Merge branch 'zh/ls-files-format'
[git.git] / merge-ort.c
blob8b7de0fbd8e87118aad09fad3be8ef77cd0bdbbc
1 /*
2 * "Ostensibly Recursive's Twin" merge strategy, or "ort" for short. Meant
3 * as a drop-in replacement for the "recursive" merge strategy, allowing one
4 * to replace
6 * git merge [-s recursive]
8 * with
10 * git merge -s ort
12 * Note: git's parser allows the space between '-s' and its argument to be
13 * missing. (Should I have backronymed "ham", "alsa", "kip", "nap, "alvo",
14 * "cale", "peedy", or "ins" instead of "ort"?)
17 #include "cache.h"
18 #include "merge-ort.h"
20 #include "alloc.h"
21 #include "attr.h"
22 #include "blob.h"
23 #include "cache-tree.h"
24 #include "commit.h"
25 #include "commit-reach.h"
26 #include "diff.h"
27 #include "diffcore.h"
28 #include "dir.h"
29 #include "entry.h"
30 #include "ll-merge.h"
31 #include "object-store.h"
32 #include "promisor-remote.h"
33 #include "revision.h"
34 #include "strmap.h"
35 #include "submodule-config.h"
36 #include "submodule.h"
37 #include "tree.h"
38 #include "unpack-trees.h"
39 #include "xdiff-interface.h"
42 * We have many arrays of size 3. Whenever we have such an array, the
43 * indices refer to one of the sides of the three-way merge. This is so
44 * pervasive that the constants 0, 1, and 2 are used in many places in the
45 * code (especially in arithmetic operations to find the other side's index
46 * or to compute a relevant mask), but sometimes these enum names are used
47 * to aid code clarity.
49 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
50 * referred to there is one of these three sides.
52 enum merge_side {
53 MERGE_BASE = 0,
54 MERGE_SIDE1 = 1,
55 MERGE_SIDE2 = 2
58 static unsigned RESULT_INITIALIZED = 0x1abe11ed; /* unlikely accidental value */
60 struct traversal_callback_data {
61 unsigned long mask;
62 unsigned long dirmask;
63 struct name_entry names[3];
66 struct deferred_traversal_data {
68 * possible_trivial_merges: directories to be explored only when needed
70 * possible_trivial_merges is a map of directory names to
71 * dir_rename_mask. When we detect that a directory is unchanged on
72 * one side, we can sometimes resolve the directory without recursing
73 * into it. Renames are the only things that can prevent such an
74 * optimization. However, for rename sources:
75 * - If no parent directory needed directory rename detection, then
76 * no path under such a directory can be a relevant_source.
77 * and for rename destinations:
78 * - If no cached rename has a target path under the directory AND
79 * - If there are no unpaired relevant_sources elsewhere in the
80 * repository
81 * then we don't need any path under this directory for a rename
82 * destination. The only way to know the last item above is to defer
83 * handling such directories until the end of collect_merge_info(),
84 * in handle_deferred_entries().
86 * For each we store dir_rename_mask, since that's the only bit of
87 * information we need, other than the path, to resume the recursive
88 * traversal.
90 struct strintmap possible_trivial_merges;
93 * trivial_merges_okay: if trivial directory merges are okay
95 * See possible_trivial_merges above. The "no unpaired
96 * relevant_sources elsewhere in the repository" is a single boolean
97 * per merge side, which we store here. Note that while 0 means no,
98 * 1 only means "maybe" rather than "yes"; we optimistically set it
99 * to 1 initially and only clear when we determine it is unsafe to
100 * do trivial directory merges.
102 unsigned trivial_merges_okay;
105 * target_dirs: ancestor directories of rename targets
107 * target_dirs contains all directory names that are an ancestor of
108 * any rename destination.
110 struct strset target_dirs;
113 struct rename_info {
115 * All variables that are arrays of size 3 correspond to data tracked
116 * for the sides in enum merge_side. Index 0 is almost always unused
117 * because we often only need to track information for MERGE_SIDE1 and
118 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
119 * are determined relative to what changed since the MERGE_BASE).
123 * pairs: pairing of filenames from diffcore_rename()
125 struct diff_queue_struct pairs[3];
128 * dirs_removed: directories removed on a given side of history.
130 * The keys of dirs_removed[side] are the directories that were removed
131 * on the given side of history. The value of the strintmap for each
132 * directory is a value from enum dir_rename_relevance.
134 struct strintmap dirs_removed[3];
137 * dir_rename_count: tracking where parts of a directory were renamed to
139 * When files in a directory are renamed, they may not all go to the
140 * same location. Each strmap here tracks:
141 * old_dir => {new_dir => int}
142 * That is, dir_rename_count[side] is a strmap to a strintmap.
144 struct strmap dir_rename_count[3];
147 * dir_renames: computed directory renames
149 * This is a map of old_dir => new_dir and is derived in part from
150 * dir_rename_count.
152 struct strmap dir_renames[3];
155 * relevant_sources: deleted paths wanted in rename detection, and why
157 * relevant_sources is a set of deleted paths on each side of
158 * history for which we need rename detection. If a path is deleted
159 * on one side of history, we need to detect if it is part of a
160 * rename if either
161 * * the file is modified/deleted on the other side of history
162 * * we need to detect renames for an ancestor directory
163 * If neither of those are true, we can skip rename detection for
164 * that path. The reason is stored as a value from enum
165 * file_rename_relevance, as the reason can inform the algorithm in
166 * diffcore_rename_extended().
168 struct strintmap relevant_sources[3];
170 struct deferred_traversal_data deferred[3];
173 * dir_rename_mask:
174 * 0: optimization removing unmodified potential rename source okay
175 * 2 or 4: optimization okay, but must check for files added to dir
176 * 7: optimization forbidden; need rename source in case of dir rename
178 unsigned dir_rename_mask:3;
181 * callback_data_*: supporting data structures for alternate traversal
183 * We sometimes need to be able to traverse through all the files
184 * in a given tree before all immediate subdirectories within that
185 * tree. Since traverse_trees() doesn't do that naturally, we have
186 * a traverse_trees_wrapper() that stores any immediate
187 * subdirectories while traversing files, then traverses the
188 * immediate subdirectories later. These callback_data* variables
189 * store the information for the subdirectories so that we can do
190 * that traversal order.
192 struct traversal_callback_data *callback_data;
193 int callback_data_nr, callback_data_alloc;
194 char *callback_data_traverse_path;
197 * merge_trees: trees passed to the merge algorithm for the merge
199 * merge_trees records the trees passed to the merge algorithm. But,
200 * this data also is stored in merge_result->priv. If a sequence of
201 * merges are being done (such as when cherry-picking or rebasing),
202 * the next merge can look at this and re-use information from
203 * previous merges under certain circumstances.
205 * See also all the cached_* variables.
207 struct tree *merge_trees[3];
210 * cached_pairs_valid_side: which side's cached info can be reused
212 * See the description for merge_trees. For repeated merges, at most
213 * only one side's cached information can be used. Valid values:
214 * MERGE_SIDE2: cached data from side2 can be reused
215 * MERGE_SIDE1: cached data from side1 can be reused
216 * 0: no cached data can be reused
217 * -1: See redo_after_renames; both sides can be reused.
219 int cached_pairs_valid_side;
222 * cached_pairs: Caching of renames and deletions.
224 * These are mappings recording renames and deletions of individual
225 * files (not directories). They are thus a map from an old
226 * filename to either NULL (for deletions) or a new filename (for
227 * renames).
229 struct strmap cached_pairs[3];
232 * cached_target_names: just the destinations from cached_pairs
234 * We sometimes want a fast lookup to determine if a given filename
235 * is one of the destinations in cached_pairs. cached_target_names
236 * is thus duplicative information, but it provides a fast lookup.
238 struct strset cached_target_names[3];
241 * cached_irrelevant: Caching of rename_sources that aren't relevant.
243 * If we try to detect a rename for a source path and succeed, it's
244 * part of a rename. If we try to detect a rename for a source path
245 * and fail, then it's a delete. If we do not try to detect a rename
246 * for a path, then we don't know if it's a rename or a delete. If
247 * merge-ort doesn't think the path is relevant, then we just won't
248 * cache anything for that path. But there's a slight problem in
249 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
250 * commit 9bd342137e ("diffcore-rename: determine which
251 * relevant_sources are no longer relevant", 2021-03-13),
252 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
253 * avoid excessive calls to diffcore_rename_extended() we still need
254 * to cache such paths, though we cannot record them as either
255 * renames or deletes. So we cache them here as a "turned out to be
256 * irrelevant *for this commit*" as they are often also irrelevant
257 * for subsequent commits, though we will have to do some extra
258 * checking to see whether such paths become relevant for rename
259 * detection when cherry-picking/rebasing subsequent commits.
261 struct strset cached_irrelevant[3];
264 * redo_after_renames: optimization flag for "restarting" the merge
266 * Sometimes it pays to detect renames, cache them, and then
267 * restart the merge operation from the beginning. The reason for
268 * this is that when we know where all the renames are, we know
269 * whether a certain directory has any paths under it affected --
270 * and if a directory is not affected then it permits us to do
271 * trivial tree merging in more cases. Doing trivial tree merging
272 * prevents the need to run process_entry() on every path
273 * underneath trees that can be trivially merged, and
274 * process_entry() is more expensive than collect_merge_info() --
275 * plus, the second collect_merge_info() will be much faster since
276 * it doesn't have to recurse into the relevant trees.
278 * Values for this flag:
279 * 0 = don't bother, not worth it (or conditions not yet checked)
280 * 1 = conditions for optimization met, optimization worthwhile
281 * 2 = we already did it (don't restart merge yet again)
283 unsigned redo_after_renames;
286 * needed_limit: value needed for inexact rename detection to run
288 * If the current rename limit wasn't high enough for inexact
289 * rename detection to run, this records the limit needed. Otherwise,
290 * this value remains 0.
292 int needed_limit;
295 struct merge_options_internal {
297 * paths: primary data structure in all of merge ort.
299 * The keys of paths:
300 * * are full relative paths from the toplevel of the repository
301 * (e.g. "drivers/firmware/raspberrypi.c").
302 * * store all relevant paths in the repo, both directories and
303 * files (e.g. drivers, drivers/firmware would also be included)
304 * * these keys serve to intern all the path strings, which allows
305 * us to do pointer comparison on directory names instead of
306 * strcmp; we just have to be careful to use the interned strings.
308 * The values of paths:
309 * * either a pointer to a merged_info, or a conflict_info struct
310 * * merged_info contains all relevant information for a
311 * non-conflicted entry.
312 * * conflict_info contains a merged_info, plus any additional
313 * information about a conflict such as the higher orders stages
314 * involved and the names of the paths those came from (handy
315 * once renames get involved).
316 * * a path may start "conflicted" (i.e. point to a conflict_info)
317 * and then a later step (e.g. three-way content merge) determines
318 * it can be cleanly merged, at which point it'll be marked clean
319 * and the algorithm will ignore any data outside the contained
320 * merged_info for that entry
321 * * If an entry remains conflicted, the merged_info portion of a
322 * conflict_info will later be filled with whatever version of
323 * the file should be placed in the working directory (e.g. an
324 * as-merged-as-possible variation that contains conflict markers).
326 struct strmap paths;
329 * conflicted: a subset of keys->values from "paths"
331 * conflicted is basically an optimization between process_entries()
332 * and record_conflicted_index_entries(); the latter could loop over
333 * ALL the entries in paths AGAIN and look for the ones that are
334 * still conflicted, but since process_entries() has to loop over
335 * all of them, it saves the ones it couldn't resolve in this strmap
336 * so that record_conflicted_index_entries() can iterate just the
337 * relevant entries.
339 struct strmap conflicted;
342 * pool: memory pool for fast allocation/deallocation
344 * We allocate room for lots of filenames and auxiliary data
345 * structures in merge_options_internal, and it tends to all be
346 * freed together too. Using a memory pool for these provides a
347 * nice speedup.
349 struct mem_pool pool;
352 * conflicts: logical conflicts and messages stored by _primary_ path
354 * This is a map of pathnames (a subset of the keys in "paths" above)
355 * to struct string_list, with each item's `util` containing a
356 * `struct logical_conflict_info`. Note, though, that for each path,
357 * it only stores the logical conflicts for which that path is the
358 * primary path; the path might be part of additional conflicts.
360 struct strmap conflicts;
363 * renames: various data relating to rename detection
365 struct rename_info renames;
368 * attr_index: hacky minimal index used for renormalization
370 * renormalization code _requires_ an index, though it only needs to
371 * find a .gitattributes file within the index. So, when
372 * renormalization is important, we create a special index with just
373 * that one file.
375 struct index_state attr_index;
378 * current_dir_name, toplevel_dir: temporary vars
380 * These are used in collect_merge_info_callback(), and will set the
381 * various merged_info.directory_name for the various paths we get;
382 * see documentation for that variable and the requirements placed on
383 * that field.
385 const char *current_dir_name;
386 const char *toplevel_dir;
388 /* call_depth: recursion level counter for merging merge bases */
389 int call_depth;
392 struct version_info {
393 struct object_id oid;
394 unsigned short mode;
397 struct merged_info {
398 /* if is_null, ignore result. otherwise result has oid & mode */
399 struct version_info result;
400 unsigned is_null:1;
403 * clean: whether the path in question is cleanly merged.
405 * see conflict_info.merged for more details.
407 unsigned clean:1;
410 * basename_offset: offset of basename of path.
412 * perf optimization to avoid recomputing offset of final '/'
413 * character in pathname (0 if no '/' in pathname).
415 size_t basename_offset;
418 * directory_name: containing directory name.
420 * Note that we assume directory_name is constructed such that
421 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
422 * i.e. string equality is equivalent to pointer equality. For this
423 * to hold, we have to be careful setting directory_name.
425 const char *directory_name;
428 struct conflict_info {
430 * merged: the version of the path that will be written to working tree
432 * WARNING: It is critical to check merged.clean and ensure it is 0
433 * before reading any conflict_info fields outside of merged.
434 * Allocated merge_info structs will always have clean set to 1.
435 * Allocated conflict_info structs will have merged.clean set to 0
436 * initially. The merged.clean field is how we know if it is safe
437 * to access other parts of conflict_info besides merged; if a
438 * conflict_info's merged.clean is changed to 1, the rest of the
439 * algorithm is not allowed to look at anything outside of the
440 * merged member anymore.
442 struct merged_info merged;
444 /* oids & modes from each of the three trees for this path */
445 struct version_info stages[3];
447 /* pathnames for each stage; may differ due to rename detection */
448 const char *pathnames[3];
450 /* Whether this path is/was involved in a directory/file conflict */
451 unsigned df_conflict:1;
454 * Whether this path is/was involved in a non-content conflict other
455 * than a directory/file conflict (e.g. rename/rename, rename/delete,
456 * file location based on possible directory rename).
458 unsigned path_conflict:1;
461 * For filemask and dirmask, the ith bit corresponds to whether the
462 * ith entry is a file (filemask) or a directory (dirmask). Thus,
463 * filemask & dirmask is always zero, and filemask | dirmask is at
464 * most 7 but can be less when a path does not appear as either a
465 * file or a directory on at least one side of history.
467 * Note that these masks are related to enum merge_side, as the ith
468 * entry corresponds to side i.
470 * These values come from a traverse_trees() call; more info may be
471 * found looking at tree-walk.h's struct traverse_info,
472 * particularly the documentation above the "fn" member (note that
473 * filemask = mask & ~dirmask from that documentation).
475 unsigned filemask:3;
476 unsigned dirmask:3;
479 * Optimization to track which stages match, to avoid the need to
480 * recompute it in multiple steps. Either 0 or at least 2 bits are
481 * set; if at least 2 bits are set, their corresponding stages match.
483 unsigned match_mask:3;
486 enum conflict_and_info_types {
487 /* "Simple" conflicts and informational messages */
488 INFO_AUTO_MERGING = 0,
489 CONFLICT_CONTENTS, /* text file that failed to merge */
490 CONFLICT_BINARY,
491 CONFLICT_FILE_DIRECTORY,
492 CONFLICT_DISTINCT_MODES,
493 CONFLICT_MODIFY_DELETE,
494 CONFLICT_PRESENT_DESPITE_SKIPPED,
496 /* Regular rename */
497 CONFLICT_RENAME_RENAME, /* same file renamed differently */
498 CONFLICT_RENAME_COLLIDES, /* rename/add or two files renamed to 1 */
499 CONFLICT_RENAME_DELETE,
501 /* Basic directory rename */
502 CONFLICT_DIR_RENAME_SUGGESTED,
503 INFO_DIR_RENAME_APPLIED,
505 /* Special directory rename cases */
506 INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME,
507 CONFLICT_DIR_RENAME_FILE_IN_WAY,
508 CONFLICT_DIR_RENAME_COLLISION,
509 CONFLICT_DIR_RENAME_SPLIT,
511 /* Basic submodule */
512 INFO_SUBMODULE_FAST_FORWARDING,
513 CONFLICT_SUBMODULE_FAILED_TO_MERGE,
515 /* Special submodule cases broken out from FAILED_TO_MERGE */
516 CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION,
517 CONFLICT_SUBMODULE_NOT_INITIALIZED,
518 CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE,
519 CONFLICT_SUBMODULE_MAY_HAVE_REWINDS,
521 /* Keep this entry _last_ in the list */
522 NB_CONFLICT_TYPES,
526 * Short description of conflict type, relied upon by external tools.
528 * We can add more entries, but DO NOT change any of these strings. Also,
529 * Order MUST match conflict_info_and_types.
531 static const char *type_short_descriptions[] = {
532 /*** "Simple" conflicts and informational messages ***/
533 [INFO_AUTO_MERGING] = "Auto-merging",
534 [CONFLICT_CONTENTS] = "CONFLICT (contents)",
535 [CONFLICT_BINARY] = "CONFLICT (binary)",
536 [CONFLICT_FILE_DIRECTORY] = "CONFLICT (file/directory)",
537 [CONFLICT_DISTINCT_MODES] = "CONFLICT (distinct modes)",
538 [CONFLICT_MODIFY_DELETE] = "CONFLICT (modify/delete)",
539 [CONFLICT_PRESENT_DESPITE_SKIPPED] =
540 "CONFLICT (upgrade your version of git)",
542 /*** Regular rename ***/
543 [CONFLICT_RENAME_RENAME] = "CONFLICT (rename/rename)",
544 [CONFLICT_RENAME_COLLIDES] = "CONFLICT (rename involved in collision)",
545 [CONFLICT_RENAME_DELETE] = "CONFLICT (rename/delete)",
547 /*** Basic directory rename ***/
548 [CONFLICT_DIR_RENAME_SUGGESTED] =
549 "CONFLICT (directory rename suggested)",
550 [INFO_DIR_RENAME_APPLIED] = "Path updated due to directory rename",
552 /*** Special directory rename cases ***/
553 [INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME] =
554 "Directory rename skipped since directory was renamed on both sides",
555 [CONFLICT_DIR_RENAME_FILE_IN_WAY] =
556 "CONFLICT (file in way of directory rename)",
557 [CONFLICT_DIR_RENAME_COLLISION] = "CONFLICT(directory rename collision)",
558 [CONFLICT_DIR_RENAME_SPLIT] = "CONFLICT(directory rename unclear split)",
560 /*** Basic submodule ***/
561 [INFO_SUBMODULE_FAST_FORWARDING] = "Fast forwarding submodule",
562 [CONFLICT_SUBMODULE_FAILED_TO_MERGE] = "CONFLICT (submodule)",
564 /*** Special submodule cases broken out from FAILED_TO_MERGE ***/
565 [CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION] =
566 "CONFLICT (submodule with possible resolution)",
567 [CONFLICT_SUBMODULE_NOT_INITIALIZED] =
568 "CONFLICT (submodule not initialized)",
569 [CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE] =
570 "CONFLICT (submodule history not available)",
571 [CONFLICT_SUBMODULE_MAY_HAVE_REWINDS] =
572 "CONFLICT (submodule may have rewinds)",
575 struct logical_conflict_info {
576 enum conflict_and_info_types type;
577 struct strvec paths;
580 /*** Function Grouping: various utility functions ***/
583 * For the next three macros, see warning for conflict_info.merged.
585 * In each of the below, mi is a struct merged_info*, and ci was defined
586 * as a struct conflict_info* (but we need to verify ci isn't actually
587 * pointed at a struct merged_info*).
589 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
590 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
591 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
593 #define INITIALIZE_CI(ci, mi) do { \
594 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
595 } while (0)
596 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
597 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
598 (ci) = (struct conflict_info *)(mi); \
599 assert((ci) && !(mi)->clean); \
600 } while (0)
602 static void free_strmap_strings(struct strmap *map)
604 struct hashmap_iter iter;
605 struct strmap_entry *entry;
607 strmap_for_each_entry(map, &iter, entry) {
608 free((char*)entry->key);
612 static void clear_or_reinit_internal_opts(struct merge_options_internal *opti,
613 int reinitialize)
615 struct rename_info *renames = &opti->renames;
616 int i;
617 void (*strmap_clear_func)(struct strmap *, int) =
618 reinitialize ? strmap_partial_clear : strmap_clear;
619 void (*strintmap_clear_func)(struct strintmap *) =
620 reinitialize ? strintmap_partial_clear : strintmap_clear;
621 void (*strset_clear_func)(struct strset *) =
622 reinitialize ? strset_partial_clear : strset_clear;
624 strmap_clear_func(&opti->paths, 0);
627 * All keys and values in opti->conflicted are a subset of those in
628 * opti->paths. We don't want to deallocate anything twice, so we
629 * don't free the keys and we pass 0 for free_values.
