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
6 * git merge [-s recursive]
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"?)
18 #include "merge-ort.h"
23 #include "cache-tree.h"
25 #include "commit-reach.h"
33 #include "object-store.h"
34 #include "promisor-remote.h"
37 #include "submodule-config.h"
38 #include "submodule.h"
40 #include "unpack-trees.h"
41 #include "xdiff-interface.h"
44 * We have many arrays of size 3. Whenever we have such an array, the
45 * indices refer to one of the sides of the three-way merge. This is so
46 * pervasive that the constants 0, 1, and 2 are used in many places in the
47 * code (especially in arithmetic operations to find the other side's index
48 * or to compute a relevant mask), but sometimes these enum names are used
49 * to aid code clarity.
51 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
52 * referred to there is one of these three sides.
60 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
62 struct traversal_callback_data
{
64 unsigned long dirmask
;
65 struct name_entry names
[3];
68 struct deferred_traversal_data
{
70 * possible_trivial_merges: directories to be explored only when needed
72 * possible_trivial_merges is a map of directory names to
73 * dir_rename_mask. When we detect that a directory is unchanged on
74 * one side, we can sometimes resolve the directory without recursing
75 * into it. Renames are the only things that can prevent such an
76 * optimization. However, for rename sources:
77 * - If no parent directory needed directory rename detection, then
78 * no path under such a directory can be a relevant_source.
79 * and for rename destinations:
80 * - If no cached rename has a target path under the directory AND
81 * - If there are no unpaired relevant_sources elsewhere in the
83 * then we don't need any path under this directory for a rename
84 * destination. The only way to know the last item above is to defer
85 * handling such directories until the end of collect_merge_info(),
86 * in handle_deferred_entries().
88 * For each we store dir_rename_mask, since that's the only bit of
89 * information we need, other than the path, to resume the recursive
92 struct strintmap possible_trivial_merges
;
95 * trivial_merges_okay: if trivial directory merges are okay
97 * See possible_trivial_merges above. The "no unpaired
98 * relevant_sources elsewhere in the repository" is a single boolean
99 * per merge side, which we store here. Note that while 0 means no,
100 * 1 only means "maybe" rather than "yes"; we optimistically set it
101 * to 1 initially and only clear when we determine it is unsafe to
102 * do trivial directory merges.
104 unsigned trivial_merges_okay
;
107 * target_dirs: ancestor directories of rename targets
109 * target_dirs contains all directory names that are an ancestor of
110 * any rename destination.
112 struct strset target_dirs
;
117 * All variables that are arrays of size 3 correspond to data tracked
118 * for the sides in enum merge_side. Index 0 is almost always unused
119 * because we often only need to track information for MERGE_SIDE1 and
120 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
121 * are determined relative to what changed since the MERGE_BASE).
125 * pairs: pairing of filenames from diffcore_rename()
127 struct diff_queue_struct pairs
[3];
130 * dirs_removed: directories removed on a given side of history.
132 * The keys of dirs_removed[side] are the directories that were removed
133 * on the given side of history. The value of the strintmap for each
134 * directory is a value from enum dir_rename_relevance.
136 struct strintmap dirs_removed
[3];
139 * dir_rename_count: tracking where parts of a directory were renamed to
141 * When files in a directory are renamed, they may not all go to the
142 * same location. Each strmap here tracks:
143 * old_dir => {new_dir => int}
144 * That is, dir_rename_count[side] is a strmap to a strintmap.
146 struct strmap dir_rename_count
[3];
149 * dir_renames: computed directory renames
151 * This is a map of old_dir => new_dir and is derived in part from
154 struct strmap dir_renames
[3];
157 * relevant_sources: deleted paths wanted in rename detection, and why
159 * relevant_sources is a set of deleted paths on each side of
160 * history for which we need rename detection. If a path is deleted
161 * on one side of history, we need to detect if it is part of a
163 * * the file is modified/deleted on the other side of history
164 * * we need to detect renames for an ancestor directory
165 * If neither of those are true, we can skip rename detection for
166 * that path. The reason is stored as a value from enum
167 * file_rename_relevance, as the reason can inform the algorithm in
168 * diffcore_rename_extended().
170 struct strintmap relevant_sources
[3];
172 struct deferred_traversal_data deferred
[3];
176 * 0: optimization removing unmodified potential rename source okay
177 * 2 or 4: optimization okay, but must check for files added to dir
178 * 7: optimization forbidden; need rename source in case of dir rename
180 unsigned dir_rename_mask
:3;
183 * callback_data_*: supporting data structures for alternate traversal
185 * We sometimes need to be able to traverse through all the files
186 * in a given tree before all immediate subdirectories within that
187 * tree. Since traverse_trees() doesn't do that naturally, we have
188 * a traverse_trees_wrapper() that stores any immediate
189 * subdirectories while traversing files, then traverses the
190 * immediate subdirectories later. These callback_data* variables
191 * store the information for the subdirectories so that we can do
192 * that traversal order.
194 struct traversal_callback_data
*callback_data
;
195 int callback_data_nr
, callback_data_alloc
;
196 char *callback_data_traverse_path
;
199 * merge_trees: trees passed to the merge algorithm for the merge
201 * merge_trees records the trees passed to the merge algorithm. But,
202 * this data also is stored in merge_result->priv. If a sequence of
203 * merges are being done (such as when cherry-picking or rebasing),
204 * the next merge can look at this and re-use information from
205 * previous merges under certain circumstances.
207 * See also all the cached_* variables.
209 struct tree
*merge_trees
[3];
212 * cached_pairs_valid_side: which side's cached info can be reused
214 * See the description for merge_trees. For repeated merges, at most
215 * only one side's cached information can be used. Valid values:
216 * MERGE_SIDE2: cached data from side2 can be reused
217 * MERGE_SIDE1: cached data from side1 can be reused
218 * 0: no cached data can be reused
219 * -1: See redo_after_renames; both sides can be reused.
221 int cached_pairs_valid_side
;
224 * cached_pairs: Caching of renames and deletions.
226 * These are mappings recording renames and deletions of individual
227 * files (not directories). They are thus a map from an old
228 * filename to either NULL (for deletions) or a new filename (for
231 struct strmap cached_pairs
[3];
234 * cached_target_names: just the destinations from cached_pairs
236 * We sometimes want a fast lookup to determine if a given filename
237 * is one of the destinations in cached_pairs. cached_target_names
238 * is thus duplicative information, but it provides a fast lookup.
240 struct strset cached_target_names
[3];
243 * cached_irrelevant: Caching of rename_sources that aren't relevant.
245 * If we try to detect a rename for a source path and succeed, it's
246 * part of a rename. If we try to detect a rename for a source path
247 * and fail, then it's a delete. If we do not try to detect a rename
248 * for a path, then we don't know if it's a rename or a delete. If
249 * merge-ort doesn't think the path is relevant, then we just won't
250 * cache anything for that path. But there's a slight problem in
251 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
252 * commit 9bd342137e ("diffcore-rename: determine which
253 * relevant_sources are no longer relevant", 2021-03-13),
254 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
255 * avoid excessive calls to diffcore_rename_extended() we still need
256 * to cache such paths, though we cannot record them as either
257 * renames or deletes. So we cache them here as a "turned out to be
258 * irrelevant *for this commit*" as they are often also irrelevant
259 * for subsequent commits, though we will have to do some extra
260 * checking to see whether such paths become relevant for rename
261 * detection when cherry-picking/rebasing subsequent commits.
