13, #3ALT */
2526 		if (!df_conflict_remote && remote_match && !head_match)
2527 			return merged_entry(head, index, o);
2528 	}
2529 
2530 	/* #1 */
2531 	if (!head && !remote && any_anc_missing)
2532 		return 0;
2533 
2534 	/*
2535 	 * Under the "aggressive" rule, we resolve mostly trivial
2536 	 * cases that we historically had git-merge-one-file resolve.
2537 	 */
2538 	if (o->aggressive) {
2539 		int head_deleted = !head;
2540 		int remote_deleted = !remote;
2541 		const struct cache_entry *ce = NULL;
2542 
2543 		if (index)
2544 			ce = index;
2545 		else if (head)
2546 			ce = head;
2547 		else if (remote)
2548 			ce = remote;
2549 		else {
2550 			for (i = 1; i < o->head_idx; i++) {
2551 				if (stages[i] && stages[i] != o->df_conflict_entry) {
2552 					ce = stages[i];
2553 					break;
2554 				}
2555 			}
2556 		}
2557 
2558 		/*
2559 		 * Deleted in both.
2560 		 * Deleted in one and unchanged in the other.
2561 		 */
2562 		if ((head_deleted && remote_deleted) ||
2563 		    (head_deleted && remote && remote_match) ||
2564 		    (remote_deleted && head && head_match)) {
2565 			if (index)
2566 				return deleted_entry(index, index, o);
2567 			if (ce && !head_deleted) {
2568 				if (verify_absent(ce, ERROR_WOULD_LOSE_UNTRACKED_REMOVED, o))
2569 					return -1;
2570 			}
2571 			return 0;
2572 		}
2573 		/*
2574 		 * Added in both, identically.
2575 		 */
2576 		if (no_anc_exists && head && remote && same(head, remote))
2577 			return merged_entry(head, index, o);
2578 
2579 	}
2580 
2581 	/* Below are "no merge" cases, which require that the index be
2582 	 * up-to-date to avoid the files getting overwritten with
2583 	 * conflict resolution files.
2584 	 */
2585 	if (index) {
2586 		if (verify_uptodate(index, o))
2587 			return -1;
2588 	}
2589 
2590 	o->nontrivial_merge = 1;
2591 
2592 	/* #2, #3, #4, #6, #7, #9, #10, #11. */
2593 	count = 0;
2594 	if (!head_match || !remote_match) {
2595 		for (i = 1; i < o->head_idx; i++) {
2596 			if (stages[i] && stages[i] != o->df_conflict_entry) {
2597 				keep_entry(stages[i], o);
2598 				count++;
2599 				break;
2600 			}
2601 		}
2602 	}
2603 #if DBRT_DEBUG
2604 	else {
2605 		fprintf(stderr, "read-tree: warning #16 detected\n");
2606 		show_stage_entry(stderr, "head   ", stages[head_match]);
2607 		show_stage_entry(stderr, "remote ", stages[remote_match]);
2608 	}
2609 #endif
2610 	if (head) { count += keep_entry(head, o); }
2611 	if (remote) { count += keep_entry(remote, o); }
2612 	return count;
2613 }
2614 
2615 /*
2616  * Two-way merge.
2617  *
2618  * The rule is to "carry forward" what is in the index without losing
2619  * information across a "fast-forward", favoring a successful merge
2620  * over a merge failure when it makes sense.  For details of the
2621  * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
2622  *
2623  */
twoway_merge(const struct cache_entry * const * src,struct unpack_trees_options * o)2624 int twoway_merge(const struct cache_entry * const *src,
2625 		 struct unpack_trees_options *o)
2626 {
2627 	const struct cache_entry *current = src[0];
2628 	const struct cache_entry *oldtree = src[1];
2629 	const struct cache_entry *newtree = src[2];
2630 
2631 	if (o->merge_size != 2)
2632 		return error("Cannot do a twoway merge of %d trees",
2633 			     o->merge_size);
2634 
2635 	if (oldtree == o->df_conflict_entry)
2636 		oldtree = NULL;
2637 	if (newtree == o->df_conflict_entry)
2638 		newtree = NULL;
2639 
2640 	if (current) {
2641 		if (current->ce_flags & CE_CONFLICTED) {
2642 			if (same(oldtree, newtree) || o->reset) {
2643 				if (!newtree)
2644 					return deleted_entry(current, current, o);
2645 				else
2646 					return merged_entry(newtree, current, o);
2647 			}
2648 			return reject_merge(current, o);
2649 		} else if ((!oldtree && !newtree) || /* 4 and 5 */
2650 			 (!oldtree && newtree &&
2651 			  same(current, newtree)) || /* 6 and 7 */
2652 			 (oldtree && newtree &&
2653 			  same(oldtree, newtree)) || /* 14 and 15 */
2654 			 (oldtree && newtree &&
2655 			  !same(oldtree, newtree) && /* 18 and 19 */
2656 			  same(current, newtree))) {
2657 			return keep_entry(current, o);
2658 		} else if (oldtree && !newtree && same(current, oldtree)) {
2659 			/* 10 or 11 */
2660 			return deleted_entry(oldtree, current, o);
2661 		} else if (oldtree && newtree &&
2662 			 same(current, oldtree) && !same(current, newtree)) {
2663 			/* 20 or 21 */
2664 			return merged_entry(newtree, current, o);
2665 		} else if (current && !oldtree && newtree &&
2666 			   S_ISSPARSEDIR(current->ce_mode) != S_ISSPARSEDIR(newtree->ce_mode) &&
2667 			   ce_stage(current) == 0) {
2668 			/*
2669 			 * This case is a directory/file conflict across the sparse-index
2670 			 * boundary. When we are changing from one path to another via
2671 			 * 'git checkout', then we want to replace one entry with another
2672 			 * via merged_entry(). If there are staged changes, then we should
2673 			 * reject the merge instead.