631 strmap_clear_func(&opti->conflicted, 0);
633 if (opti->attr_index.cache_nr) /* true iff opt->renormalize */
634 discard_index(&opti->attr_index);
636 /* Free memory used by various renames maps */
637 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) {
638 strintmap_clear_func(&renames->dirs_removed[i]);
639 strmap_clear_func(&renames->dir_renames[i], 0);
640 strintmap_clear_func(&renames->relevant_sources[i]);
641 if (!reinitialize)
642 assert(renames->cached_pairs_valid_side == 0);
643 if (i != renames->cached_pairs_valid_side &&
644 -1 != renames->cached_pairs_valid_side) {
645 strset_clear_func(&renames->cached_target_names[i]);
646 strmap_clear_func(&renames->cached_pairs[i], 1);
647 strset_clear_func(&renames->cached_irrelevant[i]);
648 partial_clear_dir_rename_count(&renames->dir_rename_count[i]);
649 if (!reinitialize)
650 strmap_clear(&renames->dir_rename_count[i], 1);
653 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) {
654 strintmap_clear_func(&renames->deferred[i].possible_trivial_merges);
655 strset_clear_func(&renames->deferred[i].target_dirs);
656 renames->deferred[i].trivial_merges_okay = 1; /* 1 == maybe */
658 renames->cached_pairs_valid_side = 0;
659 renames->dir_rename_mask = 0;
661 if (!reinitialize) {
662 struct hashmap_iter iter;
663 struct strmap_entry *e;
665 /* Release and free each strbuf found in output */
666 strmap_for_each_entry(&opti->conflicts, &iter, e) {
667 struct string_list *list = e->value;
668 for (int i = 0; i < list->nr; i++) {
669 struct logical_conflict_info *info =
670 list->items[i].util;
671 strvec_clear(&info->paths);
674 * While strictly speaking we don't need to
675 * free(conflicts) here because we could pass
676 * free_values=1 when calling strmap_clear() on
677 * opti->conflicts, that would require strmap_clear
678 * to do another strmap_for_each_entry() loop, so we
679 * just free it while we're iterating anyway.
681 string_list_clear(list, 1);
682 free(list);
684 strmap_clear(&opti->conflicts, 0);
687 mem_pool_discard(&opti->pool, 0);
689 /* Clean out callback_data as well. */
690 FREE_AND_NULL(renames->callback_data);
691 renames->callback_data_nr = renames->callback_data_alloc = 0;
694 __attribute__((format (printf, 2, 3)))
695 static int err(struct merge_options *opt, const char *err, ...)
697 va_list params;
698 struct strbuf sb = STRBUF_INIT;
700 strbuf_addstr(&sb, "error: ");
701 va_start(params, err);
702 strbuf_vaddf(&sb, err, params);
703 va_end(params);
705 error("%s", sb.buf);
706 strbuf_release(&sb);
708 return -1;
711 static void format_commit(struct strbuf *sb,
712 int indent,
713 struct repository *repo,
714 struct commit *commit)
716 struct merge_remote_desc *desc;
717 struct pretty_print_context ctx = {0};
718 ctx.abbrev = DEFAULT_ABBREV;
720 strbuf_addchars(sb, ' ', indent);
721 desc = merge_remote_util(commit);
722 if (desc) {
723 strbuf_addf(sb, "virtual %s\n", desc->name);
724 return;
727 repo_format_commit_message(repo, commit, "%h %s", sb, &ctx);
728 strbuf_addch(sb, '\n');
731 __attribute__((format (printf, 8, 9)))
732 static void path_msg(struct merge_options *opt,
733 enum conflict_and_info_types type,
734 int omittable_hint, /* skippable under --remerge-diff */
735 const char *primary_path,
736 const char *other_path_1, /* may be NULL */
737 const char *other_path_2, /* may be NULL */
738 struct string_list *other_paths, /* may be NULL */
739 const char *fmt, ...)
741 va_list ap;
742 struct string_list *path_conflicts;
743 struct logical_conflict_info *info;
744 struct strbuf buf = STRBUF_INIT;
745 struct strbuf *dest;
746 struct strbuf tmp = STRBUF_INIT;
748 /* Sanity checks */
749 assert(omittable_hint ==
750 !starts_with(type_short_descriptions[type], "CONFLICT") ||
751 type == CONFLICT_DIR_RENAME_SUGGESTED ||
752 type == CONFLICT_PRESENT_DESPITE_SKIPPED);
753 if (opt->record_conflict_msgs_as_headers && omittable_hint)
754 return; /* Do not record mere hints in headers */
755 if (opt->priv->call_depth && opt->verbosity < 5)
756 return; /* Ignore messages from inner merges */
758 /* Ensure path_conflicts (ptr to array of logical_conflict) allocated */
759 path_conflicts = strmap_get(&opt->priv->conflicts, primary_path);
760 if (!path_conflicts) {
761 path_conflicts = xmalloc(sizeof(*path_conflicts));
762 string_list_init_dup(path_conflicts);
763 strmap_put(&opt->priv->conflicts, primary_path, path_conflicts);
766 /* Add a logical_conflict at the end to store info from this call */
767 info = xcalloc(1, sizeof(*info));
768 info->type = type;
769 strvec_init(&info->paths);
771 /* Handle the list of paths */
772 strvec_push(&info->paths, primary_path);
773 if (other_path_1)
774 strvec_push(&info->paths, other_path_1);
775 if (other_path_2)
776 strvec_push(&info->paths, other_path_2);
777 if (other_paths)
778 for (int i = 0; i < other_paths->nr; i++)
779 strvec_push(&info->paths, other_paths->items[i].string);
781 /* Handle message and its format, in normal case */
782 dest = (opt->record_conflict_msgs_as_headers ? &tmp : &buf);
784 va_start(ap, fmt);
785 if (opt->priv->call_depth) {
786 strbuf_addchars(dest, ' ', 2);
787 strbuf_addstr(dest, "From inner merge:");
788 strbuf_addchars(dest, ' ', opt->priv->call_depth * 2);
790 strbuf_vaddf(dest, fmt, ap);
791 va_end(ap);
793 /* Handle specialized formatting of message under --remerge-diff */
794 if (opt->record_conflict_msgs_as_headers) {
795 int i_sb = 0, i_tmp = 0;
797 /* Start with the specified prefix */
798 if (opt->msg_header_prefix)
799 strbuf_addf(&buf, "%s ", opt->msg_header_prefix);
801 /* Copy tmp to sb, adding spaces after newlines */
802 strbuf_grow(&buf, buf.len + 2*tmp.len); /* more than sufficient */
803 for (; i_tmp < tmp.len; i_tmp++, i_sb++) {
804 /* Copy next character from tmp to sb */
805 buf.buf[buf.len + i_sb] = tmp.buf[i_tmp];
807 /* If we copied a newline, add a space */
808 if (tmp.buf[i_tmp] == '\n')
809 buf.buf[++i_sb] = ' ';
811 /* Update length and ensure it's NUL-terminated */
812 buf.len += i_sb;
813 buf.buf[buf.len] = '\0';
815 strbuf_release(&tmp);
817 string_list_append_nodup(path_conflicts, strbuf_detach(&buf, NULL))
818 ->util = info;
821 static struct diff_filespec *pool_alloc_filespec(struct mem_pool *pool,
822 const char *path)
824 /* Similar to alloc_filespec(), but allocate from pool and reuse path */
825 struct diff_filespec *spec;
827 spec = mem_pool_calloc(pool, 1, sizeof(*spec));
828 spec->path = (char*)path; /* spec won't modify it */
830 spec->count = 1;
831 spec->is_binary = -1;
832 return spec;
835 static struct diff_filepair *pool_diff_queue(struct mem_pool *pool,
836 struct diff_queue_struct *queue,
837 struct diff_filespec *one,
838 struct diff_filespec *two)
840 /* Same code as diff_queue(), except allocate from pool */
841 struct diff_filepair *dp;
843 dp = mem_pool_calloc(pool, 1, sizeof(*dp));
844 dp->one = one;
845 dp->two = two;
846 if (queue)
847 diff_q(queue, dp);
848 return dp;
851 /* add a string to a strbuf, but converting "/" to "_" */
852 static void add_flattened_path(struct strbuf *out, const char *s)
854 size_t i = out->len;
855 strbuf_addstr(out, s);
856 for (; i < out->len; i++)
857 if (out->buf[i] == '/')
858 out->buf[i] = '_';
861 static char *unique_path(struct merge_options *opt,
862 const char *path,
863 const char *branch)
865 char *ret = NULL;
866 struct strbuf newpath = STRBUF_INIT;
867 int suffix = 0;
868 size_t base_len;
869 struct strmap *existing_paths = &opt->priv->paths;
871 strbuf_addf(&newpath, "%s~", path);
872 add_flattened_path(&newpath, branch);
874 base_len = newpath.len;
875 while (strmap_contains(existing_paths, newpath.buf)) {
876 strbuf_setlen(&newpath, base_len);
877 strbuf_addf(&newpath, "_%d", suffix++);
880 /* Track the new path in our memory pool */
881 ret = mem_pool_alloc(&opt->priv->pool, newpath.len + 1);
882 memcpy(ret, newpath.buf, newpath.len + 1);
883 strbuf_release(&newpath);
884 return ret;
887 /*** Function Grouping: functions related to collect_merge_info() ***/
889 static int traverse_trees_wrapper_callback(int n,
890 unsigned long mask,
891 unsigned long dirmask,
892 struct name_entry *names,
893 struct traverse_info *info)
895 struct merge_options *opt = info->data;
896 struct rename_info *renames = &opt->priv->renames;
897 unsigned filemask = mask & ~dirmask;
899 assert(n==3);
901 if (!renames->callback_data_traverse_path)
902 renames->callback_data_traverse_path = xstrdup(info->traverse_path);
904 if (filemask && filemask == renames->dir_rename_mask)
905 renames->dir_rename_mask = 0x07;
907 ALLOC_GROW(renames->callback_data, renames->callback_data_nr + 1,
908 renames->callback_data_alloc);
909 renames->callback_data[renames->callback_data_nr].mask = mask;
910 renames->callback_data[renames->callback_data_nr].dirmask = dirmask;
911 COPY_ARRAY(renames->callback_data[renames->callback_data_nr].names,
912 names, 3);
913 renames->callback_data_nr++;
915 return mask;
919 * Much like traverse_trees(), BUT:
920 * - read all the tree entries FIRST, saving them
921 * - note that the above step provides an opportunity to compute necessary
922 * additional details before the "real" traversal
923 * - loop through the saved entries and call the original callback on them
925 static int traverse_trees_wrapper(struct index_state *istate,
926 int n,
927 struct tree_desc *t,
928 struct traverse_info *info)
930 int ret, i, old_offset;
931 traverse_callback_t old_fn;
932 char *old_callback_data_traverse_path;
933 struct merge_options *opt = info->data;
934 struct rename_info *renames = &opt->priv->renames;
936 assert(renames->dir_rename_mask == 2 || renames->dir_rename_mask == 4);
938 old_callback_data_traverse_path = renames->callback_data_traverse_path;
939 old_fn = info->fn;
940 old_offset = renames->callback_data_nr;
942 renames->callback_data_traverse_path = NULL;
943 info->fn = traverse_trees_wrapper_callback;
944 ret = traverse_trees(istate, n, t, info);
945 if (ret < 0)
946 return ret;
948 info->traverse_path = renames->callback_data_traverse_path;
949 info->fn = old_fn;
950 for (i = old_offset; i < renames->callback_data_nr; ++i) {
951 info->fn(n,
952 renames->callback_data[i].mask,
953 renames->callback_data[i].dirmask,
954 renames->callback_data[i].names,
955 info);
958 renames->callback_data_nr = old_offset;
959 free(renames->callback_data_traverse_path);
960 renames->callback_data_traverse_path = old_callback_data_traverse_path;
961 info->traverse_path = NULL;
962 return 0;
965 static void setup_path_info(struct merge_options *opt,
966 struct string_list_item *result,
967 const char *current_dir_name,
968 int current_dir_name_len,
969 char *fullpath, /* we'll take over ownership */
970 struct name_entry *names,
971 struct name_entry *merged_version,
972 unsigned is_null, /* boolean */
973 unsigned df_conflict, /* boolean */
974 unsigned filemask,
975 unsigned dirmask,
976 int resolved /* boolean */)
978 /* result->util is void*, so mi is a convenience typed variable */
979 struct merged_info *mi;
981 assert(!is_null || resolved);
982 assert(!df_conflict || !resolved); /* df_conflict implies !resolved */
983 assert(resolved == (merged_version != NULL));
985 mi = mem_pool_calloc(&opt->priv->pool, 1,
986 resolved ? sizeof(struct merged_info) :
987 sizeof(struct conflict_info));
988 mi->directory_name = current_dir_name;
989 mi->basename_offset = current_dir_name_len;
990 mi->clean = !!resolved;
991 if (resolved) {
992 mi->result.mode = merged_version->mode;
993 oidcpy(&mi->result.oid, &merged_version->oid);
994 mi->is_null = !!is_null;
995 } else {
996 int i;
997 struct conflict_info *ci;
999 ASSIGN_AND_VERIFY_CI(ci, mi);
1000 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
1001 ci->pathnames[i] = fullpath;
1002 ci->stages[i].mode = names[i].mode;
1003 oidcpy(&ci->stages[i].oid, &names[i].oid);
1005 ci->filemask = filemask;
1006 ci->dirmask = dirmask;
1007 ci->df_conflict = !!df_conflict;
1008 if (dirmask)
1010 * Assume is_null for now, but if we have entries
1011 * under the directory then when it is complete in
1012 * write_completed_directory() it'll update this.
1013 * Also, for D/F conflicts, we have to handle the
1014 * directory first, then clear this bit and process
1015 * the file to see how it is handled -- that occurs
1016 * near the top of process_entry().
1018 mi->is_null = 1;
1020 strmap_put(&opt->priv->paths, fullpath, mi);
1021 result->string = fullpath;
1022 result->util = mi;
1025 static void add_pair(struct merge_options *opt,
1026 struct name_entry *names,
1027 const char *pathname,
1028 unsigned side,
1029 unsigned is_add /* if false, is_delete */,
1030 unsigned match_mask,
1031 unsigned dir_rename_mask)
1033 struct diff_filespec *one, *two;
1034 struct rename_info *renames = &opt->priv->renames;
1035 int names_idx = is_add ? side : 0;
1037 if (is_add) {
1038 assert(match_mask == 0 || match_mask == 6);
1039 if (strset_contains(&renames->cached_target_names[side],
1040 pathname))
1041 return;
1042 } else {
1043 unsigned content_relevant = (match_mask == 0);
1044 unsigned location_relevant = (dir_rename_mask == 0x07);
1046 assert(match_mask == 0 || match_mask == 3 || match_mask == 5);
1049 * If pathname is found in cached_irrelevant[side] due to
1050 * previous pick but for this commit content is relevant,
1051 * then we need to remove it from cached_irrelevant.
1053 if (content_relevant)
1054 /* strset_remove is no-op if strset doesn't have key */
1055 strset_remove(&renames->cached_irrelevant[side],
1056 pathname);
1059 * We do not need to re-detect renames for paths that we already
1060 * know the pairing, i.e. for cached_pairs (or
1061 * cached_irrelevant). However, handle_deferred_entries() needs
1062 * to loop over the union of keys from relevant_sources[side] and
1063 * cached_pairs[side], so for simplicity we set relevant_sources
1064 * for all the cached_pairs too and then strip them back out in
1065 * prune_cached_from_relevant() at the beginning of
1066 * detect_regular_renames().
1068 if (content_relevant || location_relevant) {
1069 /* content_relevant trumps location_relevant */
1070 strintmap_set(&renames->relevant_sources[side], pathname,
1071 content_relevant ? RELEVANT_CONTENT : RELEVANT_LOCATION);
1075 * Avoid creating pair if we've already cached rename results.
1076 * Note that we do this after setting relevant_sources[side]
1077 * as noted in the comment above.
1079 if (strmap_contains(&renames->cached_pairs[side], pathname) ||
1080 strset_contains(&renames->cached_irrelevant[side], pathname))
1081 return;
1084 one = pool_alloc_filespec(&opt->priv->pool, pathname);
1085 two = pool_alloc_filespec(&opt->priv->pool, pathname);
1086 fill_filespec(is_add ? two : one,
1087 &names[names_idx].oid, 1, names[names_idx].mode);
1088 pool_diff_queue(&opt->priv->pool, &renames->pairs[side], one, two);
1091 static void collect_rename_info(struct merge_options *opt,
1092 struct name_entry *names,
1093 const char *dirname,
1094 const char *fullname,
1095 unsigned filemask,
1096 unsigned dirmask,
1097 unsigned match_mask)
1099 struct rename_info *renames = &opt->priv->renames;
1100 unsigned side;
1103 * Update dir_rename_mask (determines ignore-rename-source validity)
1105 * dir_rename_mask helps us keep track of when directory rename
1106 * detection may be relevant. Basically, whenver a directory is
1107 * removed on one side of history, and a file is added to that
1108 * directory on the other side of history, directory rename
1109 * detection is relevant (meaning we have to detect renames for all
1110 * files within that directory to deduce where the directory
1111 * moved). Also, whenever a directory needs directory rename
1112 * detection, due to the "majority rules" choice for where to move
1113 * it (see t6423 testcase 1f), we also need to detect renames for
1114 * all files within subdirectories of that directory as well.
1116 * Here we haven't looked at files within the directory yet, we are
1117 * just looking at the directory itself. So, if we aren't yet in
1118 * a case where a parent directory needed directory rename detection
1119 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
1120 * on one side of history, record the mask of the other side of
1121 * history in dir_rename_mask.
1123 if (renames->dir_rename_mask != 0x07 &&
1124 (dirmask == 3 || dirmask == 5)) {
1125 /* simple sanity check */
1126 assert(renames->dir_rename_mask == 0 ||
1127 renames->dir_rename_mask == (dirmask & ~1));
1128 /* update dir_rename_mask; have it record mask of new side */
1129 renames->dir_rename_mask = (dirmask & ~1);
1132 /* Update dirs_removed, as needed */
1133 if (dirmask == 1 || dirmask == 3 || dirmask == 5) {
1134 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1135 unsigned sides = (0x07 - dirmask)/2;
1136 unsigned relevance = (renames->dir_rename_mask == 0x07) ?
1137 RELEVANT_FOR_ANCESTOR : NOT_RELEVANT;
1139 * Record relevance of this directory. However, note that
1140 * when collect_merge_info_callback() recurses into this
1141 * directory and calls collect_rename_info() on paths
1142 * within that directory, if we find a path that was added
1143 * to this directory on the other side of history, we will
1144 * upgrade this value to RELEVANT_FOR_SELF; see below.
1146 if (sides & 1)
1147 strintmap_set(&renames->dirs_removed[1], fullname,
1148 relevance);
1149 if (sides & 2)
1150 strintmap_set(&renames->dirs_removed[2], fullname,
1151 relevance);
1155 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1156 * When we run across a file added to a directory. In such a case,
1157 * find the directory of the file and upgrade its relevance.
1159 if (renames->dir_rename_mask == 0x07 &&
1160 (filemask == 2 || filemask == 4)) {
1162 * Need directory rename for parent directory on other side
1163 * of history from added file. Thus
1164 * side = (~filemask & 0x06) >> 1
1165 * or
1166 * side = 3 - (filemask/2).
1168 unsigned side = 3 - (filemask >> 1);
1169 strintmap_set(&renames->dirs_removed[side], dirname,
1170 RELEVANT_FOR_SELF);
1173 if (filemask == 0 || filemask == 7)
1174 return;
1176 for (side = MERGE_SIDE1; side <= MERGE_SIDE2; ++side) {
1177 unsigned side_mask = (1 << side);
1179 /* Check for deletion on side */
1180 if ((filemask & 1) && !(filemask & side_mask))
1181 add_pair(opt, names, fullname, side, 0 /* delete */,
1182 match_mask & filemask,
1183 renames->dir_rename_mask);
1185 /* Check for addition on side */
1186 if (!(filemask & 1) && (filemask & side_mask))
1187 add_pair(opt, names, fullname, side, 1 /* add */,
1188 match_mask & filemask,
1189 renames->dir_rename_mask);
1193 static int collect_merge_info_callback(int n,
1194 unsigned long mask,
1195 unsigned long dirmask,
1196 struct name_entry *names,
1197 struct traverse_info *info)
1200 * n is 3. Always.
1201 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1202 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1203 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1205 struct merge_options *opt = info->data;
1206 struct merge_options_internal *opti = opt->priv;
1207 struct rename_info *renames = &opt->priv->renames;
1208 struct string_list_item pi; /* Path Info */
1209 struct conflict_info *ci; /* typed alias to pi.util (which is void*) */
1210 struct name_entry *p;
1211 size_t len;
1212 char *fullpath;
1213 const char *dirname = opti->current_dir_name;
1214 unsigned prev_dir_rename_mask = renames->dir_rename_mask;
1215 unsigned filemask = mask & ~dirmask;
1216 unsigned match_mask = 0; /* will be updated below */
1217 unsigned mbase_null = !(mask & 1);
1218 unsigned side1_null = !(mask & 2);
1219 unsigned side2_null = !(mask & 4);
1220 unsigned side1_matches_mbase = (!side1_null && !mbase_null &&
1221 names[0].mode == names[1].mode &&
1222 oideq(&names[0].oid, &names[1].oid));
1223 unsigned side2_matches_mbase = (!side2_null && !mbase_null &&
1224 names[0].mode == names[2].mode &&
1225 oideq(&names[0].oid, &names[2].oid));
1226 unsigned sides_match = (!side1_null && !side2_null &&
1227 names[1].mode == names[2].mode &&
1228 oideq(&names[1].oid, &names[2].oid));
1231 * Note: When a path is a file on one side of history and a directory
1232 * in another, we have a directory/file conflict. In such cases, if
1233 * the conflict doesn't resolve from renames and deletions, then we
1234 * always leave directories where they are and move files out of the
1235 * way. Thus, while struct conflict_info has a df_conflict field to
1236 * track such conflicts, we ignore that field for any directories at
1237 * a path and only pay attention to it for files at the given path.
1238 * The fact that we leave directories were they are also means that
1239 * we do not need to worry about getting additional df_conflict
1240 * information propagated from parent directories down to children
1241 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1242 * sets a newinfo.df_conflicts field specifically to propagate it).