263 struct strset cached_irrelevant
[3];
266 * redo_after_renames: optimization flag for "restarting" the merge
268 * Sometimes it pays to detect renames, cache them, and then
269 * restart the merge operation from the beginning. The reason for
270 * this is that when we know where all the renames are, we know
271 * whether a certain directory has any paths under it affected --
272 * and if a directory is not affected then it permits us to do
273 * trivial tree merging in more cases. Doing trivial tree merging
274 * prevents the need to run process_entry() on every path
275 * underneath trees that can be trivially merged, and
276 * process_entry() is more expensive than collect_merge_info() --
277 * plus, the second collect_merge_info() will be much faster since
278 * it doesn't have to recurse into the relevant trees.
280 * Values for this flag:
281 * 0 = don't bother, not worth it (or conditions not yet checked)
282 * 1 = conditions for optimization met, optimization worthwhile
283 * 2 = we already did it (don't restart merge yet again)
285 unsigned redo_after_renames
;
288 * needed_limit: value needed for inexact rename detection to run
290 * If the current rename limit wasn't high enough for inexact
291 * rename detection to run, this records the limit needed. Otherwise,
292 * this value remains 0.
297 struct merge_options_internal
{
299 * paths: primary data structure in all of merge ort.
302 * * are full relative paths from the toplevel of the repository
303 * (e.g. "drivers/firmware/raspberrypi.c").
304 * * store all relevant paths in the repo, both directories and
305 * files (e.g. drivers, drivers/firmware would also be included)
306 * * these keys serve to intern all the path strings, which allows
307 * us to do pointer comparison on directory names instead of
308 * strcmp; we just have to be careful to use the interned strings.
310 * The values of paths:
311 * * either a pointer to a merged_info, or a conflict_info struct
312 * * merged_info contains all relevant information for a
313 * non-conflicted entry.
314 * * conflict_info contains a merged_info, plus any additional
315 * information about a conflict such as the higher orders stages
316 * involved and the names of the paths those came from (handy
317 * once renames get involved).
318 * * a path may start "conflicted" (i.e. point to a conflict_info)
319 * and then a later step (e.g. three-way content merge) determines
320 * it can be cleanly merged, at which point it'll be marked clean
321 * and the algorithm will ignore any data outside the contained
322 * merged_info for that entry
323 * * If an entry remains conflicted, the merged_info portion of a
324 * conflict_info will later be filled with whatever version of
325 * the file should be placed in the working directory (e.g. an
326 * as-merged-as-possible variation that contains conflict markers).
331 * conflicted: a subset of keys->values from "paths"
333 * conflicted is basically an optimization between process_entries()
334 * and record_conflicted_index_entries(); the latter could loop over
335 * ALL the entries in paths AGAIN and look for the ones that are
336 * still conflicted, but since process_entries() has to loop over
337 * all of them, it saves the ones it couldn't resolve in this strmap
338 * so that record_conflicted_index_entries() can iterate just the
341 struct strmap conflicted
;
344 * pool: memory pool for fast allocation/deallocation
346 * We allocate room for lots of filenames and auxiliary data
347 * structures in merge_options_internal, and it tends to all be
348 * freed together too. Using a memory pool for these provides a
351 struct mem_pool pool
;
354 * conflicts: logical conflicts and messages stored by _primary_ path
356 * This is a map of pathnames (a subset of the keys in "paths" above)
357 * to struct string_list, with each item's `util` containing a
358 * `struct logical_conflict_info`. Note, though, that for each path,
359 * it only stores the logical conflicts for which that path is the
360 * primary path; the path might be part of additional conflicts.
362 struct strmap conflicts
;
365 * renames: various data relating to rename detection
367 struct rename_info renames
;
370 * attr_index: hacky minimal index used for renormalization
372 * renormalization code _requires_ an index, though it only needs to
373 * find a .gitattributes file within the index. So, when
374 * renormalization is important, we create a special index with just
377 struct index_state attr_index
;
380 * current_dir_name, toplevel_dir: temporary vars
382 * These are used in collect_merge_info_callback(), and will set the
383 * various merged_info.directory_name for the various paths we get;
384 * see documentation for that variable and the requirements placed on
387 const char *current_dir_name
;
388 const char *toplevel_dir
;
390 /* call_depth: recursion level counter for merging merge bases */
393 /* field that holds submodule conflict information */
394 struct string_list conflicted_submodules
;
397 struct conflicted_submodule_item
{
402 static void conflicted_submodule_item_free(void *util
, const char *str UNUSED
)
404 struct conflicted_submodule_item
*item
= util
;
410 struct version_info
{
411 struct object_id oid
;
416 /* if is_null, ignore result. otherwise result has oid & mode */
417 struct version_info result
;
421 * clean: whether the path in question is cleanly merged.
423 * see conflict_info.merged for more details.
428 * basename_offset: offset of basename of path.
430 * perf optimization to avoid recomputing offset of final '/'
431 * character in pathname (0 if no '/' in pathname).
433 size_t basename_offset
;
436 * directory_name: containing directory name.
438 * Note that we assume directory_name is constructed such that
439 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
440 * i.e. string equality is equivalent to pointer equality. For this
441 * to hold, we have to be careful setting directory_name.
443 const char *directory_name
;
446 struct conflict_info
{
448 * merged: the version of the path that will be written to working tree
450 * WARNING: It is critical to check merged.clean and ensure it is 0
451 * before reading any conflict_info fields outside of merged.
452 * Allocated merge_info structs will always have clean set to 1.
453 * Allocated conflict_info structs will have merged.clean set to 0
454 * initially. The merged.clean field is how we know if it is safe
455 * to access other parts of conflict_info besides merged; if a
456 * conflict_info's merged.clean is changed to 1, the rest of the
457 * algorithm is not allowed to look at anything outside of the
458 * merged member anymore.
460 struct merged_info merged
;
462 /* oids & modes from each of the three trees for this path */
463 struct version_info stages
[3];
465 /* pathnames for each stage; may differ due to rename detection */
466 const char *pathnames
[3];
468 /* Whether this path is/was involved in a directory/file conflict */
469 unsigned df_conflict
:1;
472 * Whether this path is/was involved in a non-content conflict other
473 * than a directory/file conflict (e.g. rename/rename, rename/delete,
474 * file location based on possible directory rename).
476 unsigned path_conflict
:1;
479 * For filemask and dirmask, the ith bit corresponds to whether the
480 * ith entry is a file (filemask) or a directory (dirmask). Thus,
481 * filemask & dirmask is always zero, and filemask | dirmask is at
482 * most 7 but can be less when a path does not appear as either a
483 * file or a directory on at least one side of history.
485 * Note that these masks are related to enum merge_side, as the ith
486 * entry corresponds to side i.
488 * These values come from a traverse_trees() call; more info may be
489 * found looking at tree-walk.h's struct traverse_info,
490 * particularly the documentation above the "fn" member (note that
491 * filemask = mask & ~dirmask from that documentation).
497 * Optimization to track which stages match, to avoid the need to
498 * recompute it in multiple steps. Either 0 or at least 2 bits are
499 * set; if at least 2 bits are set, their corresponding stages match.