2674 			 */
2675 			return merged_entry(newtree, current, o);
2676 		} else
2677 			return reject_merge(current, o);
2678 	}
2679 	else if (newtree) {
2680 		if (oldtree && !o->initial_checkout) {
2681 			/*
2682 			 * deletion of the path was staged;
2683 			 */
2684 			if (same(oldtree, newtree))
2685 				return 1;
2686 			return reject_merge(oldtree, o);
2687 		}
2688 		return merged_entry(newtree, current, o);
2689 	}
2690 	return deleted_entry(oldtree, current, o);
2691 }
2692 
2693 /*
2694  * Bind merge.
2695  *
2696  * Keep the index entries at stage0, collapse stage1 but make sure
2697  * stage0 does not have anything there.
2698  */
bind_merge(const struct cache_entry * const * src,struct unpack_trees_options * o)2699 int bind_merge(const struct cache_entry * const *src,
2700 	       struct unpack_trees_options *o)
2701 {
2702 	const struct cache_entry *old = src[0];
2703 	const struct cache_entry *a = src[1];
2704 
2705 	if (o->merge_size != 1)
2706 		return error("Cannot do a bind merge of %d trees",
2707 			     o->merge_size);
2708 	if (a && old)
2709 		return o->quiet ? -1 :
2710 			error(ERRORMSG(o, ERROR_BIND_OVERLAP),
2711 			      super_prefixed(a->name),
2712 			      super_prefixed(old->name));
2713 	if (!a)
2714 		return keep_entry(old, o);
2715 	else
2716 		return merged_entry(a, NULL, o);
2717 }
2718 
2719 /*
2720  * One-way merge.
2721  *
2722  * The rule is:
2723  * - take the stat information from stage0, take the data from stage1
2724  */
oneway_merge(const struct cache_entry * const * src,struct unpack_trees_options * o)2725 int oneway_merge(const struct cache_entry * const *src,
2726 		 struct unpack_trees_options *o)
2727 {
2728 	const struct cache_entry *old = src[0];
2729 	const struct cache_entry *a = src[1];
2730 
2731 	if (o->merge_size != 1)
2732 		return error("Cannot do a oneway merge of %d trees",
2733 			     o->merge_size);
2734 
2735 	if (!a || a == o->df_conflict_entry)
2736 		return deleted_entry(old, old, o);
2737 
2738 	if (old && same(old, a)) {
2739 		int update = 0;
2740 		if (o->reset && o->update && !ce_uptodate(old) && !ce_skip_worktree(old) &&
2741 			!(old->ce_flags & CE_FSMONITOR_VALID)) {
2742 			struct stat st;
2743 			if (lstat(old->name, &st) ||
2744 			    ie_match_stat(o->src_index, old, &st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE))
2745 				update |= CE_UPDATE;
2746 		}
2747 		if (o->update && S_ISGITLINK(old->ce_mode) &&
2748 		    should_update_submodules() && !verify_uptodate(old, o))
2749 			update |= CE_UPDATE;
2750 		add_entry(o, old, update, CE_STAGEMASK);
2751 		return 0;
2752 	}
2753 	return merged_entry(a, old, o);
2754 }
2755 
2756 /*
2757  * Merge worktree and untracked entries in a stash entry.
2758  *
2759  * Ignore all index entries. Collapse remaining trees but make sure that they
2760  * don't have any conflicting files.
2761  */
stash_worktree_untracked_merge(const struct cache_entry * const * src,struct unpack_trees_options * o)2762 int stash_worktree_untracked_merge(const struct cache_entry * const *src,
2763 				   struct unpack_trees_options *o)
2764 {
2765 	const struct cache_entry *worktree = src[1];
2766 	const struct cache_entry *untracked = src[2];
2767 
2768 	if (o->merge_size != 2)
2769 		BUG("invalid merge_size: %d", o->merge_size);
2770 
2771 	if (worktree && untracked)
2772 		return error(_("worktree and untracked commit have duplicate entries: %s"),
2773 			     super_prefixed(worktree->name));
2774 
2775 	return merged_entry(worktree ? worktree : untracked, NULL, o);
2776 }
2777