1244 unsigned df_conflict = (filemask != 0) && (dirmask != 0);
1246 /* n = 3 is a fundamental assumption. */
1247 if (n != 3)
1248 BUG("Called collect_merge_info_callback wrong");
1251 * A bunch of sanity checks verifying that traverse_trees() calls
1252 * us the way I expect. Could just remove these at some point,
1253 * though maybe they are helpful to future code readers.
1255 assert(mbase_null == is_null_oid(&names[0].oid));
1256 assert(side1_null == is_null_oid(&names[1].oid));
1257 assert(side2_null == is_null_oid(&names[2].oid));
1258 assert(!mbase_null || !side1_null || !side2_null);
1259 assert(mask > 0 && mask < 8);
1261 /* Determine match_mask */
1262 if (side1_matches_mbase)
1263 match_mask = (side2_matches_mbase ? 7 : 3);
1264 else if (side2_matches_mbase)
1265 match_mask = 5;
1266 else if (sides_match)
1267 match_mask = 6;
1270 * Get the name of the relevant filepath, which we'll pass to
1271 * setup_path_info() for tracking.
1273 p = names;
1274 while (!p->mode)
1275 p++;
1276 len = traverse_path_len(info, p->pathlen);
1278 /* +1 in both of the following lines to include the NUL byte */
1279 fullpath = mem_pool_alloc(&opt->priv->pool, len + 1);
1280 make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen);
1283 * If mbase, side1, and side2 all match, we can resolve early. Even
1284 * if these are trees, there will be no renames or anything
1285 * underneath.
1287 if (side1_matches_mbase && side2_matches_mbase) {
1288 /* mbase, side1, & side2 all match; use mbase as resolution */
1289 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
1290 names, names+0, mbase_null, 0 /* df_conflict */,
1291 filemask, dirmask, 1 /* resolved */);
1292 return mask;
1296 * If the sides match, and all three paths are present and are
1297 * files, then we can take either as the resolution. We can't do
1298 * this with trees, because there may be rename sources from the
1299 * merge_base.
1301 if (sides_match && filemask == 0x07) {
1302 /* use side1 (== side2) version as resolution */
1303 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
1304 names, names+1, side1_null, 0,
1305 filemask, dirmask, 1);
1306 return mask;
1310 * If side1 matches mbase and all three paths are present and are
1311 * files, then we can use side2 as the resolution. We cannot
1312 * necessarily do so this for trees, because there may be rename
1313 * destinations within side2.
1315 if (side1_matches_mbase && filemask == 0x07) {
1316 /* use side2 version as resolution */
1317 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
1318 names, names+2, side2_null, 0,
1319 filemask, dirmask, 1);
1320 return mask;
1323 /* Similar to above but swapping sides 1 and 2 */
1324 if (side2_matches_mbase && filemask == 0x07) {
1325 /* use side1 version as resolution */
1326 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
1327 names, names+1, side1_null, 0,
1328 filemask, dirmask, 1);
1329 return mask;
1333 * Sometimes we can tell that a source path need not be included in
1334 * rename detection -- namely, whenever either
1335 * side1_matches_mbase && side2_null
1336 * or
1337 * side2_matches_mbase && side1_null
1338 * However, we call collect_rename_info() even in those cases,
1339 * because exact renames are cheap and would let us remove both a
1340 * source and destination path. We'll cull the unneeded sources
1341 * later.
1343 collect_rename_info(opt, names, dirname, fullpath,
1344 filemask, dirmask, match_mask);
1347 * None of the special cases above matched, so we have a
1348 * provisional conflict. (Rename detection might allow us to
1349 * unconflict some more cases, but that comes later so all we can
1350 * do now is record the different non-null file hashes.)
1352 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
1353 names, NULL, 0, df_conflict, filemask, dirmask, 0);
1355 ci = pi.util;
1356 VERIFY_CI(ci);
1357 ci->match_mask = match_mask;
1359 /* If dirmask, recurse into subdirectories */
1360 if (dirmask) {
1361 struct traverse_info newinfo;
1362 struct tree_desc t[3];
1363 void *buf[3] = {NULL, NULL, NULL};
1364 const char *original_dir_name;
1365 int i, ret, side;
1368 * Check for whether we can avoid recursing due to one side
1369 * matching the merge base. The side that does NOT match is
1370 * the one that might have a rename destination we need.
1372 assert(!side1_matches_mbase || !side2_matches_mbase);
1373 side = side1_matches_mbase ? MERGE_SIDE2 :
1374 side2_matches_mbase ? MERGE_SIDE1 : MERGE_BASE;
1375 if (filemask == 0 && (dirmask == 2 || dirmask == 4)) {
1377 * Also defer recursing into new directories; set up a
1378 * few variables to let us do so.
1380 ci->match_mask = (7 - dirmask);
1381 side = dirmask / 2;
1383 if (renames->dir_rename_mask != 0x07 &&
1384 side != MERGE_BASE &&
1385 renames->deferred[side].trivial_merges_okay &&
1386 !strset_contains(&renames->deferred[side].target_dirs,
1387 pi.string)) {
1388 strintmap_set(&renames->deferred[side].possible_trivial_merges,
1389 pi.string, renames->dir_rename_mask);
1390 renames->dir_rename_mask = prev_dir_rename_mask;
1391 return mask;
1394 /* We need to recurse */
1395 ci->match_mask &= filemask;
1396 newinfo = *info;
1397 newinfo.prev = info;
1398 newinfo.name = p->path;
1399 newinfo.namelen = p->pathlen;
1400 newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1);
1402 * If this directory we are about to recurse into cared about
1403 * its parent directory (the current directory) having a D/F
1404 * conflict, then we'd propagate the masks in this way:
1405 * newinfo.df_conflicts |= (mask & ~dirmask);
1406 * But we don't worry about propagating D/F conflicts. (See
1407 * comment near setting of local df_conflict variable near
1408 * the beginning of this function).
1411 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
1412 if (i == 1 && side1_matches_mbase)
1413 t[1] = t[0];
1414 else if (i == 2 && side2_matches_mbase)
1415 t[2] = t[0];
1416 else if (i == 2 && sides_match)
1417 t[2] = t[1];
1418 else {
1419 const struct object_id *oid = NULL;
1420 if (dirmask & 1)
1421 oid = &names[i].oid;
1422 buf[i] = fill_tree_descriptor(opt->repo,
1423 t + i, oid);
1425 dirmask >>= 1;
1428 original_dir_name = opti->current_dir_name;
1429 opti->current_dir_name = pi.string;
1430 if (renames->dir_rename_mask == 0 ||
1431 renames->dir_rename_mask == 0x07)
1432 ret = traverse_trees(NULL, 3, t, &newinfo);
1433 else
1434 ret = traverse_trees_wrapper(NULL, 3, t, &newinfo);
1435 opti->current_dir_name = original_dir_name;
1436 renames->dir_rename_mask = prev_dir_rename_mask;
1438 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++)
1439 free(buf[i]);
1441 if (ret < 0)
1442 return -1;
1445 return mask;
1448 static void resolve_trivial_directory_merge(struct conflict_info *ci, int side)
1450 VERIFY_CI(ci);
1451 assert((side == 1 && ci->match_mask == 5) ||
1452 (side == 2 && ci->match_mask == 3));
1453 oidcpy(&ci->merged.result.oid, &ci->stages[side].oid);
1454 ci->merged.result.mode = ci->stages[side].mode;
1455 ci->merged.is_null = is_null_oid(&ci->stages[side].oid);
1456 ci->match_mask = 0;
1457 ci->merged.clean = 1; /* (ci->filemask == 0); */
1460 static int handle_deferred_entries(struct merge_options *opt,
1461 struct traverse_info *info)
1463 struct rename_info *renames = &opt->priv->renames;
1464 struct hashmap_iter iter;
1465 struct strmap_entry *entry;
1466 int side, ret = 0;
1467 int path_count_before, path_count_after = 0;
1469 path_count_before = strmap_get_size(&opt->priv->paths);
1470 for (side = MERGE_SIDE1; side <= MERGE_SIDE2; side++) {
1471 unsigned optimization_okay = 1;
1472 struct strintmap copy;
1474 /* Loop over the set of paths we need to know rename info for */
1475 strset_for_each_entry(&renames->relevant_sources[side],
1476 &iter, entry) {
1477 char *rename_target, *dir, *dir_marker;
1478 struct strmap_entry *e;
1481 * If we don't know delete/rename info for this path,
1482 * then we need to recurse into all trees to get all
1483 * adds to make sure we have it.
1485 if (strset_contains(&renames->cached_irrelevant[side],
1486 entry->key))
1487 continue;
1488 e = strmap_get_entry(&renames->cached_pairs[side],
1489 entry->key);
1490 if (!e) {
1491 optimization_okay = 0;
1492 break;
1495 /* If this is a delete, we have enough info already */
1496 rename_target = e->value;
1497 if (!rename_target)
1498 continue;
1500 /* If we already walked the rename target, we're good */
1501 if (strmap_contains(&opt->priv->paths, rename_target))
1502 continue;
1505 * Otherwise, we need to get a list of directories that
1506 * will need to be recursed into to get this
1507 * rename_target.
1509 dir = xstrdup(rename_target);
1510 while ((dir_marker = strrchr(dir, '/'))) {
1511 *dir_marker = '\0';
1512 if (strset_contains(&renames->deferred[side].target_dirs,
1513 dir))
1514 break;
1515 strset_add(&renames->deferred[side].target_dirs,
1516 dir);
1518 free(dir);
1520 renames->deferred[side].trivial_merges_okay = optimization_okay;
1522 * We need to recurse into any directories in
1523 * possible_trivial_merges[side] found in target_dirs[side].
1524 * But when we recurse, we may need to queue up some of the
1525 * subdirectories for possible_trivial_merges[side]. Since
1526 * we can't safely iterate through a hashmap while also adding
1527 * entries, move the entries into 'copy', iterate over 'copy',
1528 * and then we'll also iterate anything added into
1529 * possible_trivial_merges[side] once this loop is done.
1531 copy = renames->deferred[side].possible_trivial_merges;
1532 strintmap_init_with_options(&renames->deferred[side].possible_trivial_merges,
1534 &opt->priv->pool,
1536 strintmap_for_each_entry(&copy, &iter, entry) {
1537 const char *path = entry->key;
1538 unsigned dir_rename_mask = (intptr_t)entry->value;
1539 struct conflict_info *ci;
1540 unsigned dirmask;
1541 struct tree_desc t[3];
1542 void *buf[3] = {NULL,};
1543 int i;
1545 ci = strmap_get(&opt->priv->paths, path);
1546 VERIFY_CI(ci);
1547 dirmask = ci->dirmask;
1549 if (optimization_okay &&
1550 !strset_contains(&renames->deferred[side].target_dirs,
1551 path)) {
1552 resolve_trivial_directory_merge(ci, side);
1553 continue;
1556 info->name = path;
1557 info->namelen = strlen(path);
1558 info->pathlen = info->namelen + 1;
1560 for (i = 0; i < 3; i++, dirmask >>= 1) {
1561 if (i == 1 && ci->match_mask == 3)
1562 t[1] = t[0];
1563 else if (i == 2 && ci->match_mask == 5)
1564 t[2] = t[0];
1565 else if (i == 2 && ci->match_mask == 6)
1566 t[2] = t[1];
1567 else {
1568 const struct object_id *oid = NULL;
1569 if (dirmask & 1)
1570 oid = &ci->stages[i].oid;
1571 buf[i] = fill_tree_descriptor(opt->repo,
1572 t+i, oid);
1576 ci->match_mask &= ci->filemask;
1577 opt->priv->current_dir_name = path;
1578 renames->dir_rename_mask = dir_rename_mask;
1579 if (renames->dir_rename_mask == 0 ||
1580 renames->dir_rename_mask == 0x07)
1581 ret = traverse_trees(NULL, 3, t, info);
1582 else
1583 ret = traverse_trees_wrapper(NULL, 3, t, info);
1585 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++)
1586 free(buf[i]);
1588 if (ret < 0)
1589 return ret;
1591 strintmap_clear(&copy);
1592 strintmap_for_each_entry(&renames->deferred[side].possible_trivial_merges,
1593 &iter, entry) {
1594 const char *path = entry->key;
1595 struct conflict_info *ci;
1597 ci = strmap_get(&opt->priv->paths, path);
1598 VERIFY_CI(ci);
1600 assert(renames->deferred[side].trivial_merges_okay &&
1601 !strset_contains(&renames->deferred[side].target_dirs,
1602 path));
1603 resolve_trivial_directory_merge(ci, side);
1605 if (!optimization_okay || path_count_after)
1606 path_count_after = strmap_get_size(&opt->priv->paths);
1608 if (path_count_after) {
1610 * The choice of wanted_factor here does not affect
1611 * correctness, only performance. When the
1612 * path_count_after / path_count_before
1613 * ratio is high, redoing after renames is a big
1614 * performance boost. I suspect that redoing is a wash
1615 * somewhere near a value of 2, and below that redoing will
1616 * slow things down. I applied a fudge factor and picked
1617 * 3; see the commit message when this was introduced for
1618 * back of the envelope calculations for this ratio.
1620 const int wanted_factor = 3;
1622 /* We should only redo collect_merge_info one time */
1623 assert(renames->redo_after_renames == 0);
1625 if (path_count_after / path_count_before >= wanted_factor) {
1626 renames->redo_after_renames = 1;
1627 renames->cached_pairs_valid_side = -1;
1629 } else if (renames->redo_after_renames == 2)
1630 renames->redo_after_renames = 0;
1631 return ret;
1634 static int collect_merge_info(struct merge_options *opt,
1635 struct tree *merge_base,
1636 struct tree *side1,
1637 struct tree *side2)
1639 int ret;
1640 struct tree_desc t[3];
1641 struct traverse_info info;
1643 opt->priv->toplevel_dir = "";
1644 opt->priv->current_dir_name = opt->priv->toplevel_dir;
1645 setup_traverse_info(&info, opt->priv->toplevel_dir);
1646 info.fn = collect_merge_info_callback;
1647 info.data = opt;
1648 info.show_all_errors = 1;
1650 parse_tree(merge_base);
1651 parse_tree(side1);
1652 parse_tree(side2);
1653 init_tree_desc(t + 0, merge_base->buffer, merge_base->size);
1654 init_tree_desc(t + 1, side1->buffer, side1->size);
1655 init_tree_desc(t + 2, side2->buffer, side2->size);
1657 trace2_region_enter("merge", "traverse_trees", opt->repo);
1658 ret = traverse_trees(NULL, 3, t, &info);
1659 if (ret == 0)
1660 ret = handle_deferred_entries(opt, &info);
1661 trace2_region_leave("merge", "traverse_trees", opt->repo);
1663 return ret;
1666 /*** Function Grouping: functions related to threeway content merges ***/
1668 static int find_first_merges(struct repository *repo,
1669 const char *path,
1670 struct commit *a,
1671 struct commit *b,
1672 struct object_array *result)
1674 int i, j;
1675 struct object_array merges = OBJECT_ARRAY_INIT;
1676 struct commit *commit;
1677 int contains_another;
1679 char merged_revision[GIT_MAX_HEXSZ + 2];
1680 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1681 "--all", merged_revision, NULL };
1682 struct rev_info revs;
1683 struct setup_revision_opt rev_opts;
1685 memset(result, 0, sizeof(struct object_array));
1686 memset(&rev_opts, 0, sizeof(rev_opts));
1688 /* get all revisions that merge commit a */
1689 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1690 oid_to_hex(&a->object.oid));
1691 repo_init_revisions(repo, &revs, NULL);
1692 /* FIXME: can't handle linked worktrees in submodules yet */
1693 revs.single_worktree = path != NULL;
1694 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1696 /* save all revisions from the above list that contain b */
1697 if (prepare_revision_walk(&revs))
1698 die("revision walk setup failed");
1699 while ((commit = get_revision(&revs)) != NULL) {
1700 struct object *o = &(commit->object);
1701 if (repo_in_merge_bases(repo, b, commit))
1702 add_object_array(o, NULL, &merges);
1704 reset_revision_walk();
1706 /* Now we've got all merges that contain a and b. Prune all
1707 * merges that contain another found merge and save them in
1708 * result.
1710 for (i = 0; i < merges.nr; i++) {
1711 struct commit *m1 = (struct commit *) merges.objects[i].item;
1713 contains_another = 0;
1714 for (j = 0; j < merges.nr; j++) {
1715 struct commit *m2 = (struct commit *) merges.objects[j].item;
1716 if (i != j && repo_in_merge_bases(repo, m2, m1)) {
1717 contains_another = 1;
1718 break;
1722 if (!contains_another)
1723 add_object_array(merges.objects[i].item, NULL, result);
1726 object_array_clear(&merges);
1727 release_revisions(&revs);
1728 return result->nr;
1731 static int merge_submodule(struct merge_options *opt,
1732 const char *path,
1733 const struct object_id *o,
1734 const struct object_id *a,
1735 const struct object_id *b,
1736 struct object_id *result)
1738 struct repository subrepo;
1739 struct strbuf sb = STRBUF_INIT;
1740 int ret = 0;
1741 struct commit *commit_o, *commit_a, *commit_b;
1742 int parent_count;
1743 struct object_array merges;
1745 int i;
1746 int search = !opt->priv->call_depth;
1748 /* store fallback answer in result in case we fail */
1749 oidcpy(result, opt->priv->call_depth ? o : a);
1751 /* we can not handle deletion conflicts */
1752 if (is_null_oid(o))
1753 return 0;
1754 if (is_null_oid(a))
1755 return 0;
1756 if (is_null_oid(b))
1757 return 0;
1759 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1760 path_msg(opt, CONFLICT_SUBMODULE_NOT_INITIALIZED, 0,
1761 path, NULL, NULL, NULL,
1762 _("Failed to merge submodule %s (not checked out)"),
1763 path);
1764 return 0;
1767 if (!(commit_o = lookup_commit_reference(&subrepo, o)) ||
1768 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1769 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1770 path_msg(opt, CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE, 0,
1771 path, NULL, NULL, NULL,
1772 _("Failed to merge submodule %s (commits not present)"),
1773 path);
1774 goto cleanup;
1777 /* check whether both changes are forward */
1778 if (!repo_in_merge_bases(&subrepo, commit_o, commit_a) ||
1779 !repo_in_merge_bases(&subrepo, commit_o, commit_b)) {
1780 path_msg(opt, CONFLICT_SUBMODULE_MAY_HAVE_REWINDS, 0,
1781 path, NULL, NULL, NULL,
1782 _("Failed to merge submodule %s "
1783 "(commits don't follow merge-base)"),
1784 path);
1785 goto cleanup;
1788 /* Case #1: a is contained in b or vice versa */
1789 if (repo_in_merge_bases(&subrepo, commit_a, commit_b)) {
1790 oidcpy(result, b);
1791 path_msg(opt, INFO_SUBMODULE_FAST_FORWARDING, 1,
1792 path, NULL, NULL, NULL,
1793 _("Note: Fast-forwarding submodule %s to %s"),
1794 path, oid_to_hex(b));
1795 ret = 1;
1796 goto cleanup;
1798 if (repo_in_merge_bases(&subrepo, commit_b, commit_a)) {
1799 oidcpy(result, a);
1800 path_msg(opt, INFO_SUBMODULE_FAST_FORWARDING, 1,
1801 path, NULL, NULL, NULL,
1802 _("Note: Fast-forwarding submodule %s to %s"),
1803 path, oid_to_hex(a));
1804 ret = 1;
1805 goto cleanup;
1809 * Case #2: There are one or more merges that contain a and b in
1810 * the submodule. If there is only one, then present it as a
1811 * suggestion to the user, but leave it marked unmerged so the
1812 * user needs to confirm the resolution.
1815 /* Skip the search if makes no sense to the calling context. */
1816 if (!search)
1817 goto cleanup;
1819 /* find commit which merges them */
1820 parent_count = find_first_merges(&subrepo, path, commit_a, commit_b,
1821 &merges);
1822 switch (parent_count) {
1823 case 0:
1824 path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE, 0,
1825 path, NULL, NULL, NULL,
1826 _("Failed to merge submodule %s"), path);
1827 break;
1829 case 1:
1830 format_commit(&sb, 4, &subrepo,
1831 (struct commit *)merges.objects[0].item);
1832 path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION, 0,
1833 path, NULL, NULL, NULL,
1834 _("Failed to merge submodule %s, but a possible merge "
1835 "resolution exists: %s"),
1836 path, sb.buf);
1837 strbuf_release(&sb);
1838 break;
1839 default:
1840 for (i = 0; i < merges.nr; i++)
1841 format_commit(&sb, 4, &subrepo,
1842 (struct commit *)merges.objects[i].item);
1843 path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION, 0,
1844 path, NULL, NULL, NULL,
1845 _("Failed to merge submodule %s, but multiple "
1846 "possible merges exist:\n%s"), path, sb.buf);
1847 strbuf_release(&sb);
1850 object_array_clear(&merges);
1851 cleanup:
1852 repo_clear(&subrepo);
1853 return ret;
1856 static void initialize_attr_index(struct merge_options *opt)
1859 * The renormalize_buffer() functions require attributes, and
1860 * annoyingly those can only be read from the working tree or from
1861 * an index_state. merge-ort doesn't have an index_state, so we
1862 * generate a fake one containing only attribute information.