501 unsigned match_mask
:3;
504 enum conflict_and_info_types
{
505 /* "Simple" conflicts and informational messages */
506 INFO_AUTO_MERGING
= 0,
507 CONFLICT_CONTENTS
, /* text file that failed to merge */
509 CONFLICT_FILE_DIRECTORY
,
510 CONFLICT_DISTINCT_MODES
,
511 CONFLICT_MODIFY_DELETE
,
514 CONFLICT_RENAME_RENAME
, /* same file renamed differently */
515 CONFLICT_RENAME_COLLIDES
, /* rename/add or two files renamed to 1 */
516 CONFLICT_RENAME_DELETE
,
518 /* Basic directory rename */
519 CONFLICT_DIR_RENAME_SUGGESTED
,
520 INFO_DIR_RENAME_APPLIED
,
522 /* Special directory rename cases */
523 INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
,
524 CONFLICT_DIR_RENAME_FILE_IN_WAY
,
525 CONFLICT_DIR_RENAME_COLLISION
,
526 CONFLICT_DIR_RENAME_SPLIT
,
528 /* Basic submodule */
529 INFO_SUBMODULE_FAST_FORWARDING
,
530 CONFLICT_SUBMODULE_FAILED_TO_MERGE
,
532 /* Special submodule cases broken out from FAILED_TO_MERGE */
533 CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
,
534 CONFLICT_SUBMODULE_NOT_INITIALIZED
,
535 CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
,
536 CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
,
537 CONFLICT_SUBMODULE_NULL_MERGE_BASE
,
539 /* Keep this entry _last_ in the list */
544 * Short description of conflict type, relied upon by external tools.
546 * We can add more entries, but DO NOT change any of these strings. Also,
547 * Order MUST match conflict_info_and_types.
549 static const char *type_short_descriptions
[] = {
550 /*** "Simple" conflicts and informational messages ***/
551 [INFO_AUTO_MERGING
] = "Auto-merging",
552 [CONFLICT_CONTENTS
] = "CONFLICT (contents)",
553 [CONFLICT_BINARY
] = "CONFLICT (binary)",
554 [CONFLICT_FILE_DIRECTORY
] = "CONFLICT (file/directory)",
555 [CONFLICT_DISTINCT_MODES
] = "CONFLICT (distinct modes)",
556 [CONFLICT_MODIFY_DELETE
] = "CONFLICT (modify/delete)",
558 /*** Regular rename ***/
559 [CONFLICT_RENAME_RENAME
] = "CONFLICT (rename/rename)",
560 [CONFLICT_RENAME_COLLIDES
] = "CONFLICT (rename involved in collision)",
561 [CONFLICT_RENAME_DELETE
] = "CONFLICT (rename/delete)",
563 /*** Basic directory rename ***/
564 [CONFLICT_DIR_RENAME_SUGGESTED
] =
565 "CONFLICT (directory rename suggested)",
566 [INFO_DIR_RENAME_APPLIED
] = "Path updated due to directory rename",
568 /*** Special directory rename cases ***/
569 [INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
] =
570 "Directory rename skipped since directory was renamed on both sides",
571 [CONFLICT_DIR_RENAME_FILE_IN_WAY
] =
572 "CONFLICT (file in way of directory rename)",
573 [CONFLICT_DIR_RENAME_COLLISION
] = "CONFLICT(directory rename collision)",
574 [CONFLICT_DIR_RENAME_SPLIT
] = "CONFLICT(directory rename unclear split)",
576 /*** Basic submodule ***/
577 [INFO_SUBMODULE_FAST_FORWARDING
] = "Fast forwarding submodule",
578 [CONFLICT_SUBMODULE_FAILED_TO_MERGE
] = "CONFLICT (submodule)",
580 /*** Special submodule cases broken out from FAILED_TO_MERGE ***/
581 [CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
] =
582 "CONFLICT (submodule with possible resolution)",
583 [CONFLICT_SUBMODULE_NOT_INITIALIZED
] =
584 "CONFLICT (submodule not initialized)",
585 [CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
] =
586 "CONFLICT (submodule history not available)",
587 [CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
] =
588 "CONFLICT (submodule may have rewinds)",
589 [CONFLICT_SUBMODULE_NULL_MERGE_BASE
] =
590 "CONFLICT (submodule lacks merge base)"
593 struct logical_conflict_info
{
594 enum conflict_and_info_types type
;
598 /*** Function Grouping: various utility functions ***/
601 * For the next three macros, see warning for conflict_info.merged.
603 * In each of the below, mi is a struct merged_info*, and ci was defined
604 * as a struct conflict_info* (but we need to verify ci isn't actually
605 * pointed at a struct merged_info*).
607 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
608 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
609 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
611 #define INITIALIZE_CI(ci, mi) do { \
612 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
614 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
615 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
616 (ci) = (struct conflict_info *)(mi); \
617 assert((ci) && !(mi)->clean); \
620 static void free_strmap_strings(struct strmap
*map
)
622 struct hashmap_iter iter
;
623 struct strmap_entry
*entry
;
625 strmap_for_each_entry(map
, &iter
, entry
) {
626 free((char*)entry
->key
);
630 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
633 struct rename_info
*renames
= &opti
->renames
;
635 void (*strmap_clear_func
)(struct strmap
*, int) =
636 reinitialize
? strmap_partial_clear
: strmap_clear
;
637 void (*strintmap_clear_func
)(struct strintmap
*) =
638 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
639 void (*strset_clear_func
)(struct strset
*) =
640 reinitialize
? strset_partial_clear
: strset_clear
;
642 strmap_clear_func(&opti
->paths
, 0);
645 * All keys and values in opti->conflicted are a subset of those in
646 * opti->paths. We don't want to deallocate anything twice, so we
647 * don't free the keys and we pass 0 for free_values.
649 strmap_clear_func(&opti
->conflicted
, 0);
651 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
652 discard_index(&opti
->attr_index
);
654 /* Free memory used by various renames maps */
655 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
656 strintmap_clear_func(&renames
->dirs_removed
[i
]);
657 strmap_clear_func(&renames
->dir_renames
[i
], 0);
658 strintmap_clear_func(&renames
->relevant_sources
[i
]);
660 assert(renames
->cached_pairs_valid_side
== 0);
661 if (i
!= renames
->cached_pairs_valid_side
&&
662 -1 != renames
->cached_pairs_valid_side
) {
663 strset_clear_func(&renames
->cached_target_names
[i
]);
664 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
665 strset_clear_func(&renames
->cached_irrelevant
[i
]);
666 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
668 strmap_clear(&renames
->dir_rename_count
[i
], 1);
671 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
672 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
673 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
674 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
676 renames
->cached_pairs_valid_side
= 0;
677 renames
->dir_rename_mask
= 0;
680 struct hashmap_iter iter
;
681 struct strmap_entry
*e
;
683 /* Release and free each strbuf found in output */
684 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
685 struct string_list
*list
= e
->value
;
686 for (int i
= 0; i
< list
->nr
; i
++) {
687 struct logical_conflict_info
*info
=
689 strvec_clear(&info
->paths
);
692 * While strictly speaking we don't need to
693 * free(conflicts) here because we could pass
694 * free_values=1 when calling strmap_clear() on
695 * opti->conflicts, that would require strmap_clear
696 * to do another strmap_for_each_entry() loop, so we
697 * just free it while we're iterating anyway.
699 string_list_clear(list
, 1);
702 strmap_clear(&opti
->conflicts
, 0);
705 mem_pool_discard(&opti
->pool
, 0);
707 string_list_clear_func(&opti
->conflicted_submodules
,
708 conflicted_submodule_item_free
);
710 /* Clean out callback_data as well. */
711 FREE_AND_NULL(renames
->callback_data
);
712 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
715 __attribute__((format (printf
, 2, 3)))
716 static int err(struct merge_options
*opt
, const char *err
, ...)