1864 struct merged_info *mi;
1865 struct index_state *attr_index = &opt->priv->attr_index;
1866 struct cache_entry *ce;
1868 attr_index->initialized = 1;
1870 if (!opt->renormalize)
1871 return;
1873 mi = strmap_get(&opt->priv->paths, GITATTRIBUTES_FILE);
1874 if (!mi)
1875 return;
1877 if (mi->clean) {
1878 int len = strlen(GITATTRIBUTES_FILE);
1879 ce = make_empty_cache_entry(attr_index, len);
1880 ce->ce_mode = create_ce_mode(mi->result.mode);
1881 ce->ce_flags = create_ce_flags(0);
1882 ce->ce_namelen = len;
1883 oidcpy(&ce->oid, &mi->result.oid);
1884 memcpy(ce->name, GITATTRIBUTES_FILE, len);
1885 add_index_entry(attr_index, ce,
1886 ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
1887 get_stream_filter(attr_index, GITATTRIBUTES_FILE, &ce->oid);
1888 } else {
1889 int stage, len;
1890 struct conflict_info *ci;
1892 ASSIGN_AND_VERIFY_CI(ci, mi);
1893 for (stage = 0; stage < 3; stage++) {
1894 unsigned stage_mask = (1 << stage);
1896 if (!(ci->filemask & stage_mask))
1897 continue;
1898 len = strlen(GITATTRIBUTES_FILE);
1899 ce = make_empty_cache_entry(attr_index, len);
1900 ce->ce_mode = create_ce_mode(ci->stages[stage].mode);
1901 ce->ce_flags = create_ce_flags(stage);
1902 ce->ce_namelen = len;
1903 oidcpy(&ce->oid, &ci->stages[stage].oid);
1904 memcpy(ce->name, GITATTRIBUTES_FILE, len);
1905 add_index_entry(attr_index, ce,
1906 ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
1907 get_stream_filter(attr_index, GITATTRIBUTES_FILE,
1908 &ce->oid);
1913 static int merge_3way(struct merge_options *opt,
1914 const char *path,
1915 const struct object_id *o,
1916 const struct object_id *a,
1917 const struct object_id *b,
1918 const char *pathnames[3],
1919 const int extra_marker_size,
1920 mmbuffer_t *result_buf)
1922 mmfile_t orig, src1, src2;
1923 struct ll_merge_options ll_opts = {0};
1924 char *base, *name1, *name2;
1925 enum ll_merge_result merge_status;
1927 if (!opt->priv->attr_index.initialized)
1928 initialize_attr_index(opt);
1930 ll_opts.renormalize = opt->renormalize;
1931 ll_opts.extra_marker_size = extra_marker_size;
1932 ll_opts.xdl_opts = opt->xdl_opts;
1934 if (opt->priv->call_depth) {
1935 ll_opts.virtual_ancestor = 1;
1936 ll_opts.variant = 0;
1937 } else {
1938 switch (opt->recursive_variant) {
1939 case MERGE_VARIANT_OURS:
1940 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1941 break;
1942 case MERGE_VARIANT_THEIRS:
1943 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1944 break;
1945 default:
1946 ll_opts.variant = 0;
1947 break;
1951 assert(pathnames[0] && pathnames[1] && pathnames[2] && opt->ancestor);
1952 if (pathnames[0] == pathnames[1] && pathnames[1] == pathnames[2]) {
1953 base = mkpathdup("%s", opt->ancestor);
1954 name1 = mkpathdup("%s", opt->branch1);
1955 name2 = mkpathdup("%s", opt->branch2);
1956 } else {
1957 base = mkpathdup("%s:%s", opt->ancestor, pathnames[0]);
1958 name1 = mkpathdup("%s:%s", opt->branch1, pathnames[1]);
1959 name2 = mkpathdup("%s:%s", opt->branch2, pathnames[2]);
1962 read_mmblob(&orig, o);
1963 read_mmblob(&src1, a);
1964 read_mmblob(&src2, b);
1966 merge_status = ll_merge(result_buf, path, &orig, base,
1967 &src1, name1, &src2, name2,
1968 &opt->priv->attr_index, &ll_opts);
1969 if (merge_status == LL_MERGE_BINARY_CONFLICT)
1970 path_msg(opt, CONFLICT_BINARY, 0,
1971 path, NULL, NULL, NULL,
1972 "warning: Cannot merge binary files: %s (%s vs. %s)",
1973 path, name1, name2);
1975 free(base);
1976 free(name1);
1977 free(name2);
1978 free(orig.ptr);
1979 free(src1.ptr);
1980 free(src2.ptr);
1981 return merge_status;
1984 static int handle_content_merge(struct merge_options *opt,
1985 const char *path,
1986 const struct version_info *o,
1987 const struct version_info *a,
1988 const struct version_info *b,
1989 const char *pathnames[3],
1990 const int extra_marker_size,
1991 struct version_info *result)
1994 * path is the target location where we want to put the file, and
1995 * is used to determine any normalization rules in ll_merge.
1997 * The normal case is that path and all entries in pathnames are
1998 * identical, though renames can affect which path we got one of
1999 * the three blobs to merge on various sides of history.
2001 * extra_marker_size is the amount to extend conflict markers in
2002 * ll_merge; this is neeed if we have content merges of content
2003 * merges, which happens for example with rename/rename(2to1) and
2004 * rename/add conflicts.
2006 unsigned clean = 1;
2009 * handle_content_merge() needs both files to be of the same type, i.e.
2010 * both files OR both submodules OR both symlinks. Conflicting types
2011 * needs to be handled elsewhere.
2013 assert((S_IFMT & a->mode) == (S_IFMT & b->mode));
2015 /* Merge modes */
2016 if (a->mode == b->mode || a->mode == o->mode)
2017 result->mode = b->mode;
2018 else {
2019 /* must be the 100644/100755 case */
2020 assert(S_ISREG(a->mode));
2021 result->mode = a->mode;
2022 clean = (b->mode == o->mode);
2024 * FIXME: If opt->priv->call_depth && !clean, then we really
2025 * should not make result->mode match either a->mode or
2026 * b->mode; that causes t6036 "check conflicting mode for
2027 * regular file" to fail. It would be best to use some other
2028 * mode, but we'll confuse all kinds of stuff if we use one
2029 * where S_ISREG(result->mode) isn't true, and if we use
2030 * something like 0100666, then tree-walk.c's calls to
2031 * canon_mode() will just normalize that to 100644 for us and
2032 * thus not solve anything.
2034 * Figure out if there's some kind of way we can work around
2035 * this...
2040 * Trivial oid merge.
2042 * Note: While one might assume that the next four lines would
2043 * be unnecessary due to the fact that match_mask is often
2044 * setup and already handled, renames don't always take care
2045 * of that.
2047 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
2048 oidcpy(&result->oid, &b->oid);
2049 else if (oideq(&b->oid, &o->oid))
2050 oidcpy(&result->oid, &a->oid);
2052 /* Remaining rules depend on file vs. submodule vs. symlink. */
2053 else if (S_ISREG(a->mode)) {
2054 mmbuffer_t result_buf;
2055 int ret = 0, merge_status;
2056 int two_way;
2059 * If 'o' is different type, treat it as null so we do a
2060 * two-way merge.
2062 two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode));
2064 merge_status = merge_3way(opt, path,
2065 two_way ? null_oid() : &o->oid,
2066 &a->oid, &b->oid,
2067 pathnames, extra_marker_size,
2068 &result_buf);
2070 if ((merge_status < 0) || !result_buf.ptr)
2071 ret = err(opt, _("Failed to execute internal merge"));
2073 if (!ret &&
2074 write_object_file(result_buf.ptr, result_buf.size,
2075 OBJ_BLOB, &result->oid))
2076 ret = err(opt, _("Unable to add %s to database"),
2077 path);
2079 free(result_buf.ptr);
2080 if (ret)
2081 return -1;
2082 clean &= (merge_status == 0);
2083 path_msg(opt, INFO_AUTO_MERGING, 1, path, NULL, NULL, NULL,
2084 _("Auto-merging %s"), path);
2085 } else if (S_ISGITLINK(a->mode)) {
2086 int two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode));
2087 clean = merge_submodule(opt, pathnames[0],
2088 two_way ? null_oid() : &o->oid,
2089 &a->oid, &b->oid, &result->oid);
2090 if (opt->priv->call_depth && two_way && !clean) {
2091 result->mode = o->mode;
2092 oidcpy(&result->oid, &o->oid);
2094 } else if (S_ISLNK(a->mode)) {
2095 if (opt->priv->call_depth) {
2096 clean = 0;
2097 result->mode = o->mode;
2098 oidcpy(&result->oid, &o->oid);
2099 } else {
2100 switch (opt->recursive_variant) {
2101 case MERGE_VARIANT_NORMAL:
2102 clean = 0;
2103 oidcpy(&result->oid, &a->oid);
2104 break;
2105 case MERGE_VARIANT_OURS:
2106 oidcpy(&result->oid, &a->oid);
2107 break;
2108 case MERGE_VARIANT_THEIRS:
2109 oidcpy(&result->oid, &b->oid);
2110 break;
2113 } else
2114 BUG("unsupported object type in the tree: %06o for %s",
2115 a->mode, path);
2117 return clean;
2120 /*** Function Grouping: functions related to detect_and_process_renames(), ***
2121 *** which are split into directory and regular rename detection sections. ***/
2123 /*** Function Grouping: functions related to directory rename detection ***/
2125 struct collision_info {
2126 struct string_list source_files;
2127 unsigned reported_already:1;
2131 * Return a new string that replaces the beginning portion (which matches
2132 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
2133 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
2134 * NOTE:
2135 * Caller must ensure that old_path starts with rename_info->key + '/'.
2137 static char *apply_dir_rename(struct strmap_entry *rename_info,
2138 const char *old_path)
2140 struct strbuf new_path = STRBUF_INIT;
2141 const char *old_dir = rename_info->key;
2142 const char *new_dir = rename_info->value;
2143 int oldlen, newlen, new_dir_len;
2145 oldlen = strlen(old_dir);
2146 if (*new_dir == '\0')
2148 * If someone renamed/merged a subdirectory into the root
2149 * directory (e.g. 'some/subdir' -> ''), then we want to
2150 * avoid returning
2151 * '' + '/filename'
2152 * as the rename; we need to make old_path + oldlen advance
2153 * past the '/' character.
2155 oldlen++;
2156 new_dir_len = strlen(new_dir);
2157 newlen = new_dir_len + (strlen(old_path) - oldlen) + 1;
2158 strbuf_grow(&new_path, newlen);
2159 strbuf_add(&new_path, new_dir, new_dir_len);
2160 strbuf_addstr(&new_path, &old_path[oldlen]);
2162 return strbuf_detach(&new_path, NULL);
2165 static int path_in_way(struct strmap *paths, const char *path, unsigned side_mask)
2167 struct merged_info *mi = strmap_get(paths, path);
2168 struct conflict_info *ci;
2169 if (!mi)
2170 return 0;
2171 INITIALIZE_CI(ci, mi);
2172 return mi->clean || (side_mask & (ci->filemask | ci->dirmask));
2176 * See if there is a directory rename for path, and if there are any file
2177 * level conflicts on the given side for the renamed location. If there is
2178 * a rename and there are no conflicts, return the new name. Otherwise,
2179 * return NULL.
2181 static char *handle_path_level_conflicts(struct merge_options *opt,
2182 const char *path,
2183 unsigned side_index,
2184 struct strmap_entry *rename_info,
2185 struct strmap *collisions)
2187 char *new_path = NULL;
2188 struct collision_info *c_info;
2189 int clean = 1;
2190 struct strbuf collision_paths = STRBUF_INIT;
2193 * entry has the mapping of old directory name to new directory name
2194 * that we want to apply to path.
2196 new_path = apply_dir_rename(rename_info, path);
2197 if (!new_path)
2198 BUG("Failed to apply directory rename!");
2201 * The caller needs to have ensured that it has pre-populated
2202 * collisions with all paths that map to new_path. Do a quick check
2203 * to ensure that's the case.
2205 c_info = strmap_get(collisions, new_path);
2206 if (!c_info)
2207 BUG("c_info is NULL");
2210 * Check for one-sided add/add/.../add conflicts, i.e.
2211 * where implicit renames from the other side doing
2212 * directory rename(s) can affect this side of history
2213 * to put multiple paths into the same location. Warn
2214 * and bail on directory renames for such paths.
2216 if (c_info->reported_already) {
2217 clean = 0;
2218 } else if (path_in_way(&opt->priv->paths, new_path, 1 << side_index)) {
2219 c_info->reported_already = 1;
2220 strbuf_add_separated_string_list(&collision_paths, ", ",
2221 &c_info->source_files);
2222 path_msg(opt, CONFLICT_DIR_RENAME_FILE_IN_WAY, 0,
2223 new_path, NULL, NULL, &c_info->source_files,
2224 _("CONFLICT (implicit dir rename): Existing "
2225 "file/dir at %s in the way of implicit "
2226 "directory rename(s) putting the following "
2227 "path(s) there: %s."),
2228 new_path, collision_paths.buf);
2229 clean = 0;
2230 } else if (c_info->source_files.nr > 1) {
2231 c_info->reported_already = 1;
2232 strbuf_add_separated_string_list(&collision_paths, ", ",
2233 &c_info->source_files);
2234 path_msg(opt, CONFLICT_DIR_RENAME_COLLISION, 0,
2235 new_path, NULL, NULL, &c_info->source_files,
2236 _("CONFLICT (implicit dir rename): Cannot map "
2237 "more than one path to %s; implicit directory "
2238 "renames tried to put these paths there: %s"),
2239 new_path, collision_paths.buf);
2240 clean = 0;
2243 /* Free memory we no longer need */
2244 strbuf_release(&collision_paths);
2245 if (!clean && new_path) {
2246 free(new_path);
2247 return NULL;
2250 return new_path;
2253 static void get_provisional_directory_renames(struct merge_options *opt,
2254 unsigned side,
2255 int *clean)
2257 struct hashmap_iter iter;
2258 struct strmap_entry *entry;
2259 struct rename_info *renames = &opt->priv->renames;
2262 * Collapse
2263 * dir_rename_count: old_directory -> {new_directory -> count}
2264 * down to
2265 * dir_renames: old_directory -> best_new_directory
2266 * where best_new_directory is the one with the unique highest count.
2268 strmap_for_each_entry(&renames->dir_rename_count[side], &iter, entry) {
2269 const char *source_dir = entry->key;
2270 struct strintmap *counts = entry->value;
2271 struct hashmap_iter count_iter;
2272 struct strmap_entry *count_entry;
2273 int max = 0;
2274 int bad_max = 0;
2275 const char *best = NULL;
2277 strintmap_for_each_entry(counts, &count_iter, count_entry) {
2278 const char *target_dir = count_entry->key;
2279 intptr_t count = (intptr_t)count_entry->value;
2281 if (count == max)
2282 bad_max = max;
2283 else if (count > max) {
2284 max = count;
2285 best = target_dir;
2289 if (max == 0)
2290 continue;
2292 if (bad_max == max) {
2293 path_msg(opt, CONFLICT_DIR_RENAME_SPLIT, 0,
2294 source_dir, NULL, NULL, NULL,
2295 _("CONFLICT (directory rename split): "
2296 "Unclear where to rename %s to; it was "
2297 "renamed to multiple other directories, "
2298 "with no destination getting a majority of "
2299 "the files."),
2300 source_dir);
2301 *clean = 0;
2302 } else {
2303 strmap_put(&renames->dir_renames[side],
2304 source_dir, (void*)best);
2309 static void handle_directory_level_conflicts(struct merge_options *opt)
2311 struct hashmap_iter iter;
2312 struct strmap_entry *entry;
2313 struct string_list duplicated = STRING_LIST_INIT_NODUP;
2314 struct rename_info *renames = &opt->priv->renames;
2315 struct strmap *side1_dir_renames = &renames->dir_renames[MERGE_SIDE1];
2316 struct strmap *side2_dir_renames = &renames->dir_renames[MERGE_SIDE2];
2317 int i;
2319 strmap_for_each_entry(side1_dir_renames, &iter, entry) {
2320 if (strmap_contains(side2_dir_renames, entry->key))
2321 string_list_append(&duplicated, entry->key);
2324 for (i = 0; i < duplicated.nr; i++) {
2325 strmap_remove(side1_dir_renames, duplicated.items[i].string, 0);
2326 strmap_remove(side2_dir_renames, duplicated.items[i].string, 0);
2328 string_list_clear(&duplicated, 0);
2331 static struct strmap_entry *check_dir_renamed(const char *path,
2332 struct strmap *dir_renames)
2334 char *temp = xstrdup(path);
2335 char *end;
2336 struct strmap_entry *e = NULL;
2338 while ((end = strrchr(temp, '/'))) {
2339 *end = '\0';
2340 e = strmap_get_entry(dir_renames, temp);
2341 if (e)
2342 break;
2344 free(temp);
2345 return e;
2348 static void compute_collisions(struct strmap *collisions,
2349 struct strmap *dir_renames,
2350 struct diff_queue_struct *pairs)
2352 int i;
2354 strmap_init_with_options(collisions, NULL, 0);
2355 if (strmap_empty(dir_renames))
2356 return;
2359 * Multiple files can be mapped to the same path due to directory
2360 * renames done by the other side of history. Since that other
2361 * side of history could have merged multiple directories into one,
2362 * if our side of history added the same file basename to each of
2363 * those directories, then all N of them would get implicitly
2364 * renamed by the directory rename detection into the same path,
2365 * and we'd get an add/add/.../add conflict, and all those adds
2366 * from *this* side of history. This is not representable in the
2367 * index, and users aren't going to easily be able to make sense of
2368 * it. So we need to provide a good warning about what's
2369 * happening, and fall back to no-directory-rename detection
2370 * behavior for those paths.
2372 * See testcases 9e and all of section 5 from t6043 for examples.
2374 for (i = 0; i < pairs->nr; ++i) {
2375 struct strmap_entry *rename_info;
2376 struct collision_info *collision_info;
2377 char *new_path;
2378 struct diff_filepair *pair = pairs->queue[i];
2380 if (pair->status != 'A' && pair->status != 'R')
2381 continue;
2382 rename_info = check_dir_renamed(pair->two->path, dir_renames);
2383 if (!rename_info)
2384 continue;
2386 new_path = apply_dir_rename(rename_info, pair->two->path);
2387 assert(new_path);
2388 collision_info = strmap_get(collisions, new_path);
2389 if (collision_info) {
2390 free(new_path);
2391 } else {
2392 CALLOC_ARRAY(collision_info, 1);
2393 string_list_init_nodup(&collision_info->source_files);
2394 strmap_put(collisions, new_path, collision_info);
2396 string_list_insert(&collision_info->source_files,
2397 pair->two->path);
2401 static void free_collisions(struct strmap *collisions)
2403 struct hashmap_iter iter;
2404 struct strmap_entry *entry;
2406 /* Free each value in the collisions map */
2407 strmap_for_each_entry(collisions, &iter, entry) {
2408 struct collision_info *info = entry->value;
2409 string_list_clear(&info->source_files, 0);
2412 * In compute_collisions(), we set collisions.strdup_strings to 0
2413 * so that we wouldn't have to make another copy of the new_path
2414 * allocated by apply_dir_rename(). But now that we've used them
2415 * and have no other references to these strings, it is time to
2416 * deallocate them.
2418 free_strmap_strings(collisions);
2419 strmap_clear(collisions, 1);
2422 static char *check_for_directory_rename(struct merge_options *opt,
2423 const char *path,
2424 unsigned side_index,
2425 struct strmap *dir_renames,
2426 struct strmap *dir_rename_exclusions,
2427 struct strmap *collisions,
2428 int *clean_merge)
2430 char *new_path;
2431 struct strmap_entry *rename_info;
2432 struct strmap_entry *otherinfo;
2433 const char *new_dir;
2434 int other_side = 3 - side_index;
2437 * Cases where we don't have or don't want a directory rename for
2438 * this path.
2440 if (strmap_empty(dir_renames))
2441 return NULL;
2442 if (strmap_get(&collisions[other_side], path))
2443 return NULL;
2444 rename_info = check_dir_renamed(path, dir_renames);
2445 if (!rename_info)
2446 return NULL;
2449 * This next part is a little weird. We do not want to do an
2450 * implicit rename into a directory we renamed on our side, because
2451 * that will result in a spurious rename/rename(1to2) conflict. An
2452 * example:
2453 * Base commit: dumbdir/afile, otherdir/bfile
2454 * Side 1: smrtdir/afile, otherdir/bfile
2455 * Side 2: dumbdir/afile, dumbdir/bfile
2456 * Here, while working on Side 1, we could notice that otherdir was
2457 * renamed/merged to dumbdir, and change the diff_filepair for
2458 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2459 * 2 will notice the rename from dumbdir to smrtdir, and do the
2460 * transitive rename to move it from dumbdir/bfile to
2461 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2462 * smrtdir, a rename/rename(1to2) conflict. We really just want
2463 * the file to end up in smrtdir. And the way to achieve that is
2464 * to not let Side1 do the rename to dumbdir, since we know that is
2465 * the source of one of our directory renames.
2467 * That's why otherinfo and dir_rename_exclusions is here.
2469 * As it turns out, this also prevents N-way transient rename
2470 * confusion; See testcases 9c and 9d of t6043.
2472 new_dir = rename_info->value; /* old_dir = rename_info->key; */
2473 otherinfo = strmap_get_entry(dir_rename_exclusions, new_dir);
2474 if (otherinfo) {
2475 path_msg(opt, INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME, 1,
2476 rename_info->key, path, new_dir, NULL,
2477 _("WARNING: Avoiding applying %s -> %s rename "
2478 "to %s, because %s itself was renamed."),
2479 rename_info->key, new_dir, path, new_dir);
2480 return NULL;
2483 new_path = handle_path_level_conflicts(opt, path, side_index,
2484 rename_info,
2485 &collisions[side_index]);
2486 *clean_merge &= (new_path != NULL);
2488 return new_path;
2491 static void apply_directory_rename_modifications(struct merge_options *opt,
2492 struct diff_filepair *pair,
2493 char *new_path)
2496 * The basic idea is to get the conflict_info from opt->priv->paths
2497 * at old path, and insert it into new_path; basically just this:
2498 * ci = strmap_get(&opt->priv->paths, old_path);
2499 * strmap_remove(&opt->priv->paths, old_path, 0);
2500 * strmap_put(&opt->priv->paths, new_path, ci);
2501 * However, there are some factors complicating this:
2502 * - opt->priv->paths may already have an entry at new_path
2503 * - Each ci tracks its containing directory, so we need to
2504 * update that
2505 * - If another ci has the same containing directory, then
2506 * the two char*'s MUST point to the same location. See the
2507 * comment in struct merged_info. strcmp equality is not
2508 * enough; we need pointer equality.
2509 * - opt->priv->paths must hold the parent directories of any
2510 * entries that are added. So, if this directory rename
2511 * causes entirely new directories, we must recursively add
2512 * parent directories.
2513 * - For each parent directory added to opt->priv->paths, we
2514 * also need to get its parent directory stored in its
2515 * conflict_info->merged.directory_name with all the same
2516 * requirements about pointer equality.
2518 struct string_list dirs_to_insert = STRING_LIST_INIT_NODUP;
2519 struct conflict_info *ci, *new_ci;
2520 struct strmap_entry *entry;
2521 const char *branch_with_new_path, *branch_with_dir_rename;
2522 const char *old_path = pair->two->path;
2523 const char *parent_name;
2524 const char *cur_path;
2525 int i, len;
2527 entry = strmap_get_entry(&opt->priv->paths, old_path);
2528 old_path = entry->key;
2529 ci = entry->value;
2530 VERIFY_CI(ci);
2532 /* Find parent directories missing from opt->priv->paths */
2533 cur_path = mem_pool_strdup(&opt->priv->pool, new_path);
2534 free((char*)new_path);
2535 new_path = (char *)cur_path;
2537 while (1) {
2538 /* Find the parent directory of cur_path */
2539 char *last_slash = strrchr(cur_path, '/');
2540 if (last_slash) {
2541 parent_name = mem_pool_strndup(&opt->priv->pool,
2542 cur_path,
2543 last_slash - cur_path);
2544 } else {
2545 parent_name = opt->priv->toplevel_dir;
2546 break;
2549 /* Look it up in opt->priv->paths */
2550 entry = strmap_get_entry(&opt->priv->paths, parent_name);
2551 if (entry) {
2552 parent_name = entry->key; /* reuse known pointer */
2553 break;
2556 /* Record this is one of the directories we need to insert */
2557 string_list_append(&dirs_to_insert, parent_name);
2558 cur_path = parent_name;
2561 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2562 for (i = dirs_to_insert.nr-1; i >= 0; --i) {
2563 struct conflict_info *dir_ci;
2564 char *cur_dir = dirs_to_insert.items[i].string;
2566 CALLOC_ARRAY(dir_ci, 1);
2568 dir_ci->merged.directory_name = parent_name;
2569 len = strlen(parent_name);
2570 /* len+1 because of trailing '/' character */
2571 dir_ci->merged.basename_offset = (len > 0 ? len+1 : len);
2572 dir_ci->dirmask = ci->filemask;
2573 strmap_put(&opt->priv->paths, cur_dir, dir_ci);
2575 parent_name = cur_dir;
2578 assert(ci->filemask == 2 || ci->filemask == 4);
2579 assert(ci->dirmask == 0);
2580 strmap_remove(&opt->priv->paths, old_path, 0);
2582 branch_with_new_path = (ci->filemask == 2) ? opt->branch1 : opt->branch2;
2583 branch_with_dir_rename = (ci->filemask == 2) ? opt->branch2 : opt->branch1;
2585 /* Now, finally update ci and stick it into opt->priv->paths */
2586 ci->merged.directory_name = parent_name;
2587 len = strlen(parent_name);
2588 ci->merged.basename_offset = (len > 0 ? len+1 : len);
2589 new_ci = strmap_get(&opt->priv->paths, new_path);
2590 if (!new_ci) {
2591 /* Place ci back into opt->priv->paths, but at new_path */
2592 strmap_put(&opt->priv->paths, new_path, ci);
2593 } else {
2594 int index;
2596 /* A few sanity checks */
2597 VERIFY_CI(new_ci);
2598 assert(ci->filemask == 2 || ci->filemask == 4);
2599 assert((new_ci->filemask & ci->filemask) == 0);
2600 assert(!new_ci->merged.clean);
2602 /* Copy stuff from ci into new_ci */
2603 new_ci->filemask |= ci->filemask;
2604 if (new_ci->dirmask)
2605 new_ci->df_conflict = 1;
2606 index = (ci->filemask >> 1);
2607 new_ci->pathnames[index] = ci->pathnames[index];
2608 new_ci->stages[index].mode = ci->stages[index].mode;
2609 oidcpy(&new_ci->stages[index].oid, &ci->stages[index].oid);
2611 ci = new_ci;
2614 if (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) {
2615 /* Notify user of updated path */
2616 if (pair->status == 'A')
2617 path_msg(opt, INFO_DIR_RENAME_APPLIED, 1,
2618 new_path, old_path, NULL, NULL,
2619 _("Path updated: %s added in %s inside a "
2620 "directory that was renamed in %s; moving "
2621 "it to %s."),
2622 old_path, branch_with_new_path,
2623 branch_with_dir_rename, new_path);
2624 else
2625 path_msg(opt, INFO_DIR_RENAME_APPLIED, 1,
2626 new_path, old_path, NULL, NULL,
2627 _("Path updated: %s renamed to %s in %s, "
2628 "inside a directory that was renamed in %s; "
2629 "moving it to %s."),
2630 pair->one->path, old_path, branch_with_new_path,
2631 branch_with_dir_rename, new_path);
2632 } else {
2634 * opt->detect_directory_renames has the value
2635 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2637 ci->path_conflict = 1;
2638 if (pair->status == 'A')
2639 path_msg(opt, CONFLICT_DIR_RENAME_SUGGESTED, 1,
2640 new_path, old_path, NULL, NULL,
2641 _("CONFLICT (file location): %s added in %s "
2642 "inside a directory that was renamed in %s, "
2643 "suggesting it should perhaps be moved to "
2644 "%s."),
2645 old_path, branch_with_new_path,
2646 branch_with_dir_rename, new_path);
2647 else
2648 path_msg(opt, CONFLICT_DIR_RENAME_SUGGESTED, 1,
2649 new_path, old_path, NULL, NULL,
2650 _("CONFLICT (file location): %s renamed to %s "
2651 "in %s, inside a directory that was renamed "
2652 "in %s, suggesting it should perhaps be "
2653 "moved to %s."),
2654 pair->one->path, old_path, branch_with_new_path,
2655 branch_with_dir_rename, new_path);
2659 * Finally, record the new location.
2661 pair->two->path = new_path;
2664 /*** Function Grouping: functions related to regular rename detection ***/
2666 static int process_renames(struct merge_options *opt,
2667 struct diff_queue_struct *renames)
2669 int clean_merge = 1, i;
2671 for (i = 0; i < renames->nr; ++i) {
2672 const char *oldpath = NULL, *newpath;
2673 struct diff_filepair *pair = renames->queue[i];
2674 struct conflict_info *oldinfo = NULL, *newinfo = NULL;
2675 struct strmap_entry *old_ent, *new_ent;
2676 unsigned int old_sidemask;
2677 int target_index, other_source_index;
2678 int source_deleted, collision, type_changed;
2679 const char *rename_branch = NULL, *delete_branch = NULL;
2681 old_ent = strmap_get_entry(&opt->priv->paths, pair->one->path);
2682 new_ent = strmap_get_entry(&opt->priv->paths, pair->two->path);
2683 if (old_ent) {
2684 oldpath = old_ent->key;
2685 oldinfo = old_ent->value;
2687 newpath = pair->two->path;
2688 if (new_ent) {
2689 newpath = new_ent->key;
2690 newinfo = new_ent->value;
2694 * If pair->one->path isn't in opt->priv->paths, that means
2695 * that either directory rename detection removed that
2696 * path, or a parent directory of oldpath was resolved and
2697 * we don't even need the rename; in either case, we can
2698 * skip it. If oldinfo->merged.clean, then the other side
2699 * of history had no changes to oldpath and we don't need
2700 * the rename and can skip it.
2702 if (!oldinfo || oldinfo->merged.clean)
2703 continue;
2706 * diff_filepairs have copies of pathnames, thus we have to
2707 * use standard 'strcmp()' (negated) instead of '=='.
2709 if (i + 1 < renames->nr &&
2710 !strcmp(oldpath, renames->queue[i+1]->one->path)) {
2711 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2712 const char *pathnames[3];
2713 struct version_info merged;
2714 struct conflict_info *base, *side1, *side2;
2715 unsigned was_binary_blob = 0;
2717 pathnames[0] = oldpath;
2718 pathnames[1] = newpath;
2719 pathnames[2] = renames->queue[i+1]->two->path;
2721 base = strmap_get(&opt->priv->paths, pathnames[0]);
2722 side1 = strmap_get(&opt->priv->paths, pathnames[1]);
2723 side2 = strmap_get(&opt->priv->paths, pathnames[2]);
2725 VERIFY_CI(base);
2726 VERIFY_CI(side1);
2727 VERIFY_CI(side2);
2729 if (!strcmp(pathnames[1], pathnames[2])) {
2730 struct rename_info *ri = &opt->priv->renames;
2731 int j;
2733 /* Both sides renamed the same way */
2734 assert(side1 == side2);
2735 memcpy(&side1->stages[0], &base->stages[0],
2736 sizeof(merged));
2737 side1->filemask |= (1 << MERGE_BASE);
2738 /* Mark base as resolved by removal */
2739 base->merged.is_null = 1;
2740 base->merged.clean = 1;
2743 * Disable remembering renames optimization;
2744 * rename/rename(1to1) is incredibly rare, and
2745 * just disabling the optimization is easier
2746 * than purging cached_pairs,
2747 * cached_target_names, and dir_rename_counts.
2749 for (j = 0; j < 3; j++)
2750 ri->merge_trees[j] = NULL;
2752 /* We handled both renames, i.e. i+1 handled */
2753 i++;
2754 /* Move to next rename */
2755 continue;
2758 /* This is a rename/rename(1to2) */
2759 clean_merge = handle_content_merge(opt,
2760 pair->one->path,
2761 &base->stages[0],
2762 &side1->stages[1],
2763 &side2->stages[2],
2764 pathnames,
2765 1 + 2 * opt->priv->call_depth,
2766 &merged);
2767 if (!clean_merge &&
2768 merged.mode == side1->stages[1].mode &&
2769 oideq(&merged.oid, &side1->stages[1].oid))
2770 was_binary_blob = 1;
2771 memcpy(&side1->stages[1], &merged, sizeof(merged));
2772 if (was_binary_blob) {
2774 * Getting here means we were attempting to
2775 * merge a binary blob.
2777 * Since we can't merge binaries,
2778 * handle_content_merge() just takes one
2779 * side. But we don't want to copy the
2780 * contents of one side to both paths. We
2781 * used the contents of side1 above for
2782 * side1->stages, let's use the contents of
2783 * side2 for side2->stages below.
2785 oidcpy(&merged.oid, &side2->stages[2].oid);
2786 merged.mode = side2->stages[2].mode;
2788 memcpy(&side2->stages[2], &merged, sizeof(merged));
2790 side1->path_conflict = 1;
2791 side2->path_conflict = 1;
2793 * TODO: For renames we normally remove the path at the
2794 * old name. It would thus seem consistent to do the
2795 * same for rename/rename(1to2) cases, but we haven't
2796 * done so traditionally and a number of the regression
2797 * tests now encode an expectation that the file is
2798 * left there at stage 1. If we ever decide to change
2799 * this, add the following two lines here:
2800 * base->merged.is_null = 1;
2801 * base->merged.clean = 1;
2802 * and remove the setting of base->path_conflict to 1.
2804 base->path_conflict = 1;
2805 path_msg(opt, CONFLICT_RENAME_RENAME, 0,
2806 pathnames[0], pathnames[1], pathnames[2], NULL,
2807 _("CONFLICT (rename/rename): %s renamed to "
2808 "%s in %s and to %s in %s."),
2809 pathnames[0],
2810 pathnames[1], opt->branch1,
2811 pathnames[2], opt->branch2);
2813 i++; /* We handled both renames, i.e. i+1 handled */
2814 continue;
2817 VERIFY_CI(oldinfo);
2818 VERIFY_CI(newinfo);
2819 target_index = pair->score; /* from collect_renames() */
2820 assert(target_index == 1 || target_index == 2);
2821 other_source_index = 3 - target_index;
2822 old_sidemask = (1 << other_source_index); /* 2 or 4 */
2823 source_deleted = (oldinfo->filemask == 1);
2824 collision = ((newinfo->filemask & old_sidemask) != 0);
2825 type_changed = !source_deleted &&
2826 (S_ISREG(oldinfo->stages[other_source_index].mode) !=
2827 S_ISREG(newinfo->stages[target_index].mode));
2828 if (type_changed && collision) {
2830 * special handling so later blocks can handle this...
2832 * if type_changed && collision are both true, then this
2833 * was really a double rename, but one side wasn't
2834 * detected due to lack of break detection. I.e.
2835 * something like
2836 * orig: has normal file 'foo'
2837 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2838 * side2: renames 'foo' to 'bar'
2839 * In this case, the foo->bar rename on side1 won't be
2840 * detected because the new symlink named 'foo' is
2841 * there and we don't do break detection. But we detect
2842 * this here because we don't want to merge the content
2843 * of the foo symlink with the foo->bar file, so we
2844 * have some logic to handle this special case. The
2845 * easiest way to do that is make 'bar' on side1 not
2846 * be considered a colliding file but the other part
2847 * of a normal rename. If the file is very different,
2848 * well we're going to get content merge conflicts
2849 * anyway so it doesn't hurt. And if the colliding
2850 * file also has a different type, that'll be handled
2851 * by the content merge logic in process_entry() too.
2853 * See also t6430, 'rename vs. rename/symlink'
2855 collision = 0;
2857 if (source_deleted) {
2858 if (target_index == 1) {
2859 rename_branch = opt->branch1;
2860 delete_branch = opt->branch2;
2861 } else {
2862 rename_branch = opt->branch2;
2863 delete_branch = opt->branch1;
2867 assert(source_deleted || oldinfo->filemask & old_sidemask);
2869 /* Need to check for special types of rename conflicts... */
2870 if (collision && !source_deleted) {
2871 /* collision: rename/add or rename/rename(2to1) */
2872 const char *pathnames[3];
2873 struct version_info merged;
2875 struct conflict_info *base, *side1, *side2;
2876 unsigned clean;
2878 pathnames[0] = oldpath;
2879 pathnames[other_source_index] = oldpath;
2880 pathnames[target_index] = newpath;
2882 base = strmap_get(&opt->priv->paths, pathnames[0]);
2883 side1 = strmap_get(&opt->priv->paths, pathnames[1]);
2884 side2 = strmap_get(&opt->priv->paths, pathnames[2]);
2886 VERIFY_CI(base);
2887 VERIFY_CI(side1);
2888 VERIFY_CI(side2);
2890 clean = handle_content_merge(opt, pair->one->path,
2891 &base->stages[0],
2892 &side1->stages[1],
2893 &side2->stages[2],
2894 pathnames,
2895 1 + 2 * opt->priv->call_depth,
2896 &merged);
2898 memcpy(&newinfo->stages[target_index], &merged,
2899 sizeof(merged));
2900 if (!clean) {
2901 path_msg(opt, CONFLICT_RENAME_COLLIDES, 0,
2902 newpath, oldpath, NULL, NULL,
2903 _("CONFLICT (rename involved in "
2904 "collision): rename of %s -> %s has "
2905 "content conflicts AND collides "
2906 "with another path; this may result "
2907 "in nested conflict markers."),
2908 oldpath, newpath);
2910 } else if (collision && source_deleted) {
2912 * rename/add/delete or rename/rename(2to1)/delete:
2913 * since oldpath was deleted on the side that didn't
2914 * do the rename, there's not much of a content merge
2915 * we can do for the rename. oldinfo->merged.is_null
2916 * was already set, so we just leave things as-is so
2917 * they look like an add/add conflict.
2920 newinfo->path_conflict = 1;
2921 path_msg(opt, CONFLICT_RENAME_DELETE, 0,
2922 newpath, oldpath, NULL, NULL,
2923 _("CONFLICT (rename/delete): %s renamed "
2924 "to %s in %s, but deleted in %s."),
2925 oldpath, newpath, rename_branch, delete_branch);
2926 } else {
2928 * a few different cases...start by copying the
2929 * existing stage(s) from oldinfo over the newinfo
2930 * and update the pathname(s).
2932 memcpy(&newinfo->stages[0], &oldinfo->stages[0],
2933 sizeof(newinfo->stages[0]));
2934 newinfo->filemask |= (1 << MERGE_BASE);
2935 newinfo->pathnames[0] = oldpath;
2936 if (type_changed) {
2937 /* rename vs. typechange */
2938 /* Mark the original as resolved by removal */
2939 memcpy(&oldinfo->stages[0].oid, null_oid(),
2940 sizeof(oldinfo->stages[0].oid));
2941 oldinfo->stages[0].mode = 0;
2942 oldinfo->filemask &= 0x06;
2943 } else if (source_deleted) {
2944 /* rename/delete */
2945 newinfo->path_conflict = 1;
2946 path_msg(opt, CONFLICT_RENAME_DELETE, 0,
2947 newpath, oldpath, NULL, NULL,
2948 _("CONFLICT (rename/delete): %s renamed"
2949 " to %s in %s, but deleted in %s."),
2950 oldpath, newpath,
2951 rename_branch, delete_branch);
2952 } else {
2953 /* normal rename */
2954 memcpy(&newinfo->stages[other_source_index],
2955 &oldinfo->stages[other_source_index],
2956 sizeof(newinfo->stages[0]));
2957 newinfo->filemask |= (1 << other_source_index);
2958 newinfo->pathnames[other_source_index] = oldpath;
2962 if (!type_changed) {
2963 /* Mark the original as resolved by removal */
2964 oldinfo->merged.is_null = 1;
2965 oldinfo->merged.clean = 1;
2970 return clean_merge;
2973 static inline int possible_side_renames(struct rename_info *renames,
2974 unsigned side_index)
2976 return renames->pairs[side_index].nr > 0 &&
2977 !strintmap_empty(&renames->relevant_sources[side_index]);
2980 static inline int possible_renames(struct rename_info *renames)
2982 return possible_side_renames(renames, 1) ||
2983 possible_side_renames(renames, 2) ||
2984 !strmap_empty(&renames->cached_pairs[1]) ||
2985 !strmap_empty(&renames->cached_pairs[2]);
2988 static void resolve_diffpair_statuses(struct diff_queue_struct *q)
2991 * A simplified version of diff_resolve_rename_copy(); would probably
2992 * just use that function but it's static...
2994 int i;
2995 struct diff_filepair *p;
2997 for (i = 0; i < q->nr; ++i) {
2998 p = q->queue[i];
2999 p->status = 0; /* undecided */
3000 if (!DIFF_FILE_VALID(p->one))
3001 p->status = DIFF_STATUS_ADDED;
3002 else if (!DIFF_FILE_VALID(p->two))
3003 p->status = DIFF_STATUS_DELETED;
3004 else if (DIFF_PAIR_RENAME(p))
3005 p->status = DIFF_STATUS_RENAMED;
3009 static void prune_cached_from_relevant(struct rename_info *renames,
3010 unsigned side)
3012 /* Reason for this function described in add_pair() */
3013 struct hashmap_iter iter;
3014 struct strmap_entry *entry;
3016 /* Remove from relevant_sources all entries in cached_pairs[side] */
3017 strmap_for_each_entry(&renames->cached_pairs[side], &iter, entry) {
3018 strintmap_remove(&renames->relevant_sources[side],
3019 entry->key);
3021 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
3022 strset_for_each_entry(&renames->cached_irrelevant[side], &iter, entry) {
3023 strintmap_remove(&renames->relevant_sources[side],
3024 entry->key);
3028 static void use_cached_pairs(struct merge_options *opt,
3029 struct strmap *cached_pairs,
3030 struct diff_queue_struct *pairs)
3032 struct hashmap_iter iter;
3033 struct strmap_entry *entry;
3036 * Add to side_pairs all entries from renames->cached_pairs[side_index].
3037 * (Info in cached_irrelevant[side_index] is not relevant here.)
3039 strmap_for_each_entry(cached_pairs, &iter, entry) {
3040 struct diff_filespec *one, *two;
3041 const char *old_name = entry->key;
3042 const char *new_name = entry->value;
3043 if (!new_name)
3044 new_name = old_name;
3047 * cached_pairs has *copies* of old_name and new_name,
3048 * because it has to persist across merges. Since
3049 * pool_alloc_filespec() will just re-use the existing
3050 * filenames, which will also get re-used by
3051 * opt->priv->paths if they become renames, and then
3052 * get freed at the end of the merge, that would leave
3053 * the copy in cached_pairs dangling. Avoid this by
3054 * making a copy here.
3056 old_name = mem_pool_strdup(&opt->priv->pool, old_name);
3057 new_name = mem_pool_strdup(&opt->priv->pool, new_name);
3059 /* We don't care about oid/mode, only filenames and status */
3060 one = pool_alloc_filespec(&opt->priv->pool, old_name);
3061 two = pool_alloc_filespec(&opt->priv->pool, new_name);
3062 pool_diff_queue(&opt->priv->pool, pairs, one, two);
3063 pairs->queue[pairs->nr-1]->status = entry->value ? 'R' : 'D';
3067 static void cache_new_pair(struct rename_info *renames,
3068 int side,
3069 char *old_path,
3070 char *new_path,
3071 int free_old_value)
3073 char *old_value;
3074 new_path = xstrdup(new_path);
3075 old_value = strmap_put(&renames->cached_pairs[side],
3076 old_path, new_path);
3077 strset_add(&renames->cached_target_names[side], new_path);
3078 if (free_old_value)
3079 free(old_value);
3080 else
3081 assert(!old_value);
3084 static void possibly_cache_new_pair(struct rename_info *renames,
3085 struct diff_filepair *p,
3086 unsigned side,
3087 char *new_path)
3089 int dir_renamed_side = 0;
3091 if (new_path) {
3093 * Directory renames happen on the other side of history from
3094 * the side that adds new files to the old directory.
3096 dir_renamed_side = 3 - side;
3097 } else {
3098 int val = strintmap_get(&renames->relevant_sources[side],
3099 p->one->path);
3100 if (val == RELEVANT_NO_MORE) {
3101 assert(p->status == 'D');
3102 strset_add(&renames->cached_irrelevant[side],
3103 p->one->path);
3105 if (val <= 0)
3106 return;
3109 if (p->status == 'D') {
3111 * If we already had this delete, we'll just set it's value
3112 * to NULL again, so no harm.