719 struct strbuf sb
= STRBUF_INIT
;
721 strbuf_addstr(&sb
, "error: ");
722 va_start(params
, err
);
723 strbuf_vaddf(&sb
, err
, params
);
732 static void format_commit(struct strbuf
*sb
,
734 struct repository
*repo
,
735 struct commit
*commit
)
737 struct merge_remote_desc
*desc
;
738 struct pretty_print_context ctx
= {0};
739 ctx
.abbrev
= DEFAULT_ABBREV
;
741 strbuf_addchars(sb
, ' ', indent
);
742 desc
= merge_remote_util(commit
);
744 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
748 repo_format_commit_message(repo
, commit
, "%h %s", sb
, &ctx
);
749 strbuf_addch(sb
, '\n');
752 __attribute__((format (printf
, 8, 9)))
753 static void path_msg(struct merge_options
*opt
,
754 enum conflict_and_info_types type
,
755 int omittable_hint
, /* skippable under --remerge-diff */
756 const char *primary_path
,
757 const char *other_path_1
, /* may be NULL */
758 const char *other_path_2
, /* may be NULL */
759 struct string_list
*other_paths
, /* may be NULL */
760 const char *fmt
, ...)
763 struct string_list
*path_conflicts
;
764 struct logical_conflict_info
*info
;
765 struct strbuf buf
= STRBUF_INIT
;
767 struct strbuf tmp
= STRBUF_INIT
;
770 assert(omittable_hint
==
771 !starts_with(type_short_descriptions
[type
], "CONFLICT") ||
772 type
== CONFLICT_DIR_RENAME_SUGGESTED
);
773 if (opt
->record_conflict_msgs_as_headers
&& omittable_hint
)
774 return; /* Do not record mere hints in headers */
775 if (opt
->priv
->call_depth
&& opt
->verbosity
< 5)
776 return; /* Ignore messages from inner merges */
778 /* Ensure path_conflicts (ptr to array of logical_conflict) allocated */
779 path_conflicts
= strmap_get(&opt
->priv
->conflicts
, primary_path
);
780 if (!path_conflicts
) {
781 path_conflicts
= xmalloc(sizeof(*path_conflicts
));
782 string_list_init_dup(path_conflicts
);
783 strmap_put(&opt
->priv
->conflicts
, primary_path
, path_conflicts
);
786 /* Add a logical_conflict at the end to store info from this call */
787 info
= xcalloc(1, sizeof(*info
));
789 strvec_init(&info
->paths
);
791 /* Handle the list of paths */
792 strvec_push(&info
->paths
, primary_path
);
794 strvec_push(&info
->paths
, other_path_1
);
796 strvec_push(&info
->paths
, other_path_2
);
798 for (int i
= 0; i
< other_paths
->nr
; i
++)
799 strvec_push(&info
->paths
, other_paths
->items
[i
].string
);
801 /* Handle message and its format, in normal case */
802 dest
= (opt
->record_conflict_msgs_as_headers
? &tmp
: &buf
);
805 if (opt
->priv
->call_depth
) {
806 strbuf_addchars(dest
, ' ', 2);
807 strbuf_addstr(dest
, "From inner merge:");
808 strbuf_addchars(dest
, ' ', opt
->priv
->call_depth
* 2);
810 strbuf_vaddf(dest
, fmt
, ap
);
813 /* Handle specialized formatting of message under --remerge-diff */
814 if (opt
->record_conflict_msgs_as_headers
) {
815 int i_sb
= 0, i_tmp
= 0;
817 /* Start with the specified prefix */
818 if (opt
->msg_header_prefix
)
819 strbuf_addf(&buf
, "%s ", opt
->msg_header_prefix
);
821 /* Copy tmp to sb, adding spaces after newlines */
822 strbuf_grow(&buf
, buf
.len
+ 2*tmp
.len
); /* more than sufficient */
823 for (; i_tmp
< tmp
.len
; i_tmp
++, i_sb
++) {
824 /* Copy next character from tmp to sb */
825 buf
.buf
[buf
.len
+ i_sb
] = tmp
.buf
[i_tmp
];
827 /* If we copied a newline, add a space */
828 if (tmp
.buf
[i_tmp
] == '\n')
829 buf
.buf
[++i_sb
] = ' ';
831 /* Update length and ensure it's NUL-terminated */
833 buf
.buf
[buf
.len
] = '\0';
835 strbuf_release(&tmp
);
837 string_list_append_nodup(path_conflicts
, strbuf_detach(&buf
, NULL
))
841 static struct diff_filespec
*pool_alloc_filespec(struct mem_pool
*pool
,
844 /* Similar to alloc_filespec(), but allocate from pool and reuse path */
845 struct diff_filespec
*spec
;
847 spec
= mem_pool_calloc(pool
, 1, sizeof(*spec
));
848 spec
->path
= (char*)path
; /* spec won't modify it */
851 spec
->is_binary
= -1;
855 static struct diff_filepair
*pool_diff_queue(struct mem_pool
*pool
,
856 struct diff_queue_struct
*queue
,
857 struct diff_filespec
*one
,
858 struct diff_filespec
*two
)
860 /* Same code as diff_queue(), except allocate from pool */
861 struct diff_filepair
*dp
;
863 dp
= mem_pool_calloc(pool
, 1, sizeof(*dp
));
871 /* add a string to a strbuf, but converting "/" to "_" */
872 static void add_flattened_path(struct strbuf
*out
, const char *s
)
875 strbuf_addstr(out
, s
);
876 for (; i
< out
->len
; i
++)
877 if (out
->buf
[i
] == '/')
881 static char *unique_path(struct merge_options
*opt
,
886 struct strbuf newpath
= STRBUF_INIT
;
889 struct strmap
*existing_paths
= &opt
->priv
->paths
;
891 strbuf_addf(&newpath
, "%s~", path
);
892 add_flattened_path(&newpath
, branch
);
894 base_len
= newpath
.len
;
895 while (strmap_contains(existing_paths
, newpath
.buf
)) {
896 strbuf_setlen(&newpath
, base_len
);
897 strbuf_addf(&newpath
, "_%d", suffix
++);
900 /* Track the new path in our memory pool */
901 ret
= mem_pool_alloc(&opt
->priv
->pool
, newpath
.len
+ 1);
902 memcpy(ret
, newpath
.buf
, newpath
.len
+ 1);
903 strbuf_release(&newpath
);
907 /*** Function Grouping: functions related to collect_merge_info() ***/
909 static int traverse_trees_wrapper_callback(int n
,
911 unsigned long dirmask
,
912 struct name_entry
*names
,
913 struct traverse_info
*info
)
915 struct merge_options
*opt
= info
->data
;
916 struct rename_info
*renames
= &opt
->priv
->renames
;
917 unsigned filemask
= mask
& ~dirmask
;
921 if (!renames
->callback_data_traverse_path
)
922 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
924 if (filemask
&& filemask
== renames
->dir_rename_mask
)
925 renames
->dir_rename_mask
= 0x07;
927 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
928 renames
->callback_data_alloc
);
929 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
930 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
931 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
933 renames
->callback_data_nr
++;
939 * Much like traverse_trees(), BUT:
940 * - read all the tree entries FIRST, saving them
941 * - note that the above step provides an opportunity to compute necessary
942 * additional details before the "real" traversal
943 * - loop through the saved entries and call the original callback on them
945 static int traverse_trees_wrapper(struct index_state
*istate
,
948 struct traverse_info
*info
)
950 int ret
, i
, old_offset
;
951 traverse_callback_t old_fn
;
952 char *old_callback_data_traverse_path
;
953 struct merge_options
*opt
= info
->data
;
954 struct rename_info
*renames
= &opt
->priv
->renames
;
956 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
958 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
960 old_offset
= renames
->callback_data_nr
;
962 renames
->callback_data_traverse_path
= NULL
;
963 info
->fn
= traverse_trees_wrapper_callback
;
964 ret
= traverse_trees(istate
, n
, t
, info
);
968 info
->traverse_path
= renames
->callback_data_traverse_path
;
970 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
972 renames
->callback_data
[i
].mask
,
973 renames
->callback_data
[i
].dirmask
,
974 renames
->callback_data
[i
].names
,
978 renames
->callback_data_nr
= old_offset
;
979 free(renames
->callback_data_traverse_path
);
980 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
981 info
->traverse_path
= NULL
;
985 static void setup_path_info(struct merge_options
*opt
,
986 struct string_list_item
*result
,
987 const char *current_dir_name
,
988 int current_dir_name_len
,
989 char *fullpath
, /* we'll take over ownership */
990 struct name_entry
*names
,
991 struct name_entry
*merged_version
,
992 unsigned is_null
, /* boolean */
993 unsigned df_conflict
, /* boolean */
996 int resolved
/* boolean */)
998 /* result->util is void*, so mi is a convenience typed variable */
999 struct merged_info
*mi
;
1001 assert(!is_null
|| resolved
);
1002 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
1003 assert(resolved
== (merged_version
!= NULL
));
1005 mi
= mem_pool_calloc(&opt
->priv
->pool
, 1,
1006 resolved
? sizeof(struct merged_info
) :
1007 sizeof(struct conflict_info
));
1008 mi
->directory_name
= current_dir_name
;
1009 mi
->basename_offset
= current_dir_name_len
;
1010 mi
->clean
= !!resolved
;
1012 mi
->result
.mode
= merged_version
->mode
;
1013 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
1014 mi
->is_null
= !!is_null
;
1017 struct conflict_info
*ci
;
1019 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1020 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1021 ci
->pathnames
[i
] = fullpath
;
1022 ci
->stages
[i
].mode
= names
[i
].mode
;
1023 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
1025 ci
->filemask
= filemask
;
1026 ci
->dirmask
= dirmask
;
1027 ci
->df_conflict
= !!df_conflict
;
1030 * Assume is_null for now, but if we have entries
1031 * under the directory then when it is complete in
1032 * write_completed_directory() it'll update this.