3114 strmap_put(&renames->cached_pairs[side], p->one->path, NULL);
3115 } else if (p->status == 'R') {
3116 if (!new_path)
3117 new_path = p->two->path;
3118 else
3119 cache_new_pair(renames, dir_renamed_side,
3120 p->two->path, new_path, 0);
3121 cache_new_pair(renames, side, p->one->path, new_path, 1);
3122 } else if (p->status == 'A' && new_path) {
3123 cache_new_pair(renames, dir_renamed_side,
3124 p->two->path, new_path, 0);
3128 static int compare_pairs(const void *a_, const void *b_)
3130 const struct diff_filepair *a = *((const struct diff_filepair **)a_);
3131 const struct diff_filepair *b = *((const struct diff_filepair **)b_);
3133 return strcmp(a->one->path, b->one->path);
3136 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
3137 static int detect_regular_renames(struct merge_options *opt,
3138 unsigned side_index)
3140 struct diff_options diff_opts;
3141 struct rename_info *renames = &opt->priv->renames;
3143 prune_cached_from_relevant(renames, side_index);
3144 if (!possible_side_renames(renames, side_index)) {
3146 * No rename detection needed for this side, but we still need
3147 * to make sure 'adds' are marked correctly in case the other
3148 * side had directory renames.
3150 resolve_diffpair_statuses(&renames->pairs[side_index]);
3151 return 0;
3154 partial_clear_dir_rename_count(&renames->dir_rename_count[side_index]);
3155 repo_diff_setup(opt->repo, &diff_opts);
3156 diff_opts.flags.recursive = 1;
3157 diff_opts.flags.rename_empty = 0;
3158 diff_opts.detect_rename = DIFF_DETECT_RENAME;
3159 diff_opts.rename_limit = opt->rename_limit;
3160 if (opt->rename_limit <= 0)
3161 diff_opts.rename_limit = 7000;
3162 diff_opts.rename_score = opt->rename_score;
3163 diff_opts.show_rename_progress = opt->show_rename_progress;
3164 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
3165 diff_setup_done(&diff_opts);
3167 diff_queued_diff = renames->pairs[side_index];
3168 trace2_region_enter("diff", "diffcore_rename", opt->repo);
3169 diffcore_rename_extended(&diff_opts,
3170 &opt->priv->pool,
3171 &renames->relevant_sources[side_index],
3172 &renames->dirs_removed[side_index],
3173 &renames->dir_rename_count[side_index],
3174 &renames->cached_pairs[side_index]);
3175 trace2_region_leave("diff", "diffcore_rename", opt->repo);
3176 resolve_diffpair_statuses(&diff_queued_diff);
3178 if (diff_opts.needed_rename_limit > 0)
3179 renames->redo_after_renames = 0;
3180 if (diff_opts.needed_rename_limit > renames->needed_limit)
3181 renames->needed_limit = diff_opts.needed_rename_limit;
3183 renames->pairs[side_index] = diff_queued_diff;
3185 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
3186 diff_queued_diff.nr = 0;
3187 diff_queued_diff.queue = NULL;
3188 diff_flush(&diff_opts);
3190 return 1;
3194 * Get information of all renames which occurred in 'side_pairs', making use
3195 * of any implicit directory renames in side_dir_renames (also making use of
3196 * implicit directory renames rename_exclusions as needed by
3197 * check_for_directory_rename()). Add all (updated) renames into result.
3199 static int collect_renames(struct merge_options *opt,
3200 struct diff_queue_struct *result,
3201 unsigned side_index,
3202 struct strmap *collisions,
3203 struct strmap *dir_renames_for_side,
3204 struct strmap *rename_exclusions)
3206 int i, clean = 1;
3207 struct diff_queue_struct *side_pairs;
3208 struct rename_info *renames = &opt->priv->renames;
3210 side_pairs = &renames->pairs[side_index];
3212 for (i = 0; i < side_pairs->nr; ++i) {
3213 struct diff_filepair *p = side_pairs->queue[i];
3214 char *new_path; /* non-NULL only with directory renames */
3216 if (p->status != 'A' && p->status != 'R') {
3217 possibly_cache_new_pair(renames, p, side_index, NULL);
3218 pool_diff_free_filepair(&opt->priv->pool, p);
3219 continue;
3222 new_path = check_for_directory_rename(opt, p->two->path,
3223 side_index,
3224 dir_renames_for_side,
3225 rename_exclusions,
3226 collisions,
3227 &clean);
3229 possibly_cache_new_pair(renames, p, side_index, new_path);
3230 if (p->status != 'R' && !new_path) {
3231 pool_diff_free_filepair(&opt->priv->pool, p);
3232 continue;
3235 if (new_path)
3236 apply_directory_rename_modifications(opt, p, new_path);
3239 * p->score comes back from diffcore_rename_extended() with
3240 * the similarity of the renamed file. The similarity is
3241 * was used to determine that the two files were related
3242 * and are a rename, which we have already used, but beyond
3243 * that we have no use for the similarity. So p->score is
3244 * now irrelevant. However, process_renames() will need to
3245 * know which side of the merge this rename was associated
3246 * with, so overwrite p->score with that value.
3248 p->score = side_index;
3249 result->queue[result->nr++] = p;
3252 return clean;
3255 static int detect_and_process_renames(struct merge_options *opt,
3256 struct tree *merge_base,
3257 struct tree *side1,
3258 struct tree *side2)
3260 struct diff_queue_struct combined = { 0 };
3261 struct rename_info *renames = &opt->priv->renames;
3262 struct strmap collisions[3];
3263 int need_dir_renames, s, i, clean = 1;
3264 unsigned detection_run = 0;
3266 if (!possible_renames(renames))
3267 goto cleanup;
3269 trace2_region_enter("merge", "regular renames", opt->repo);
3270 detection_run |= detect_regular_renames(opt, MERGE_SIDE1);
3271 detection_run |= detect_regular_renames(opt, MERGE_SIDE2);
3272 if (renames->needed_limit) {
3273 renames->cached_pairs_valid_side = 0;
3274 renames->redo_after_renames = 0;
3276 if (renames->redo_after_renames && detection_run) {
3277 int i, side;
3278 struct diff_filepair *p;
3280 /* Cache the renames, we found */
3281 for (side = MERGE_SIDE1; side <= MERGE_SIDE2; side++) {
3282 for (i = 0; i < renames->pairs[side].nr; ++i) {
3283 p = renames->pairs[side].queue[i];
3284 possibly_cache_new_pair(renames, p, side, NULL);
3288 /* Restart the merge with the cached renames */
3289 renames->redo_after_renames = 2;
3290 trace2_region_leave("merge", "regular renames", opt->repo);
3291 goto cleanup;
3293 use_cached_pairs(opt, &renames->cached_pairs[1], &renames->pairs[1]);
3294 use_cached_pairs(opt, &renames->cached_pairs[2], &renames->pairs[2]);
3295 trace2_region_leave("merge", "regular renames", opt->repo);
3297 trace2_region_enter("merge", "directory renames", opt->repo);
3298 need_dir_renames =
3299 !opt->priv->call_depth &&
3300 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE ||
3301 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT);
3303 if (need_dir_renames) {
3304 get_provisional_directory_renames(opt, MERGE_SIDE1, &clean);
3305 get_provisional_directory_renames(opt, MERGE_SIDE2, &clean);
3306 handle_directory_level_conflicts(opt);
3309 ALLOC_GROW(combined.queue,
3310 renames->pairs[1].nr + renames->pairs[2].nr,
3311 combined.alloc);
3312 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) {
3313 int other_side = 3 - i;
3314 compute_collisions(&collisions[i],
3315 &renames->dir_renames[other_side],
3316 &renames->pairs[i]);
3318 clean &= collect_renames(opt, &combined, MERGE_SIDE1,
3319 collisions,
3320 &renames->dir_renames[2],
3321 &renames->dir_renames[1]);
3322 clean &= collect_renames(opt, &combined, MERGE_SIDE2,
3323 collisions,
3324 &renames->dir_renames[1],
3325 &renames->dir_renames[2]);
3326 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++)
3327 free_collisions(&collisions[i]);
3328 STABLE_QSORT(combined.queue, combined.nr, compare_pairs);
3329 trace2_region_leave("merge", "directory renames", opt->repo);
3331 trace2_region_enter("merge", "process renames", opt->repo);
3332 clean &= process_renames(opt, &combined);
3333 trace2_region_leave("merge", "process renames", opt->repo);
3335 goto simple_cleanup; /* collect_renames() handles some of cleanup */
3337 cleanup:
3339 * Free now unneeded filepairs, which would have been handled
3340 * in collect_renames() normally but we skipped that code.
3342 for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) {
3343 struct diff_queue_struct *side_pairs;
3344 int i;
3346 side_pairs = &renames->pairs[s];
3347 for (i = 0; i < side_pairs->nr; ++i) {
3348 struct diff_filepair *p = side_pairs->queue[i];
3349 pool_diff_free_filepair(&opt->priv->pool, p);
3353 simple_cleanup:
3354 /* Free memory for renames->pairs[] and combined */
3355 for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) {
3356 free(renames->pairs[s].queue);
3357 DIFF_QUEUE_CLEAR(&renames->pairs[s]);
3359 for (i = 0; i < combined.nr; i++)
3360 pool_diff_free_filepair(&opt->priv->pool, combined.queue[i]);
3361 free(combined.queue);
3363 return clean;
3366 /*** Function Grouping: functions related to process_entries() ***/
3368 static int sort_dirs_next_to_their_children(const char *one, const char *two)
3370 unsigned char c1, c2;
3373 * Here we only care that entries for directories appear adjacent
3374 * to and before files underneath the directory. We can achieve
3375 * that by pretending to add a trailing slash to every file and
3376 * then sorting. In other words, we do not want the natural
3377 * sorting of
3378 * foo
3379 * foo.txt
3380 * foo/bar
3381 * Instead, we want "foo" to sort as though it were "foo/", so that
3382 * we instead get
3383 * foo.txt
3384 * foo
3385 * foo/bar
3386 * To achieve this, we basically implement our own strcmp, except that
3387 * if we get to the end of either string instead of comparing NUL to
3388 * another character, we compare '/' to it.
3390 * If this unusual "sort as though '/' were appended" perplexes
3391 * you, perhaps it will help to note that this is not the final
3392 * sort. write_tree() will sort again without the trailing slash
3393 * magic, but just on paths immediately under a given tree.
3395 * The reason to not use df_name_compare directly was that it was
3396 * just too expensive (we don't have the string lengths handy), so
3397 * it was reimplemented.
3401 * NOTE: This function will never be called with two equal strings,
3402 * because it is used to sort the keys of a strmap, and strmaps have
3403 * unique keys by construction. That simplifies our c1==c2 handling
3404 * below.
3407 while (*one && (*one == *two)) {
3408 one++;
3409 two++;
3412 c1 = *one ? *one : '/';
3413 c2 = *two ? *two : '/';
3415 if (c1 == c2) {
3416 /* Getting here means one is a leading directory of the other */
3417 return (*one) ? 1 : -1;
3418 } else
3419 return c1 - c2;
3422 static int read_oid_strbuf(struct merge_options *opt,
3423 const struct object_id *oid,
3424 struct strbuf *dst)
3426 void *buf;
3427 enum object_type type;
3428 unsigned long size;
3429 buf = read_object_file(oid, &type, &size);
3430 if (!buf)
3431 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3432 if (type != OBJ_BLOB) {
3433 free(buf);
3434 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3436 strbuf_attach(dst, buf, size, size + 1);
3437 return 0;
3440 static int blob_unchanged(struct merge_options *opt,
3441 const struct version_info *base,
3442 const struct version_info *side,
3443 const char *path)
3445 struct strbuf basebuf = STRBUF_INIT;
3446 struct strbuf sidebuf = STRBUF_INIT;
3447 int ret = 0; /* assume changed for safety */
3448 struct index_state *idx = &opt->priv->attr_index;
3450 if (!idx->initialized)
3451 initialize_attr_index(opt);
3453 if (base->mode != side->mode)
3454 return 0;
3455 if (oideq(&base->oid, &side->oid))
3456 return 1;
3458 if (read_oid_strbuf(opt, &base->oid, &basebuf) ||
3459 read_oid_strbuf(opt, &side->oid, &sidebuf))
3460 goto error_return;
3462 * Note: binary | is used so that both renormalizations are
3463 * performed. Comparison can be skipped if both files are
3464 * unchanged since their sha1s have already been compared.
3466 if (renormalize_buffer(idx, path, basebuf.buf, basebuf.len, &basebuf) |
3467 renormalize_buffer(idx, path, sidebuf.buf, sidebuf.len, &sidebuf))
3468 ret = (basebuf.len == sidebuf.len &&
3469 !memcmp(basebuf.buf, sidebuf.buf, basebuf.len));
3471 error_return:
3472 strbuf_release(&basebuf);
3473 strbuf_release(&sidebuf);
3474 return ret;
3477 struct directory_versions {
3479 * versions: list of (basename -> version_info)
3481 * The basenames are in reverse lexicographic order of full pathnames,
3482 * as processed in process_entries(). This puts all entries within
3483 * a directory together, and covers the directory itself after
3484 * everything within it, allowing us to write subtrees before needing
3485 * to record information for the tree itself.
3487 struct string_list versions;
3490 * offsets: list of (full relative path directories -> integer offsets)
3492 * Since versions contains basenames from files in multiple different
3493 * directories, we need to know which entries in versions correspond
3494 * to which directories. Values of e.g.
3495 * "" 0
3496 * src 2
3497 * src/moduleA 5
3498 * Would mean that entries 0-1 of versions are files in the toplevel
3499 * directory, entries 2-4 are files under src/, and the remaining
3500 * entries starting at index 5 are files under src/moduleA/.
3502 struct string_list offsets;
3505 * last_directory: directory that previously processed file found in
3507 * last_directory starts NULL, but records the directory in which the
3508 * previous file was found within. As soon as
3509 * directory(current_file) != last_directory
3510 * then we need to start updating accounting in versions & offsets.
3511 * Note that last_directory is always the last path in "offsets" (or
3512 * NULL if "offsets" is empty) so this exists just for quick access.
3514 const char *last_directory;
3516 /* last_directory_len: cached computation of strlen(last_directory) */
3517 unsigned last_directory_len;
3520 static int tree_entry_order(const void *a_, const void *b_)
3522 const struct string_list_item *a = a_;
3523 const struct string_list_item *b = b_;
3525 const struct merged_info *ami = a->util;
3526 const struct merged_info *bmi = b->util;
3527 return base_name_compare(a->string, strlen(a->string), ami->result.mode,
3528 b->string, strlen(b->string), bmi->result.mode);
3531 static void write_tree(struct object_id *result_oid,
3532 struct string_list *versions,
3533 unsigned int offset,
3534 size_t hash_size)
3536 size_t maxlen = 0, extra;
3537 unsigned int nr;
3538 struct strbuf buf = STRBUF_INIT;
3539 int i;
3541 assert(offset <= versions->nr);
3542 nr = versions->nr - offset;
3543 if (versions->nr)
3544 /* No need for STABLE_QSORT -- filenames must be unique */
3545 QSORT(versions->items + offset, nr, tree_entry_order);
3547 /* Pre-allocate some space in buf */
3548 extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3549 for (i = 0; i < nr; i++) {
3550 maxlen += strlen(versions->items[offset+i].string) + extra;
3552 strbuf_grow(&buf, maxlen);
3554 /* Write each entry out to buf */
3555 for (i = 0; i < nr; i++) {
3556 struct merged_info *mi = versions->items[offset+i].util;
3557 struct version_info *ri = &mi->result;
3558 strbuf_addf(&buf, "%o %s%c",
3559 ri->mode,
3560 versions->items[offset+i].string, '\0');
3561 strbuf_add(&buf, ri->oid.hash, hash_size);
3564 /* Write this object file out, and record in result_oid */
3565 write_object_file(buf.buf, buf.len, OBJ_TREE, result_oid);
3566 strbuf_release(&buf);
3569 static void record_entry_for_tree(struct directory_versions *dir_metadata,
3570 const char *path,
3571 struct merged_info *mi)
3573 const char *basename;
3575 if (mi->is_null)
3576 /* nothing to record */
3577 return;
3579 basename = path + mi->basename_offset;
3580 assert(strchr(basename, '/') == NULL);
3581 string_list_append(&dir_metadata->versions,
3582 basename)->util = &mi->result;
3585 static void write_completed_directory(struct merge_options *opt,
3586 const char *new_directory_name,
3587 struct directory_versions *info)
3589 const char *prev_dir;
3590 struct merged_info *dir_info = NULL;
3591 unsigned int offset;
3594 * Some explanation of info->versions and info->offsets...
3596 * process_entries() iterates over all relevant files AND
3597 * directories in reverse lexicographic order, and calls this
3598 * function. Thus, an example of the paths that process_entries()
3599 * could operate on (along with the directories for those paths
3600 * being shown) is:
3602 * xtract.c ""
3603 * tokens.txt ""
3604 * src/moduleB/umm.c src/moduleB
3605 * src/moduleB/stuff.h src/moduleB
3606 * src/moduleB/baz.c src/moduleB
3607 * src/moduleB src
3608 * src/moduleA/foo.c src/moduleA
3609 * src/moduleA/bar.c src/moduleA
3610 * src/moduleA src
3611 * src ""
3612 * Makefile ""
3614 * info->versions:
3616 * always contains the unprocessed entries and their
3617 * version_info information. For example, after the first five
3618 * entries above, info->versions would be:
3620 * xtract.c <xtract.c's version_info>
3621 * token.txt <token.txt's version_info>
3622 * umm.c <src/moduleB/umm.c's version_info>
3623 * stuff.h <src/moduleB/stuff.h's version_info>
3624 * baz.c <src/moduleB/baz.c's version_info>
3626 * Once a subdirectory is completed we remove the entries in
3627 * that subdirectory from info->versions, writing it as a tree
3628 * (write_tree()). Thus, as soon as we get to src/moduleB,
3629 * info->versions would be updated to
3631 * xtract.c <xtract.c's version_info>
3632 * token.txt <token.txt's version_info>
3633 * moduleB <src/moduleB's version_info>
3635 * info->offsets:
3637 * helps us track which entries in info->versions correspond to
3638 * which directories. When we are N directories deep (e.g. 4
3639 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3640 * directories (+1 because of toplevel dir). Corresponding to
3641 * the info->versions example above, after processing five entries
3642 * info->offsets will be:
3644 * "" 0
3645 * src/moduleB 2
3647 * which is used to know that xtract.c & token.txt are from the
3648 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3649 * src/moduleB directory. Again, following the example above,
3650 * once we need to process src/moduleB, then info->offsets is
3651 * updated to
3653 * "" 0
3654 * src 2
3656 * which says that moduleB (and only moduleB so far) is in the
3657 * src directory.
3659 * One unique thing to note about info->offsets here is that
3660 * "src" was not added to info->offsets until there was a path
3661 * (a file OR directory) immediately below src/ that got
3662 * processed.
3664 * Since process_entry() just appends new entries to info->versions,
3665 * write_completed_directory() only needs to do work if the next path
3666 * is in a directory that is different than the last directory found
3667 * in info->offsets.
3671 * If we are working with the same directory as the last entry, there
3672 * is no work to do. (See comments above the directory_name member of
3673 * struct merged_info for why we can use pointer comparison instead of
3674 * strcmp here.)
3676 if (new_directory_name == info->last_directory)
3677 return;
3680 * If we are just starting (last_directory is NULL), or last_directory
3681 * is a prefix of the current directory, then we can just update
3682 * info->offsets to record the offset where we started this directory
3683 * and update last_directory to have quick access to it.
3685 if (info->last_directory == NULL ||
3686 !strncmp(new_directory_name, info->last_directory,
3687 info->last_directory_len)) {
3688 uintptr_t offset = info->versions.nr;
3690 info->last_directory = new_directory_name;
3691 info->last_directory_len = strlen(info->last_directory);
3693 * Record the offset into info->versions where we will
3694 * start recording basenames of paths found within
3695 * new_directory_name.
3697 string_list_append(&info->offsets,
3698 info->last_directory)->util = (void*)offset;
3699 return;
3703 * The next entry that will be processed will be within
3704 * new_directory_name. Since at this point we know that
3705 * new_directory_name is within a different directory than
3706 * info->last_directory, we have all entries for info->last_directory
3707 * in info->versions and we need to create a tree object for them.
3709 dir_info = strmap_get(&opt->priv->paths, info->last_directory);
3710 assert(dir_info);
3711 offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util;
3712 if (offset == info->versions.nr) {
3714 * Actually, we don't need to create a tree object in this
3715 * case. Whenever all files within a directory disappear
3716 * during the merge (e.g. unmodified on one side and
3717 * deleted on the other, or files were renamed elsewhere),
3718 * then we get here and the directory itself needs to be
3719 * omitted from its parent tree as well.
3721 dir_info->is_null = 1;
3722 } else {
3724 * Write out the tree to the git object directory, and also
3725 * record the mode and oid in dir_info->result.
3727 dir_info->is_null = 0;
3728 dir_info->result.mode = S_IFDIR;
3729 write_tree(&dir_info->result.oid, &info->versions, offset,
3730 opt->repo->hash_algo->rawsz);
3734 * We've now used several entries from info->versions and one entry
3735 * from info->offsets, so we get rid of those values.
3737 info->offsets.nr--;
3738 info->versions.nr = offset;
3741 * Now we've taken care of the completed directory, but we need to
3742 * prepare things since future entries will be in
3743 * new_directory_name. (In particular, process_entry() will be
3744 * appending new entries to info->versions.) So, we need to make
3745 * sure new_directory_name is the last entry in info->offsets.
3747 prev_dir = info->offsets.nr == 0 ? NULL :
3748 info->offsets.items[info->offsets.nr-1].string;
3749 if (new_directory_name != prev_dir) {
3750 uintptr_t c = info->versions.nr;
3751 string_list_append(&info->offsets,
3752 new_directory_name)->util = (void*)c;
3755 /* And, of course, we need to update last_directory to match. */
3756 info->last_directory = new_directory_name;
3757 info->last_directory_len = strlen(info->last_directory);
3760 /* Per entry merge function */
3761 static void process_entry(struct merge_options *opt,
3762 const char *path,
3763 struct conflict_info *ci,
3764 struct directory_versions *dir_metadata)
3766 int df_file_index = 0;
3768 VERIFY_CI(ci);
3769 assert(ci->filemask >= 0 && ci->filemask <= 7);
3770 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3771 assert(ci->match_mask == 0 || ci->match_mask == 3 ||
3772 ci->match_mask == 5 || ci->match_mask == 6);
3774 if (ci->dirmask) {
3775 record_entry_for_tree(dir_metadata, path, &ci->merged);
3776 if (ci->filemask == 0)
3777 /* nothing else to handle */
3778 return;
3779 assert(ci->df_conflict);
3782 if (ci->df_conflict && ci->merged.result.mode == 0) {
3783 int i;
3786 * directory no longer in the way, but we do have a file we
3787 * need to place here so we need to clean away the "directory
3788 * merges to nothing" result.
3790 ci->df_conflict = 0;
3791 assert(ci->filemask != 0);
3792 ci->merged.clean = 0;
3793 ci->merged.is_null = 0;
3794 /* and we want to zero out any directory-related entries */
3795 ci->match_mask = (ci->match_mask & ~ci->dirmask);
3796 ci->dirmask = 0;
3797 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
3798 if (ci->filemask & (1 << i))
3799 continue;
3800 ci->stages[i].mode = 0;
3801 oidcpy(&ci->stages[i].oid, null_oid());
3803 } else if (ci->df_conflict && ci->merged.result.mode != 0) {
3805 * This started out as a D/F conflict, and the entries in
3806 * the competing directory were not removed by the merge as
3807 * evidenced by write_completed_directory() writing a value
3808 * to ci->merged.result.mode.
3810 struct conflict_info *new_ci;
3811 const char *branch;
3812 const char *old_path = path;
3813 int i;
3815 assert(ci->merged.result.mode == S_IFDIR);
3818 * If filemask is 1, we can just ignore the file as having
3819 * been deleted on both sides. We do not want to overwrite
3820 * ci->merged.result, since it stores the tree for all the
3821 * files under it.
3823 if (ci->filemask == 1) {
3824 ci->filemask = 0;
3825 return;
3829 * This file still exists on at least one side, and we want
3830 * the directory to remain here, so we need to move this
3831 * path to some new location.
3833 new_ci = mem_pool_calloc(&opt->priv->pool, 1, sizeof(*new_ci));
3835 /* We don't really want new_ci->merged.result copied, but it'll
3836 * be overwritten below so it doesn't matter. We also don't
3837 * want any directory mode/oid values copied, but we'll zero
3838 * those out immediately. We do want the rest of ci copied.
3840 memcpy(new_ci, ci, sizeof(*ci));
3841 new_ci->match_mask = (new_ci->match_mask & ~new_ci->dirmask);
3842 new_ci->dirmask = 0;
3843 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
3844 if (new_ci->filemask & (1 << i))
3845 continue;
3846 /* zero out any entries related to directories */
3847 new_ci->stages[i].mode = 0;
3848 oidcpy(&new_ci->stages[i].oid, null_oid());
3852 * Find out which side this file came from; note that we
3853 * cannot just use ci->filemask, because renames could cause
3854 * the filemask to go back to 7. So we use dirmask, then
3855 * pick the opposite side's index.
3857 df_file_index = (ci->dirmask & (1 << 1)) ? 2 : 1;
3858 branch = (df_file_index == 1) ? opt->branch1 : opt->branch2;
3859 path = unique_path(opt, path, branch);
3860 strmap_put(&opt->priv->paths, path, new_ci);
3862 path_msg(opt, CONFLICT_FILE_DIRECTORY, 0,
3863 path, old_path, NULL, NULL,
3864 _("CONFLICT (file/directory): directory in the way "
3865 "of %s from %s; moving it to %s instead."),
3866 old_path, branch, path);
3869 * Zero out the filemask for the old ci. At this point, ci
3870 * was just an entry for a directory, so we don't need to
3871 * do anything more with it.
3873 ci->filemask = 0;
3876 * Now note that we're working on the new entry (path was
3877 * updated above.
3879 ci = new_ci;
3883 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3884 * which the code goes through even for the df_conflict cases
3885 * above.
3887 if (ci->match_mask) {
3888 ci->merged.clean = !ci->df_conflict && !ci->path_conflict;
3889 if (ci->match_mask == 6) {
3890 /* stages[1] == stages[2] */
3891 ci->merged.result.mode = ci->stages[1].mode;
3892 oidcpy(&ci->merged.result.oid, &ci->stages[1].oid);
3893 } else {
3894 /* determine the mask of the side that didn't match */
3895 unsigned int othermask = 7 & ~ci->match_mask;
3896 int side = (othermask == 4) ? 2 : 1;
3898 ci->merged.result.mode = ci->stages[side].mode;
3899 ci->merged.is_null = !ci->merged.result.mode;
3900 if (ci->merged.is_null)
3901 ci->merged.clean = 1;
3902 oidcpy(&ci->merged.result.oid, &ci->stages[side].oid);
3904 assert(othermask == 2 || othermask == 4);
3905 assert(ci->merged.is_null ==
3906 (ci->filemask == ci->match_mask));
3908 } else if (ci->filemask >= 6 &&
3909 (S_IFMT & ci->stages[1].mode) !=
3910 (S_IFMT & ci->stages[2].mode)) {
3911 /* Two different items from (file/submodule/symlink) */
3912 if (opt->priv->call_depth) {
3913 /* Just use the version from the merge base */
3914 ci->merged.clean = 0;
3915 oidcpy(&ci->merged.result.oid, &ci->stages[0].oid);
3916 ci->merged.result.mode = ci->stages[0].mode;
3917 ci->merged.is_null = (ci->merged.result.mode == 0);
3918 } else {
3919 /* Handle by renaming one or both to separate paths. */
3920 unsigned o_mode = ci->stages[0].mode;
3921 unsigned a_mode = ci->stages[1].mode;
3922 unsigned b_mode = ci->stages[2].mode;
3923 struct conflict_info *new_ci;
3924 const char *a_path = NULL, *b_path = NULL;
3925 int rename_a = 0, rename_b = 0;
3927 new_ci = mem_pool_alloc(&opt->priv->pool,
3928 sizeof(*new_ci));
3930 if (S_ISREG(a_mode))
3931 rename_a = 1;
3932 else if (S_ISREG(b_mode))
3933 rename_b = 1;
3934 else {
3935 rename_a = 1;
3936 rename_b = 1;
3939 if (rename_a)
3940 a_path = unique_path(opt, path, opt->branch1);
3941 if (rename_b)
3942 b_path = unique_path(opt, path, opt->branch2);
3944 if (rename_a && rename_b) {
3945 path_msg(opt, CONFLICT_DISTINCT_MODES, 0,
3946 path, a_path, b_path, NULL,
3947 _("CONFLICT (distinct types): %s had "
3948 "different types on each side; "
3949 "renamed both of them so each can "
3950 "be recorded somewhere."),
3951 path);
3952 } else {
3953 path_msg(opt, CONFLICT_DISTINCT_MODES, 0,
3954 path, rename_a ? a_path : b_path,
3955 NULL, NULL,
3956 _("CONFLICT (distinct types): %s had "
3957 "different types on each side; "
3958 "renamed one of them so each can be "
3959 "recorded somewhere."),
3960 path);
3963 ci->merged.clean = 0;
3964 memcpy(new_ci, ci, sizeof(*new_ci));
3966 /* Put b into new_ci, removing a from stages */
3967 new_ci->merged.result.mode = ci->stages[2].mode;
3968 oidcpy(&new_ci->merged.result.oid, &ci->stages[2].oid);
3969 new_ci->stages[1].mode = 0;
3970 oidcpy(&new_ci->stages[1].oid, null_oid());
3971 new_ci->filemask = 5;
3972 if ((S_IFMT & b_mode) != (S_IFMT & o_mode)) {
3973 new_ci->stages[0].mode = 0;
3974 oidcpy(&new_ci->stages[0].oid, null_oid());
3975 new_ci->filemask = 4;
3978 /* Leave only a in ci, fixing stages. */
3979 ci->merged.result.mode = ci->stages[1].mode;
3980 oidcpy(&ci->merged.result.oid, &ci->stages[1].oid);
3981 ci->stages[2].mode = 0;
3982 oidcpy(&ci->stages[2].oid, null_oid());
3983 ci->filemask = 3;
3984 if ((S_IFMT & a_mode) != (S_IFMT & o_mode)) {
3985 ci->stages[0].mode = 0;
3986 oidcpy(&ci->stages[0].oid, null_oid());
3987 ci->filemask = 2;
3990 /* Insert entries into opt->priv_paths */
3991 assert(rename_a || rename_b);
3992 if (rename_a)
3993 strmap_put(&opt->priv->paths, a_path, ci);
3995 if (!rename_b)
3996 b_path = path;
3997 strmap_put(&opt->priv->paths, b_path, new_ci);
3999 if (rename_a && rename_b)
4000 strmap_remove(&opt->priv->paths, path, 0);
4003 * Do special handling for b_path since process_entry()
4004 * won't be called on it specially.
4006 strmap_put(&opt->priv->conflicted, b_path, new_ci);
4007 record_entry_for_tree(dir_metadata, b_path,
4008 &new_ci->merged);
4011 * Remaining code for processing this entry should
4012 * think in terms of processing a_path.
4014 if (a_path)
4015 path = a_path;
4017 } else if (ci->filemask >= 6) {
4018 /* Need a two-way or three-way content merge */
4019 struct version_info merged_file;
4020 unsigned clean_merge;
4021 struct version_info *o = &ci->stages[0];
4022 struct version_info *a = &ci->stages[1];
4023 struct version_info *b = &ci->stages[2];
4025 clean_merge = handle_content_merge(opt, path, o, a, b,
4026 ci->pathnames,
4027 opt->priv->call_depth * 2,
4028 &merged_file);
4029 ci->merged.clean = clean_merge &&
4030 !ci->df_conflict && !ci->path_conflict;
4031 ci->merged.result.mode = merged_file.mode;
4032 ci->merged.is_null = (merged_file.mode == 0);
4033 oidcpy(&ci->merged.result.oid, &merged_file.oid);
4034 if (clean_merge && ci->df_conflict) {
4035 assert(df_file_index == 1 || df_file_index == 2);
4036 ci->filemask = 1 << df_file_index;
4037 ci->stages[df_file_index].mode = merged_file.mode;
4038 oidcpy(&ci->stages[df_file_index].oid, &merged_file.oid);
4040 if (!clean_merge) {
4041 const char *reason = _("content");
4042 if (ci->filemask == 6)
4043 reason = _("add/add");
4044 if (S_ISGITLINK(merged_file.mode))
4045 reason = _("submodule");
4046 path_msg(opt, CONFLICT_CONTENTS, 0,
4047 path, NULL, NULL, NULL,
4048 _("CONFLICT (%s): Merge conflict in %s"),
4049 reason, path);
4051 } else if (ci->filemask == 3 || ci->filemask == 5) {
4052 /* Modify/delete */
4053 const char *modify_branch, *delete_branch;
4054 int side = (ci->filemask == 5) ? 2 : 1;
4055 int index = opt->priv->call_depth ? 0 : side;
4057 ci->merged.result.mode = ci->stages[index].mode;
4058 oidcpy(&ci->merged.result.oid, &ci->stages[index].oid);
4059 ci->merged.clean = 0;
4061 modify_branch = (side == 1) ? opt->branch1 : opt->branch2;
4062 delete_branch = (side == 1) ? opt->branch2 : opt->branch1;
4064 if (opt->renormalize &&
4065 blob_unchanged(opt, &ci->stages[0], &ci->stages[side],
4066 path)) {
4067 if (!ci->path_conflict) {
4069 * Blob unchanged after renormalization, so
4070 * there's no modify/delete conflict after all;
4071 * we can just remove the file.
4073 ci->merged.is_null = 1;
4074 ci->merged.clean = 1;
4076 * file goes away => even if there was a
4077 * directory/file conflict there isn't one now.
4079 ci->df_conflict = 0;
4080 } else {
4081 /* rename/delete, so conflict remains */
4083 } else if (ci->path_conflict &&
4084 oideq(&ci->stages[0].oid, &ci->stages[side].oid)) {
4086 * This came from a rename/delete; no action to take,
4087 * but avoid printing "modify/delete" conflict notice
4088 * since the contents were not modified.
4090 } else {
4091 path_msg(opt, CONFLICT_MODIFY_DELETE, 0,
4092 path, NULL, NULL, NULL,
4093 _("CONFLICT (modify/delete): %s deleted in %s "
4094 "and modified in %s. Version %s of %s left "
4095 "in tree."),
4096 path, delete_branch, modify_branch,
4097 modify_branch, path);
4099 } else if (ci->filemask == 2 || ci->filemask == 4) {
4100 /* Added on one side */
4101 int side = (ci->filemask == 4) ? 2 : 1;
4102 ci->merged.result.mode = ci->stages[side].mode;
4103 oidcpy(&ci->merged.result.oid, &ci->stages[side].oid);
4104 ci->merged.clean = !ci->df_conflict && !ci->path_conflict;
4105 } else if (ci->filemask == 1) {
4106 /* Deleted on both sides */
4107 ci->merged.is_null = 1;
4108 ci->merged.result.mode = 0;
4109 oidcpy(&ci->merged.result.oid, null_oid());
4110 assert(!ci->df_conflict);
4111 ci->merged.clean = !ci->path_conflict;
4115 * If still conflicted, record it separately. This allows us to later
4116 * iterate over just conflicted entries when updating the index instead
4117 * of iterating over all entries.
4119 if (!ci->merged.clean)
4120 strmap_put(&opt->priv->conflicted, path, ci);
4122 /* Record metadata for ci->merged in dir_metadata */
4123 record_entry_for_tree(dir_metadata, path, &ci->merged);
4126 static void prefetch_for_content_merges(struct merge_options *opt,
4127 struct string_list *plist)
4129 struct string_list_item *e;
4130 struct oid_array to_fetch = OID_ARRAY_INIT;
4132 if (opt->repo != the_repository || !has_promisor_remote())
4133 return;
4135 for (e = &plist->items[plist->nr-1]; e >= plist->items; --e) {
4136 /* char *path = e->string; */
4137 struct conflict_info *ci = e->util;
4138 int i;
4140 /* Ignore clean entries */
4141 if (ci->merged.clean)
4142 continue;
4144 /* Ignore entries that don't need a content merge */
4145 if (ci->match_mask || ci->filemask < 6 ||
4146 !S_ISREG(ci->stages[1].mode) ||
4147 !S_ISREG(ci->stages[2].mode) ||
4148 oideq(&ci->stages[1].oid, &ci->stages[2].oid))
4149 continue;
4151 /* Also don't need content merge if base matches either side */
4152 if (ci->filemask == 7 &&
4153 S_ISREG(ci->stages[0].mode) &&
4154 (oideq(&ci->stages[0].oid, &ci->stages[1].oid) ||
4155 oideq(&ci->stages[0].oid, &ci->stages[2].oid)))
4156 continue;
4158 for (i = 0; i < 3; i++) {
4159 unsigned side_mask = (1 << i);
4160 struct version_info *vi = &ci->stages[i];
4162 if ((ci->filemask & side_mask) &&
4163 S_ISREG(vi->mode) &&
4164 oid_object_info_extended(opt->repo, &vi->oid, NULL,
4165 OBJECT_INFO_FOR_PREFETCH))
4166 oid_array_append(&to_fetch, &vi->oid);
4170 promisor_remote_get_direct(opt->repo, to_fetch.oid, to_fetch.nr);
4171 oid_array_clear(&to_fetch);
4174 static void process_entries(struct merge_options *opt,
4175 struct object_id *result_oid)
4177 struct hashmap_iter iter;
4178 struct strmap_entry *e;
4179 struct string_list plist = STRING_LIST_INIT_NODUP;
4180 struct string_list_item *entry;
4181 struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP,
4182 STRING_LIST_INIT_NODUP,
4183 NULL, 0 };
4185 trace2_region_enter("merge", "process_entries setup", opt->repo);
4186 if (strmap_empty(&opt->priv->paths)) {
4187 oidcpy(result_oid, opt->repo->hash_algo->empty_tree);
4188 return;
4191 /* Hack to pre-allocate plist to the desired size */
4192 trace2_region_enter("merge", "plist grow", opt->repo);
4193 ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc);
4194 trace2_region_leave("merge", "plist grow", opt->repo);
4196 /* Put every entry from paths into plist, then sort */
4197 trace2_region_enter("merge", "plist copy", opt->repo);
4198 strmap_for_each_entry(&opt->priv->paths, &iter, e) {
4199 string_list_append(&plist, e->key)->util = e->value;
4201 trace2_region_leave("merge", "plist copy", opt->repo);
4203 trace2_region_enter("merge", "plist special sort", opt->repo);
4204 plist.cmp = sort_dirs_next_to_their_children;
4205 string_list_sort(&plist);
4206 trace2_region_leave("merge", "plist special sort", opt->repo);
4208 trace2_region_leave("merge", "process_entries setup", opt->repo);
4211 * Iterate over the items in reverse order, so we can handle paths
4212 * below a directory before needing to handle the directory itself.
4214 * This allows us to write subtrees before we need to write trees,
4215 * and it also enables sane handling of directory/file conflicts
4216 * (because it allows us to know whether the directory is still in
4217 * the way when it is time to process the file at the same path).
4219 trace2_region_enter("merge", "processing", opt->repo);
4220 prefetch_for_content_merges(opt, &plist);
4221 for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) {
4222 char *path = entry->string;
4224 * NOTE: mi may actually be a pointer to a conflict_info, but
4225 * we have to check mi->clean first to see if it's safe to
4226 * reassign to such a pointer type.
4228 struct merged_info *mi = entry->util;
4230 write_completed_directory(opt, mi->directory_name,
4231 &dir_metadata);
4232 if (mi->clean)
4233 record_entry_for_tree(&dir_metadata, path, mi);
4234 else {
4235 struct conflict_info *ci = (struct conflict_info *)mi;
4236 process_entry(opt, path, ci, &dir_metadata);
4239 trace2_region_leave("merge", "processing", opt->repo);
4241 trace2_region_enter("merge", "process_entries cleanup", opt->repo);
4242 if (dir_metadata.offsets.nr != 1 ||
4243 (uintptr_t)dir_metadata.offsets.items[0].util != 0) {
4244 printf("dir_metadata.offsets.nr = %"PRIuMAX" (should be 1)\n",
4245 (uintmax_t)dir_metadata.offsets.nr);
4246 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4247 (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util);
4248 fflush(stdout);
4249 BUG("dir_metadata accounting completely off; shouldn't happen");
4251 write_tree(result_oid, &dir_metadata.versions, 0,
4252 opt->repo->hash_algo->rawsz);
4253 string_list_clear(&plist, 0);
4254 string_list_clear(&dir_metadata.versions, 0);
4255 string_list_clear(&dir_metadata.offsets, 0);
4256 trace2_region_leave("merge", "process_entries cleanup", opt->repo);
4259 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4261 static int checkout(struct merge_options *opt,
4262 struct tree *prev,
4263 struct tree *next)
4265 /* Switch the index/working copy from old to new */
4266 int ret;
4267 struct tree_desc trees[2];
4268 struct unpack_trees_options unpack_opts;
4270 memset(&unpack_opts, 0, sizeof(unpack_opts));
4271 unpack_opts.head_idx = -1;
4272 unpack_opts.src_index = opt->repo->index;
4273 unpack_opts.dst_index = opt->repo->index;
4275 setup_unpack_trees_porcelain(&unpack_opts, "merge");
4278 * NOTE: if this were just "git checkout" code, we would probably
4279 * read or refresh the cache and check for a conflicted index, but
4280 * builtin/merge.c or sequencer.c really needs to read the index
4281 * and check for conflicted entries before starting merging for a
4282 * good user experience (no sense waiting for merges/rebases before
4283 * erroring out), so there's no reason to duplicate that work here.
4286 /* 2-way merge to the new branch */
4287 unpack_opts.update = 1;
4288 unpack_opts.merge = 1;
4289 unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */
4290 unpack_opts.verbose_update = (opt->verbosity > 2);
4291 unpack_opts.fn = twoway_merge;
4292 unpack_opts.preserve_ignored = 0; /* FIXME: !opts->overwrite_ignore */
4293 parse_tree(prev);
4294 init_tree_desc(&trees[0], prev->buffer, prev->size);
4295 parse_tree(next);
4296 init_tree_desc(&trees[1], next->buffer, next->size);
4298 ret = unpack_trees(2, trees, &unpack_opts);
4299 clear_unpack_trees_porcelain(&unpack_opts);
4300 return ret;
4303 static int record_conflicted_index_entries(struct merge_options *opt)
4305 struct hashmap_iter iter;
4306 struct strmap_entry *e;
4307 struct index_state *index = opt->repo->index;
4308 struct checkout state = CHECKOUT_INIT;
4309 int errs = 0;
4310 int original_cache_nr;
4312 if (strmap_empty(&opt->priv->conflicted))
4313 return 0;
4316 * We are in a conflicted state. These conflicts might be inside
4317 * sparse-directory entries, so check if any entries are outside
4318 * of the sparse-checkout cone preemptively.
4320 * We set original_cache_nr below, but that might change if
4321 * index_name_pos() calls ask for paths within sparse directories.
4323 strmap_for_each_entry(&opt->priv->conflicted, &iter, e) {
4324 if (!path_in_sparse_checkout(e->key, index)) {
4325 ensure_full_index(index);
4326 break;
4330 /* If any entries have skip_worktree set, we'll have to check 'em out */
4331 state.force = 1;
4332 state.quiet = 1;
4333 state.refresh_cache = 1;
4334 state.istate = index;
4335 original_cache_nr = index->cache_nr;
4337 /* Append every entry from conflicted into index, then sort */
4338 strmap_for_each_entry(&opt->priv->conflicted, &iter, e) {
4339 const char *path = e->key;
4340 struct conflict_info *ci = e->value;
4341 int pos;
4342 struct cache_entry *ce;
4343 int i;
4345 VERIFY_CI(ci);
4348 * The index will already have a stage=0 entry for this path,
4349 * because we created an as-merged-as-possible version of the
4350 * file and checkout() moved the working copy and index over
4351 * to that version.