1033 * Also, for D/F conflicts, we have to handle the
1034 * directory first, then clear this bit and process
1035 * the file to see how it is handled -- that occurs
1036 * near the top of process_entry().
1040 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
1041 result
->string
= fullpath
;
1045 static void add_pair(struct merge_options
*opt
,
1046 struct name_entry
*names
,
1047 const char *pathname
,
1049 unsigned is_add
/* if false, is_delete */,
1050 unsigned match_mask
,
1051 unsigned dir_rename_mask
)
1053 struct diff_filespec
*one
, *two
;
1054 struct rename_info
*renames
= &opt
->priv
->renames
;
1055 int names_idx
= is_add
? side
: 0;
1058 assert(match_mask
== 0 || match_mask
== 6);
1059 if (strset_contains(&renames
->cached_target_names
[side
],
1063 unsigned content_relevant
= (match_mask
== 0);
1064 unsigned location_relevant
= (dir_rename_mask
== 0x07);
1066 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
1069 * If pathname is found in cached_irrelevant[side] due to
1070 * previous pick but for this commit content is relevant,
1071 * then we need to remove it from cached_irrelevant.
1073 if (content_relevant
)
1074 /* strset_remove is no-op if strset doesn't have key */
1075 strset_remove(&renames
->cached_irrelevant
[side
],
1079 * We do not need to re-detect renames for paths that we already
1080 * know the pairing, i.e. for cached_pairs (or
1081 * cached_irrelevant). However, handle_deferred_entries() needs
1082 * to loop over the union of keys from relevant_sources[side] and
1083 * cached_pairs[side], so for simplicity we set relevant_sources
1084 * for all the cached_pairs too and then strip them back out in
1085 * prune_cached_from_relevant() at the beginning of
1086 * detect_regular_renames().
1088 if (content_relevant
|| location_relevant
) {
1089 /* content_relevant trumps location_relevant */
1090 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
1091 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
1095 * Avoid creating pair if we've already cached rename results.
1096 * Note that we do this after setting relevant_sources[side]
1097 * as noted in the comment above.
1099 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
1100 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
1104 one
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1105 two
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1106 fill_filespec(is_add
? two
: one
,
1107 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
1108 pool_diff_queue(&opt
->priv
->pool
, &renames
->pairs
[side
], one
, two
);
1111 static void collect_rename_info(struct merge_options
*opt
,
1112 struct name_entry
*names
,
1113 const char *dirname
,
1114 const char *fullname
,
1117 unsigned match_mask
)
1119 struct rename_info
*renames
= &opt
->priv
->renames
;
1123 * Update dir_rename_mask (determines ignore-rename-source validity)
1125 * dir_rename_mask helps us keep track of when directory rename
1126 * detection may be relevant. Basically, whenver a directory is
1127 * removed on one side of history, and a file is added to that
1128 * directory on the other side of history, directory rename
1129 * detection is relevant (meaning we have to detect renames for all
1130 * files within that directory to deduce where the directory
1131 * moved). Also, whenever a directory needs directory rename
1132 * detection, due to the "majority rules" choice for where to move
1133 * it (see t6423 testcase 1f), we also need to detect renames for
1134 * all files within subdirectories of that directory as well.
1136 * Here we haven't looked at files within the directory yet, we are
1137 * just looking at the directory itself. So, if we aren't yet in
1138 * a case where a parent directory needed directory rename detection
1139 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
1140 * on one side of history, record the mask of the other side of
1141 * history in dir_rename_mask.
1143 if (renames
->dir_rename_mask
!= 0x07 &&
1144 (dirmask
== 3 || dirmask
== 5)) {
1145 /* simple sanity check */
1146 assert(renames
->dir_rename_mask
== 0 ||
1147 renames
->dir_rename_mask
== (dirmask
& ~1));
1148 /* update dir_rename_mask; have it record mask of new side */
1149 renames
->dir_rename_mask
= (dirmask
& ~1);
1152 /* Update dirs_removed, as needed */
1153 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
1154 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1155 unsigned sides
= (0x07 - dirmask
)/2;
1156 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
1157 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
1159 * Record relevance of this directory. However, note that
1160 * when collect_merge_info_callback() recurses into this
1161 * directory and calls collect_rename_info() on paths
1162 * within that directory, if we find a path that was added
1163 * to this directory on the other side of history, we will
1164 * upgrade this value to RELEVANT_FOR_SELF; see below.
1167 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1170 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1175 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1176 * When we run across a file added to a directory. In such a case,
1177 * find the directory of the file and upgrade its relevance.
1179 if (renames
->dir_rename_mask
== 0x07 &&
1180 (filemask
== 2 || filemask
== 4)) {
1182 * Need directory rename for parent directory on other side
1183 * of history from added file. Thus
1184 * side = (~filemask & 0x06) >> 1
1186 * side = 3 - (filemask/2).
1188 unsigned side
= 3 - (filemask
>> 1);
1189 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1193 if (filemask
== 0 || filemask
== 7)
1196 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1197 unsigned side_mask
= (1 << side
);
1199 /* Check for deletion on side */
1200 if ((filemask
& 1) && !(filemask
& side_mask
))
1201 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1202 match_mask
& filemask
,
1203 renames
->dir_rename_mask
);
1205 /* Check for addition on side */
1206 if (!(filemask
& 1) && (filemask
& side_mask
))
1207 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1208 match_mask
& filemask
,
1209 renames
->dir_rename_mask
);
1213 static int collect_merge_info_callback(int n
,
1215 unsigned long dirmask
,
1216 struct name_entry
*names
,
1217 struct traverse_info
*info
)
1221 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1222 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1223 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1225 struct merge_options
*opt
= info
->data
;
1226 struct merge_options_internal
*opti
= opt
->priv
;
1227 struct rename_info
*renames
= &opt
->priv
->renames
;
1228 struct string_list_item pi
; /* Path Info */
1229 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1230 struct name_entry
*p
;
1233 const char *dirname
= opti
->current_dir_name
;
1234 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1235 unsigned filemask
= mask
& ~dirmask
;
1236 unsigned match_mask
= 0; /* will be updated below */
1237 unsigned mbase_null
= !(mask
& 1);
1238 unsigned side1_null
= !(mask
& 2);
1239 unsigned side2_null
= !(mask
& 4);
1240 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1241 names
[0].mode
== names
[1].mode
&&
1242 oideq(&names
[0].oid
, &names
[1].oid
));
1243 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1244 names
[0].mode
== names
[2].mode
&&
1245 oideq(&names
[0].oid
, &names
[2].oid
));
1246 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1247 names
[1].mode
== names
[2].mode
&&
1248 oideq(&names
[1].oid
, &names
[2].oid
));
1251 * Note: When a path is a file on one side of history and a directory
1252 * in another, we have a directory/file conflict. In such cases, if
1253 * the conflict doesn't resolve from renames and deletions, then we
1254 * always leave directories where they are and move files out of the
1255 * way. Thus, while struct conflict_info has a df_conflict field to
1256 * track such conflicts, we ignore that field for any directories at
1257 * a path and only pay attention to it for files at the given path.