4353 * However, previous iterations through this loop will have
4354 * added unstaged entries to the end of the cache which
4355 * ignore the standard alphabetical ordering of cache
4356 * entries and break invariants needed for index_name_pos()
4357 * to work. However, we know the entry we want is before
4358 * those appended cache entries, so do a temporary swap on
4359 * cache_nr to only look through entries of interest.
4361 SWAP(index->cache_nr, original_cache_nr);
4362 pos = index_name_pos(index, path, strlen(path));
4363 SWAP(index->cache_nr, original_cache_nr);
4364 if (pos < 0) {
4365 if (ci->filemask != 1)
4366 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path);
4367 cache_tree_invalidate_path(index, path);
4368 } else {
4369 ce = index->cache[pos];
4372 * Clean paths with CE_SKIP_WORKTREE set will not be
4373 * written to the working tree by the unpack_trees()
4374 * call in checkout(). Our conflicted entries would
4375 * have appeared clean to that code since we ignored
4376 * the higher order stages. Thus, we need override
4377 * the CE_SKIP_WORKTREE bit and manually write those
4378 * files to the working disk here.
4380 if (ce_skip_worktree(ce)) {
4381 struct stat st;
4383 if (!lstat(path, &st)) {
4384 char *new_name = unique_path(opt,
4385 path,
4386 "cruft");
4388 path_msg(opt, CONFLICT_PRESENT_DESPITE_SKIPPED, 1,
4389 path, NULL, NULL, NULL,
4390 _("Note: %s not up to date and in way of checking out conflicted version; old copy renamed to %s"),
4391 path, new_name);
4392 errs |= rename(path, new_name);
4394 errs |= checkout_entry(ce, &state, NULL, NULL);
4398 * Mark this cache entry for removal and instead add
4399 * new stage>0 entries corresponding to the
4400 * conflicts. If there are many conflicted entries, we
4401 * want to avoid memmove'ing O(NM) entries by
4402 * inserting the new entries one at a time. So,
4403 * instead, we just add the new cache entries to the
4404 * end (ignoring normal index requirements on sort
4405 * order) and sort the index once we're all done.
4407 ce->ce_flags |= CE_REMOVE;
4410 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
4411 struct version_info *vi;
4412 if (!(ci->filemask & (1ul << i)))
4413 continue;
4414 vi = &ci->stages[i];
4415 ce = make_cache_entry(index, vi->mode, &vi->oid,
4416 path, i+1, 0);
4417 add_index_entry(index, ce, ADD_CACHE_JUST_APPEND);
4422 * Remove the unused cache entries (and invalidate the relevant
4423 * cache-trees), then sort the index entries to get the conflicted
4424 * entries we added to the end into their right locations.
4426 remove_marked_cache_entries(index, 1);
4428 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4429 * on filename and secondarily on stage, and (name, stage #) are a
4430 * unique tuple.
4432 QSORT(index->cache, index->cache_nr, cmp_cache_name_compare);
4434 return errs;
4437 void merge_display_update_messages(struct merge_options *opt,
4438 int detailed,
4439 struct merge_result *result)
4441 struct merge_options_internal *opti = result->priv;
4442 struct hashmap_iter iter;
4443 struct strmap_entry *e;
4444 struct string_list olist = STRING_LIST_INIT_NODUP;
4446 if (opt->record_conflict_msgs_as_headers)
4447 BUG("Either display conflict messages or record them as headers, not both");
4449 trace2_region_enter("merge", "display messages", opt->repo);
4451 /* Hack to pre-allocate olist to the desired size */
4452 ALLOC_GROW(olist.items, strmap_get_size(&opti->conflicts),
4453 olist.alloc);
4455 /* Put every entry from output into olist, then sort */
4456 strmap_for_each_entry(&opti->conflicts, &iter, e) {
4457 string_list_append(&olist, e->key)->util = e->value;
4459 string_list_sort(&olist);
4461 /* Iterate over the items, printing them */
4462 for (int path_nr = 0; path_nr < olist.nr; ++path_nr) {
4463 struct string_list *conflicts = olist.items[path_nr].util;
4464 for (int i = 0; i < conflicts->nr; i++) {
4465 struct logical_conflict_info *info =
4466 conflicts->items[i].util;
4468 if (detailed) {
4469 printf("%lu", (unsigned long)info->paths.nr);
4470 putchar('\0');
4471 for (int n = 0; n < info->paths.nr; n++) {
4472 fputs(info->paths.v[n], stdout);
4473 putchar('\0');
4475 fputs(type_short_descriptions[info->type],
4476 stdout);
4477 putchar('\0');
4479 puts(conflicts->items[i].string);
4480 if (detailed)
4481 putchar('\0');
4484 string_list_clear(&olist, 0);
4486 /* Also include needed rename limit adjustment now */
4487 diff_warn_rename_limit("merge.renamelimit",
4488 opti->renames.needed_limit, 0);
4490 trace2_region_leave("merge", "display messages", opt->repo);
4493 void merge_get_conflicted_files(struct merge_result *result,
4494 struct string_list *conflicted_files)
4496 struct hashmap_iter iter;
4497 struct strmap_entry *e;
4498 struct merge_options_internal *opti = result->priv;
4500 strmap_for_each_entry(&opti->conflicted, &iter, e) {
4501 const char *path = e->key;
4502 struct conflict_info *ci = e->value;
4503 int i;
4505 VERIFY_CI(ci);
4507 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
4508 struct stage_info *si;
4510 if (!(ci->filemask & (1ul << i)))
4511 continue;
4513 si = xmalloc(sizeof(*si));
4514 si->stage = i+1;
4515 si->mode = ci->stages[i].mode;
4516 oidcpy(&si->oid, &ci->stages[i].oid);
4517 string_list_append(conflicted_files, path)->util = si;
4520 /* string_list_sort() uses a stable sort, so we're good */
4521 string_list_sort(conflicted_files);
4524 void merge_switch_to_result(struct merge_options *opt,
4525 struct tree *head,
4526 struct merge_result *result,
4527 int update_worktree_and_index,
4528 int display_update_msgs)
4530 assert(opt->priv == NULL);
4531 if (result->clean >= 0 && update_worktree_and_index) {
4532 const char *filename;
4533 FILE *fp;
4535 trace2_region_enter("merge", "checkout", opt->repo);
4536 if (checkout(opt, head, result->tree)) {
4537 /* failure to function */
4538 result->clean = -1;
4539 return;
4541 trace2_region_leave("merge", "checkout", opt->repo);
4543 trace2_region_enter("merge", "record_conflicted", opt->repo);
4544 opt->priv = result->priv;
4545 if (record_conflicted_index_entries(opt)) {
4546 /* failure to function */
4547 opt->priv = NULL;
4548 result->clean = -1;
4549 return;
4551 opt->priv = NULL;
4552 trace2_region_leave("merge", "record_conflicted", opt->repo);
4554 trace2_region_enter("merge", "write_auto_merge", opt->repo);
4555 filename = git_path_auto_merge(opt->repo);
4556 fp = xfopen(filename, "w");
4557 fprintf(fp, "%s\n", oid_to_hex(&result->tree->object.oid));
4558 fclose(fp);
4559 trace2_region_leave("merge", "write_auto_merge", opt->repo);
4561 if (display_update_msgs)
4562 merge_display_update_messages(opt, /* detailed */ 0, result);
4564 merge_finalize(opt, result);
4567 void merge_finalize(struct merge_options *opt,
4568 struct merge_result *result)
4570 struct merge_options_internal *opti = result->priv;
4572 if (opt->renormalize)
4573 git_attr_set_direction(GIT_ATTR_CHECKIN);
4574 assert(opt->priv == NULL);
4576 clear_or_reinit_internal_opts(opti, 0);
4577 FREE_AND_NULL(opti);
4580 /*** Function Grouping: helper functions for merge_incore_*() ***/
4582 static struct tree *shift_tree_object(struct repository *repo,
4583 struct tree *one, struct tree *two,
4584 const char *subtree_shift)
4586 struct object_id shifted;
4588 if (!*subtree_shift) {
4589 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
4590 } else {
4591 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
4592 subtree_shift);
4594 if (oideq(&two->object.oid, &shifted))
4595 return two;
4596 return lookup_tree(repo, &shifted);
4599 static inline void set_commit_tree(struct commit *c, struct tree *t)
4601 c->maybe_tree = t;
4604 static struct commit *make_virtual_commit(struct repository *repo,
4605 struct tree *tree,
4606 const char *comment)
4608 struct commit *commit = alloc_commit_node(repo);
4610 set_merge_remote_desc(commit, comment, (struct object *)commit);
4611 set_commit_tree(commit, tree);
4612 commit->object.parsed = 1;
4613 return commit;
4616 static void merge_start(struct merge_options *opt, struct merge_result *result)
4618 struct rename_info *renames;
4619 int i;
4620 struct mem_pool *pool = NULL;
4622 /* Sanity checks on opt */
4623 trace2_region_enter("merge", "sanity checks", opt->repo);
4624 assert(opt->repo);
4626 assert(opt->branch1 && opt->branch2);
4628 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
4629 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
4630 assert(opt->rename_limit >= -1);
4631 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
4632 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
4634 assert(opt->xdl_opts >= 0);
4635 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
4636 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
4638 if (opt->msg_header_prefix)
4639 assert(opt->record_conflict_msgs_as_headers);
4642 * detect_renames, verbosity, buffer_output, and obuf are ignored
4643 * fields that were used by "recursive" rather than "ort" -- but
4644 * sanity check them anyway.
4646 assert(opt->detect_renames >= -1 &&
4647 opt->detect_renames <= DIFF_DETECT_COPY);
4648 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
4649 assert(opt->buffer_output <= 2);
4650 assert(opt->obuf.len == 0);
4652 assert(opt->priv == NULL);
4653 if (result->_properly_initialized != 0 &&
4654 result->_properly_initialized != RESULT_INITIALIZED)
4655 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4656 assert(!!result->priv == !!result->_properly_initialized);
4657 if (result->priv) {
4658 opt->priv = result->priv;
4659 result->priv = NULL;
4661 * opt->priv non-NULL means we had results from a previous
4662 * run; do a few sanity checks that user didn't mess with
4663 * it in an obvious fashion.
4665 assert(opt->priv->call_depth == 0);
4666 assert(!opt->priv->toplevel_dir ||
4667 0 == strlen(opt->priv->toplevel_dir));
4669 trace2_region_leave("merge", "sanity checks", opt->repo);
4671 /* Default to histogram diff. Actually, just hardcode it...for now. */
4672 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
4674 /* Handle attr direction stuff for renormalization */
4675 if (opt->renormalize)
4676 git_attr_set_direction(GIT_ATTR_CHECKOUT);
4678 /* Initialization of opt->priv, our internal merge data */
4679 trace2_region_enter("merge", "allocate/init", opt->repo);
4680 if (opt->priv) {
4681 clear_or_reinit_internal_opts(opt->priv, 1);
4682 trace2_region_leave("merge", "allocate/init", opt->repo);
4683 return;
4685 opt->priv = xcalloc(1, sizeof(*opt->priv));
4687 /* Initialization of various renames fields */
4688 renames = &opt->priv->renames;
4689 mem_pool_init(&opt->priv->pool, 0);
4690 pool = &opt->priv->pool;
4691 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) {
4692 strintmap_init_with_options(&renames->dirs_removed[i],
4693 NOT_RELEVANT, pool, 0);
4694 strmap_init_with_options(&renames->dir_rename_count[i],
4695 NULL, 1);
4696 strmap_init_with_options(&renames->dir_renames[i],
4697 NULL, 0);
4699 * relevant_sources uses -1 for the default, because we need
4700 * to be able to distinguish not-in-strintmap from valid
4701 * relevant_source values from enum file_rename_relevance.
4702 * In particular, possibly_cache_new_pair() expects a negative
4703 * value for not-found entries.
4705 strintmap_init_with_options(&renames->relevant_sources[i],
4706 -1 /* explicitly invalid */,
4707 pool, 0);
4708 strmap_init_with_options(&renames->cached_pairs[i],
4709 NULL, 1);
4710 strset_init_with_options(&renames->cached_irrelevant[i],
4711 NULL, 1);
4712 strset_init_with_options(&renames->cached_target_names[i],
4713 NULL, 0);
4715 for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) {
4716 strintmap_init_with_options(&renames->deferred[i].possible_trivial_merges,
4717 0, pool, 0);
4718 strset_init_with_options(&renames->deferred[i].target_dirs,
4719 pool, 1);
4720 renames->deferred[i].trivial_merges_okay = 1; /* 1 == maybe */
4724 * Although we initialize opt->priv->paths with strdup_strings=0,
4725 * that's just to avoid making yet another copy of an allocated
4726 * string. Putting the entry into paths means we are taking
4727 * ownership, so we will later free it.
4729 * In contrast, conflicted just has a subset of keys from paths, so
4730 * we don't want to free those (it'd be a duplicate free).
4732 strmap_init_with_options(&opt->priv->paths, pool, 0);
4733 strmap_init_with_options(&opt->priv->conflicted, pool, 0);
4736 * keys & string_lists in conflicts will sometimes need to outlive
4737 * "paths", so it will have a copy of relevant keys. It's probably
4738 * a small subset of the overall paths that have special output.
4740 strmap_init(&opt->priv->conflicts);
4742 trace2_region_leave("merge", "allocate/init", opt->repo);
4745 static void merge_check_renames_reusable(struct merge_options *opt,
4746 struct merge_result *result,
4747 struct tree *merge_base,
4748 struct tree *side1,
4749 struct tree *side2)
4751 struct rename_info *renames;
4752 struct tree **merge_trees;
4753 struct merge_options_internal *opti = result->priv;
4755 if (!opti)
4756 return;
4758 renames = &opti->renames;
4759 merge_trees = renames->merge_trees;
4762 * Handle case where previous merge operation did not want cache to
4763 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4765 if (!merge_trees[0]) {
4766 assert(!merge_trees[0] && !merge_trees[1] && !merge_trees[2]);
4767 renames->cached_pairs_valid_side = 0; /* neither side valid */
4768 return;
4772 * Handle other cases; note that merge_trees[0..2] will only
4773 * be NULL if opti is, or if all three were manually set to
4774 * NULL by e.g. rename/rename(1to1) handling.
4776 assert(merge_trees[0] && merge_trees[1] && merge_trees[2]);
4778 /* Check if we meet a condition for re-using cached_pairs */
4779 if (oideq(&merge_base->object.oid, &merge_trees[2]->object.oid) &&
4780 oideq(&side1->object.oid, &result->tree->object.oid))
4781 renames->cached_pairs_valid_side = MERGE_SIDE1;
4782 else if (oideq(&merge_base->object.oid, &merge_trees[1]->object.oid) &&
4783 oideq(&side2->object.oid, &result->tree->object.oid))
4784 renames->cached_pairs_valid_side = MERGE_SIDE2;
4785 else
4786 renames->cached_pairs_valid_side = 0; /* neither side valid */
4789 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4792 * Originally from merge_trees_internal(); heavily adapted, though.
4794 static void merge_ort_nonrecursive_internal(struct merge_options *opt,
4795 struct tree *merge_base,
4796 struct tree *side1,
4797 struct tree *side2,
4798 struct merge_result *result)
4800 struct object_id working_tree_oid;
4802 if (opt->subtree_shift) {
4803 side2 = shift_tree_object(opt->repo, side1, side2,
4804 opt->subtree_shift);
4805 merge_base = shift_tree_object(opt->repo, side1, merge_base,
4806 opt->subtree_shift);
4809 redo:
4810 trace2_region_enter("merge", "collect_merge_info", opt->repo);
4811 if (collect_merge_info(opt, merge_base, side1, side2) != 0) {
4813 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4814 * base, and 2-3) the trees for the two trees we're merging.
4816 err(opt, _("collecting merge info failed for trees %s, %s, %s"),
4817 oid_to_hex(&merge_base->object.oid),
4818 oid_to_hex(&side1->object.oid),
4819 oid_to_hex(&side2->object.oid));
4820 result->clean = -1;
4821 return;
4823 trace2_region_leave("merge", "collect_merge_info", opt->repo);
4825 trace2_region_enter("merge", "renames", opt->repo);
4826 result->clean = detect_and_process_renames(opt, merge_base,
4827 side1, side2);
4828 trace2_region_leave("merge", "renames", opt->repo);
4829 if (opt->priv->renames.redo_after_renames == 2) {
4830 trace2_region_enter("merge", "reset_maps", opt->repo);
4831 clear_or_reinit_internal_opts(opt->priv, 1);
4832 trace2_region_leave("merge", "reset_maps", opt->repo);
4833 goto redo;
4836 trace2_region_enter("merge", "process_entries", opt->repo);
4837 process_entries(opt, &working_tree_oid);
4838 trace2_region_leave("merge", "process_entries", opt->repo);
4840 /* Set return values */
4841 result->path_messages = &opt->priv->conflicts;
4843 result->tree = parse_tree_indirect(&working_tree_oid);
4844 /* existence of conflicted entries implies unclean */
4845 result->clean &= strmap_empty(&opt->priv->conflicted);
4846 if (!opt->priv->call_depth) {
4847 result->priv = opt->priv;
4848 result->_properly_initialized = RESULT_INITIALIZED;
4849 opt->priv = NULL;
4854 * Originally from merge_recursive_internal(); somewhat adapted, though.
4856 static void merge_ort_internal(struct merge_options *opt,
4857 struct commit_list *merge_bases,
4858 struct commit *h1,
4859 struct commit *h2,
4860 struct merge_result *result)
4862 struct commit *next;
4863 struct commit *merged_merge_bases;
4864 const char *ancestor_name;
4865 struct strbuf merge_base_abbrev = STRBUF_INIT;
4867 if (!merge_bases) {
4868 merge_bases = get_merge_bases(h1, h2);
4869 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
4870 merge_bases = reverse_commit_list(merge_bases);
4873 merged_merge_bases = pop_commit(&merge_bases);
4874 if (!merged_merge_bases) {
4875 /* if there is no common ancestor, use an empty tree */
4876 struct tree *tree;
4878 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
4879 merged_merge_bases = make_virtual_commit(opt->repo, tree,
4880 "ancestor");
4881 ancestor_name = "empty tree";
4882 } else if (merge_bases) {
4883 ancestor_name = "merged common ancestors";
4884 } else {
4885 strbuf_add_unique_abbrev(&merge_base_abbrev,
4886 &merged_merge_bases->object.oid,
4887 DEFAULT_ABBREV);
4888 ancestor_name = merge_base_abbrev.buf;
4891 for (next = pop_commit(&merge_bases); next;
4892 next = pop_commit(&merge_bases)) {
4893 const char *saved_b1, *saved_b2;
4894 struct commit *prev = merged_merge_bases;
4896 opt->priv->call_depth++;
4898 * When the merge fails, the result contains files
4899 * with conflict markers. The cleanness flag is
4900 * ignored (unless indicating an error), it was never
4901 * actually used, as result of merge_trees has always
4902 * overwritten it: the committed "conflicts" were
4903 * already resolved.
4905 saved_b1 = opt->branch1;
4906 saved_b2 = opt->branch2;
4907 opt->branch1 = "Temporary merge branch 1";
4908 opt->branch2 = "Temporary merge branch 2";
4909 merge_ort_internal(opt, NULL, prev, next, result);
4910 if (result->clean < 0)
4911 return;
4912 opt->branch1 = saved_b1;
4913 opt->branch2 = saved_b2;
4914 opt->priv->call_depth--;
4916 merged_merge_bases = make_virtual_commit(opt->repo,
4917 result->tree,
4918 "merged tree");
4919 commit_list_insert(prev, &merged_merge_bases->parents);
4920 commit_list_insert(next, &merged_merge_bases->parents->next);
4922 clear_or_reinit_internal_opts(opt->priv, 1);
4925 opt->ancestor = ancestor_name;
4926 merge_ort_nonrecursive_internal(opt,
4927 repo_get_commit_tree(opt->repo,
4928 merged_merge_bases),
4929 repo_get_commit_tree(opt->repo, h1),
4930 repo_get_commit_tree(opt->repo, h2),
4931 result);
4932 strbuf_release(&merge_base_abbrev);
4933 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
4936 void merge_incore_nonrecursive(struct merge_options *opt,
4937 struct tree *merge_base,
4938 struct tree *side1,
4939 struct tree *side2,
4940 struct merge_result *result)
4942 trace2_region_enter("merge", "incore_nonrecursive", opt->repo);
4944 trace2_region_enter("merge", "merge_start", opt->repo);
4945 assert(opt->ancestor != NULL);
4946 merge_check_renames_reusable(opt, result, merge_base, side1, side2);
4947 merge_start(opt, result);
4949 * Record the trees used in this merge, so if there's a next merge in
4950 * a cherry-pick or rebase sequence it might be able to take advantage
4951 * of the cached_pairs in that next merge.
4953 opt->priv->renames.merge_trees[0] = merge_base;
4954 opt->priv->renames.merge_trees[1] = side1;
4955 opt->priv->renames.merge_trees[2] = side2;
4956 trace2_region_leave("merge", "merge_start", opt->repo);
4958 merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result);
4959 trace2_region_leave("merge", "incore_nonrecursive", opt->repo);
4962 void merge_incore_recursive(struct merge_options *opt,
4963 struct commit_list *merge_bases,
4964 struct commit *side1,
4965 struct commit *side2,
4966 struct merge_result *result)
4968 trace2_region_enter("merge", "incore_recursive", opt->repo);
4970 /* We set the ancestor label based on the merge_bases */
4971 assert(opt->ancestor == NULL);
4973 trace2_region_enter("merge", "merge_start", opt->repo);
4974 merge_start(opt, result);
4975 trace2_region_leave("merge", "merge_start", opt->repo);
4977 merge_ort_internal(opt, merge_bases, side1, side2, result);
4978 trace2_region_leave("merge", "incore_recursive", opt->repo);