1258 * The fact that we leave directories were they are also means that
1259 * we do not need to worry about getting additional df_conflict
1260 * information propagated from parent directories down to children
1261 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1262 * sets a newinfo.df_conflicts field specifically to propagate it).
1264 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1266 /* n = 3 is a fundamental assumption. */
1268 BUG("Called collect_merge_info_callback wrong");
1271 * A bunch of sanity checks verifying that traverse_trees() calls
1272 * us the way I expect. Could just remove these at some point,
1273 * though maybe they are helpful to future code readers.
1275 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1276 assert(side1_null
== is_null_oid(&names
[1].oid
));
1277 assert(side2_null
== is_null_oid(&names
[2].oid
));
1278 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1279 assert(mask
> 0 && mask
< 8);
1281 /* Determine match_mask */
1282 if (side1_matches_mbase
)
1283 match_mask
= (side2_matches_mbase
? 7 : 3);
1284 else if (side2_matches_mbase
)
1286 else if (sides_match
)
1290 * Get the name of the relevant filepath, which we'll pass to
1291 * setup_path_info() for tracking.
1296 len
= traverse_path_len(info
, p
->pathlen
);
1298 /* +1 in both of the following lines to include the NUL byte */
1299 fullpath
= mem_pool_alloc(&opt
->priv
->pool
, len
+ 1);
1300 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1303 * If mbase, side1, and side2 all match, we can resolve early. Even
1304 * if these are trees, there will be no renames or anything
1307 if (side1_matches_mbase
&& side2_matches_mbase
) {
1308 /* mbase, side1, & side2 all match; use mbase as resolution */
1309 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1310 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1311 filemask
, dirmask
, 1 /* resolved */);
1316 * If the sides match, and all three paths are present and are
1317 * files, then we can take either as the resolution. We can't do
1318 * this with trees, because there may be rename sources from the
1321 if (sides_match
&& filemask
== 0x07) {
1322 /* use side1 (== side2) version as resolution */
1323 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1324 names
, names
+1, side1_null
, 0,
1325 filemask
, dirmask
, 1);
1330 * If side1 matches mbase and all three paths are present and are
1331 * files, then we can use side2 as the resolution. We cannot
1332 * necessarily do so this for trees, because there may be rename
1333 * destinations within side2.
1335 if (side1_matches_mbase
&& filemask
== 0x07) {
1336 /* use side2 version as resolution */
1337 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1338 names
, names
+2, side2_null
, 0,
1339 filemask
, dirmask
, 1);
1343 /* Similar to above but swapping sides 1 and 2 */
1344 if (side2_matches_mbase
&& filemask
== 0x07) {
1345 /* use side1 version as resolution */
1346 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1347 names
, names
+1, side1_null
, 0,
1348 filemask
, dirmask
, 1);
1353 * Sometimes we can tell that a source path need not be included in
1354 * rename detection -- namely, whenever either
1355 * side1_matches_mbase && side2_null
1357 * side2_matches_mbase && side1_null
1358 * However, we call collect_rename_info() even in those cases,
1359 * because exact renames are cheap and would let us remove both a
1360 * source and destination path. We'll cull the unneeded sources
1363 collect_rename_info(opt
, names
, dirname
, fullpath
,
1364 filemask
, dirmask
, match_mask
);
1367 * None of the special cases above matched, so we have a
1368 * provisional conflict. (Rename detection might allow us to
1369 * unconflict some more cases, but that comes later so all we can
1370 * do now is record the different non-null file hashes.)
1372 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1373 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1377 ci
->match_mask
= match_mask
;
1379 /* If dirmask, recurse into subdirectories */
1381 struct traverse_info newinfo
;
1382 struct tree_desc t
[3];
1383 void *buf
[3] = {NULL
, NULL
, NULL
};
1384 const char *original_dir_name
;
1388 * Check for whether we can avoid recursing due to one side
1389 * matching the merge base. The side that does NOT match is
1390 * the one that might have a rename destination we need.
1392 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1393 side
= side1_matches_mbase
? MERGE_SIDE2
:
1394 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1395 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1397 * Also defer recursing into new directories; set up a
1398 * few variables to let us do so.
1400 ci
->match_mask
= (7 - dirmask
);
1403 if (renames
->dir_rename_mask
!= 0x07 &&
1404 side
!= MERGE_BASE
&&
1405 renames
->deferred
[side
].trivial_merges_okay
&&
1406 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1408 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1409 pi
.string
, renames
->dir_rename_mask
);
1410 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1414 /* We need to recurse */
1415 ci
->match_mask
&= filemask
;
1417 newinfo
.prev
= info
;
1418 newinfo
.name
= p
->path
;
1419 newinfo
.namelen
= p
->pathlen
;
1420 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1422 * If this directory we are about to recurse into cared about
1423 * its parent directory (the current directory) having a D/F
1424 * conflict, then we'd propagate the masks in this way:
1425 * newinfo.df_conflicts |= (mask & ~dirmask);
1426 * But we don't worry about propagating D/F conflicts. (See
1427 * comment near setting of local df_conflict variable near
1428 * the beginning of this function).
1431 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1432 if (i
== 1 && side1_matches_mbase
)
1434 else if (i
== 2 && side2_matches_mbase
)
1436 else if (i
== 2 && sides_match
)
1439 const struct object_id
*oid
= NULL
;
1441 oid
= &names
[i
].oid
;
1442 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1448 original_dir_name
= opti
->current_dir_name
;
1449 opti
->current_dir_name
= pi
.string
;
1450 if (renames
->dir_rename_mask
== 0 ||
1451 renames
->dir_rename_mask
== 0x07)
1452 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1454 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1455 opti
->current_dir_name
= original_dir_name
;
1456 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1458 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1468 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1471 assert((side
== 1 && ci
->match_mask
== 5) ||
1472 (side
== 2 && ci
->match_mask
== 3));
1473 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1474 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1475 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1477 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1480 static int handle_deferred_entries(struct merge_options
*opt
,
1481 struct traverse_info
*info
)
1483 struct rename_info
*renames
= &opt
->priv
->renames
;
1484 struct hashmap_iter iter
;
1485 struct strmap_entry
*entry
;
1487 int path_count_before
, path_count_after
= 0;
1489 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1490 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1491 unsigned optimization_okay
= 1;
1492 struct strintmap copy
;
1494 /* Loop over the set of paths we need to know rename info for */
1495 strset_for_each_entry(&renames
->relevant_sources
[side
],
1497 char *rename_target
, *dir
, *dir_marker
;
1498 struct strmap_entry
*e
;
1501 * If we don't know delete/rename info for this path,
1502 * then we need to recurse into all trees to get all
1503 * adds to make sure we have it.
1505 if (strset_contains(&renames
->cached_irrelevant
[side
],
1508 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1511 optimization_okay
= 0;
1515 /* If this is a delete, we have enough info already */
1516 rename_target
= e
->value
;
1520 /* If we already walked the rename target, we're good */
1521 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1525 * Otherwise, we need to get a list of directories that
1526 * will need to be recursed into to get this
1529 dir
= xstrdup(rename_target
);
1530 while ((dir_marker
= strrchr(dir
, '/'))) {
1532 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1535 strset_add(&renames
->deferred
[side
].target_dirs
,
1540 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1542 * We need to recurse into any directories in
1543 * possible_trivial_merges[side] found in target_dirs[side].
1544 * But when we recurse, we may need to queue up some of the
1545 * subdirectories for possible_trivial_merges[side]. Since
1546 * we can't safely iterate through a hashmap while also adding
1547 * entries, move the entries into 'copy', iterate over 'copy',
1548 * and then we'll also iterate anything added into
1549 * possible_trivial_merges[side] once this loop is done.
1551 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1552 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1556 strintmap_for_each_entry(©
, &iter
, entry
) {
1557 const char *path
= entry
->key
;
1558 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1559 struct conflict_info
*ci
;
1561 struct tree_desc t
[3];
1562 void *buf
[3] = {NULL
,};
1565 ci
= strmap_get(&opt
->priv
->paths
, path
);
1567 dirmask
= ci
->dirmask
;
1569 if (optimization_okay
&&
1570 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1572 resolve_trivial_directory_merge(ci
, side
);
1577 info
->namelen
= strlen(path
);
1578 info
->pathlen
= info
->namelen
+ 1;
1580 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1581 if (i
== 1 && ci
->match_mask
== 3)
1583 else if (i
== 2 && ci
->match_mask
== 5)
1585 else if (i
== 2 && ci
->match_mask
== 6)
1588 const struct object_id
*oid
= NULL
;
1590 oid
= &ci
->stages
[i
].oid
;
1591 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1596 ci
->match_mask
&= ci
->filemask
;
1597 opt
->priv
->current_dir_name
= path
;
1598 renames
->dir_rename_mask
= dir_rename_mask
;
1599 if (renames
->dir_rename_mask
== 0 ||
1600 renames
->dir_rename_mask
== 0x07)
1601 ret
= traverse_trees(NULL
, 3, t
, info
);
1603 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1605 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1611 strintmap_clear(©
);
1612 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1614 const char *path
= entry
->key
;
1615 struct conflict_info
*ci
;
1617 ci
= strmap_get(&opt
->priv
->paths
, path
);
1620 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1621 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1623 resolve_trivial_directory_merge(ci
, side
);
1625 if (!optimization_okay
|| path_count_after
)
1626 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1628 if (path_count_after
) {
1630 * The choice of wanted_factor here does not affect
1631 * correctness, only performance. When the
1632 * path_count_after / path_count_before
1633 * ratio is high, redoing after renames is a big
1634 * performance boost. I suspect that redoing is a wash
1635 * somewhere near a value of 2, and below that redoing will
1636 * slow things down. I applied a fudge factor and picked
1637 * 3; see the commit message when this was introduced for
1638 * back of the envelope calculations for this ratio.
1640 const int wanted_factor
= 3;
1642 /* We should only redo collect_merge_info one time */
1643 assert(renames
->redo_after_renames
== 0);
1645 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1646 renames
->redo_after_renames
= 1;
1647 renames
->cached_pairs_valid_side
= -1;
1649 } else if (renames
->redo_after_renames
== 2)
1650 renames
->redo_after_renames
= 0;
1654 static int collect_merge_info(struct merge_options
*opt
,
1655 struct tree
*merge_base
,
1660 struct tree_desc t
[3];
1661 struct traverse_info info
;
1663 opt
->priv
->toplevel_dir
= "";
1664 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1665 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1666 info
.fn
= collect_merge_info_callback
;
1668 info
.show_all_errors
= 1;
1670 parse_tree(merge_base
);
1673 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1674 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1675 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1677 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1678 ret
= traverse_trees(NULL
, 3, t
, &info
);
1680 ret
= handle_deferred_entries(opt
, &info
);
1681 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1686 /*** Function Grouping: functions related to threeway content merges ***/
1688 static int find_first_merges(struct repository
*repo
,
1692 struct object_array
*result
)
1695 struct object_array merges
= OBJECT_ARRAY_INIT
;
1696 struct commit
*commit
;
1697 int contains_another
;
1699 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1700 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1701 "--all", merged_revision
, NULL
};
1702 struct rev_info revs
;
1703 struct setup_revision_opt rev_opts
;
1705 memset(result
, 0, sizeof(struct object_array
));
1706 memset(&rev_opts
, 0, sizeof(rev_opts
));
1708 /* get all revisions that merge commit a */
1709 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1710 oid_to_hex(&a
->object
.oid
));
1711 repo_init_revisions(repo
, &revs
, NULL
);
1712 /* FIXME: can't handle linked worktrees in submodules yet */
1713 revs
.single_worktree
= path
!= NULL
;
1714 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1716 /* save all revisions from the above list that contain b */
1717 if (prepare_revision_walk(&revs
))
1718 die("revision walk setup failed");
1719 while ((commit
= get_revision(&revs
)) != NULL
) {
1720 struct object
*o
= &(commit
->object
);
1721 if (repo_in_merge_bases(repo
, b
, commit
))
1722 add_object_array(o
, NULL
, &merges
);
1724 reset_revision_walk();
1726 /* Now we've got all merges that contain a and b. Prune all
1727 * merges that contain another found merge and save them in
1730 for (i
= 0; i
< merges
.nr
; i
++) {
1731 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1733 contains_another
= 0;
1734 for (j
= 0; j
< merges
.nr
; j
++) {
1735 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1736 if (i
!= j
&& repo_in_merge_bases(repo
, m2
, m1
)) {
1737 contains_another
= 1;
1742 if (!contains_another
)
1743 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1746 object_array_clear(&merges
);
1747 release_revisions(&revs
);
1751 static int merge_submodule(struct merge_options
*opt
,
1753 const struct object_id
*o
,
1754 const struct object_id
*a
,
1755 const struct object_id
*b
,
1756 struct object_id
*result
)
1758 struct repository subrepo
;
1759 struct strbuf sb
= STRBUF_INIT
;
1761 struct commit
*commit_o
, *commit_a
, *commit_b
;
1763 struct object_array merges
;
1766 int search
= !opt
->priv
->call_depth
;
1767 int sub_not_initialized
= 1;
1768 int sub_flag
= CONFLICT_SUBMODULE_FAILED_TO_MERGE
;
1770 /* store fallback answer in result in case we fail */
1771 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1773 /* we can not handle deletion conflicts */
1774 if (is_null_oid(a
) || is_null_oid(b
))
1775 BUG("submodule deleted on one side; this should be handled outside of merge_submodule()");
1777 if ((sub_not_initialized
= repo_submodule_init(&subrepo
,
1778 opt
->repo
, path
, null_oid()))) {
1779 path_msg(opt
, CONFLICT_SUBMODULE_NOT_INITIALIZED
, 0,
1780 path
, NULL
, NULL
, NULL
,
1781 _("Failed to merge submodule %s (not checked out)"),
1783 sub_flag
= CONFLICT_SUBMODULE_NOT_INITIALIZED
;
1787 if (is_null_oid(o
)) {
1788 path_msg(opt
, CONFLICT_SUBMODULE_NULL_MERGE_BASE
, 0,
1789 path
, NULL
, NULL
, NULL
,
1790 _("Failed to merge submodule %s (no merge base)"),
1795 if (!(commit_o
= lookup_commit_reference(&subrepo
, o
)) ||
1796 !(commit_a
= lookup_commit_reference(&subrepo
, a
)) ||
1797 !(commit_b
= lookup_commit_reference(&subrepo
, b
))) {
1798 path_msg(opt
, CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
, 0,
1799 path
, NULL
, NULL
, NULL
,
1800 _("Failed to merge submodule %s (commits not present)"),
1802 sub_flag
= CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
;
1806 /* check whether both changes are forward */
1807 if (!repo_in_merge_bases(&subrepo
, commit_o
, commit_a
) ||
1808 !repo_in_merge_bases(&subrepo
, commit_o
, commit_b
)) {
1809 path_msg(opt
, CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
, 0,
1810 path
, NULL
, NULL
, NULL
,
1811 _("Failed to merge submodule %s "
1812 "(commits don't follow merge-base)"),
1817 /* Case #1: a is contained in b or vice versa */
1818 if (repo_in_merge_bases(&subrepo
, commit_a
, commit_b
)) {
1820 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1821 path
, NULL
, NULL
, NULL
,
1822 _("Note: Fast-forwarding submodule %s to %s"),
1823 path
, oid_to_hex(b
));
1827 if (repo_in_merge_bases(&subrepo
, commit_b
, commit_a
)) {
1829 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1830 path
, NULL
, NULL
, NULL
,
1831 _("Note: Fast-forwarding submodule %s to %s"),
1832 path
, oid_to_hex(a
));
1838 * Case #2: There are one or more merges that contain a and b in
1839 * the submodule. If there is only one, then present it as a
1840 * suggestion to the user, but leave it marked unmerged so the
1841 * user needs to confirm the resolution.
1844 /* Skip the search if makes no sense to the calling context. */
1848 /* find commit which merges them */
1849 parent_count
= find_first_merges(&subrepo
, path
, commit_a
, commit_b
,
1851 switch (parent_count
) {
1853 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE
, 0,
1854 path
, NULL
, NULL
, NULL
,
1855 _("Failed to merge submodule %s"), path
);
1859 format_commit(&sb
, 4, &subrepo
,
1860 (struct commit
*)merges
.objects
[0].item
);
1861 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1862 path
, NULL
, NULL
, NULL
,
1863 _("Failed to merge submodule %s, but a possible merge "
1864 "resolution exists: %s"),
1866 strbuf_release(&sb
);
1869 for (i
= 0; i
< merges
.nr
; i
++)
1870 format_commit(&sb
, 4, &subrepo
,
1871 (struct commit
*)merges
.objects
[i
].item
);
1872 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1873 path
, NULL
, NULL
, NULL
,
1874 _("Failed to merge submodule %s, but multiple "
1875 "possible merges exist:\n%s"), path
, sb
.buf
);
1876 strbuf_release(&sb
);
1879 object_array_clear(&merges
);
1881 if (!opt
->priv
->call_depth
&& !ret
) {
1882 struct string_list
*csub
= &opt
->priv
->conflicted_submodules
;
1883 struct conflicted_submodule_item
*util
;
1886 util
= xmalloc(sizeof(*util
));
1887 util
->flag
= sub_flag
;
1888 util
->abbrev
= NULL
;
1889 if (!sub_not_initialized
) {
1890 abbrev
= repo_find_unique_abbrev(&subrepo
, b
, DEFAULT_ABBREV
);
1891 util
->abbrev
= xstrdup(abbrev
);
1893 string_list_append(csub
, path
)->util
= util
;
1896 if (!sub_not_initialized
)
1897 repo_clear(&subrepo
);
1901 static void initialize_attr_index(struct merge_options
*opt
)
1904 * The renormalize_buffer() functions require attributes, and
1905 * annoyingly those can only be read from the working tree or from
1906 * an index_state. merge-ort doesn't have an index_state, so we
1907 * generate a fake one containing only attribute information.
1909 struct merged_info
*mi
;
1910 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1911 struct cache_entry
*ce
;
1913 attr_index
->initialized
= 1;
1915 if (!opt
->renormalize
)
1918 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1923 int len
= strlen(GITATTRIBUTES_FILE
);
1924 ce
= make_empty_cache_entry(attr_index
, len
);
1925 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1926 ce
->ce_flags
= create_ce_flags(0);
1927 ce
->ce_namelen
= len
;
1928 oidcpy(&ce
->oid
, &mi
->result
.oid
);
1929 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1930 add_index_entry(attr_index
, ce
,
1931 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1932 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
, &ce
->oid
);
1935 struct conflict_info
*ci
;
1937 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1938 for (stage
= 0; stage
< 3; stage
++) {
1939 unsigned stage_mask
= (1 << stage
);
1941 if (!(ci
->filemask
& stage_mask
))
1943 len
= strlen(GITATTRIBUTES_FILE
);
1944 ce
= make_empty_cache_entry(attr_index
, len
);
1945 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1946 ce
->ce_flags
= create_ce_flags(stage
);
1947 ce
->ce_namelen
= len
;
1948 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1949 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1950 add_index_entry(attr_index
, ce
,
1951 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1952 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1958 static int merge_3way(struct merge_options
*opt
,
1960 const struct object_id
*o
,
1961 const struct object_id
*a
,
1962 const struct object_id
*b
,
1963 const char *pathnames
[3],
1964 const int extra_marker_size
,
1965 mmbuffer_t
*result_buf
)
1967 mmfile_t orig
, src1
, src2
;
1968 struct ll_merge_options ll_opts
= {0};
1969 char *base
, *name1
, *name2
;
1970 enum ll_merge_result merge_status
;
1972 if (!opt
->priv
->attr_index
.initialized
)
1973 initialize_attr_index(opt
);
1975 ll_opts
.renormalize
= opt
->renormalize
;
1976 ll_opts
.extra_marker_size
= extra_marker_size
;
1977 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1979 if (opt
->priv
->call_depth
) {
1980 ll_opts
.virtual_ancestor
= 1;
1981 ll_opts
.variant
= 0;
1983 switch (opt
->recursive_variant
) {
1984 case MERGE_VARIANT_OURS
:
1985 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1987 case MERGE_VARIANT_THEIRS
:
1988 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1991 ll_opts
.variant
= 0;
1996 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1997 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
1998 base
= mkpathdup("%s", opt
->ancestor
);
1999 name1
= mkpathdup("%s", opt
->branch1
);
2000 name2
= mkpathdup("%s", opt
->branch2
);
2002 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
2003 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
2004 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
2007 read_mmblob(&orig
, o
);
2008 read_mmblob(&src1
, a
);
2009 read_mmblob(&src2
, b
);
2011 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
2012 &src1
, name1
, &src2
, name2
,
2013 &opt
->priv
->attr_index
, &ll_opts
);
2014 if (merge_status
== LL_MERGE_BINARY_CONFLICT
)
2015 path_msg(opt
, CONFLICT_BINARY
, 0,
2016 path
, NULL
, NULL
, NULL
,
2017 "warning: Cannot merge binary files: %s (%s vs. %s)",
2018 path
, name1
, name2
);
2026 return merge_status
;
2029 static int handle_content_merge(struct merge_options
*opt
,
2031 const struct version_info
*o
,
2032 const struct version_info
*a
,
2033 const struct version_info
*b
,
2034 const char *pathnames
[3],
2035 const int extra_marker_size
,
2036 struct version_info
*result
)
2039 * path is the target location where we want to put the file, and
2040 * is used to determine any normalization rules in ll_merge.
2042 * The normal case is that path and all entries in pathnames are
2043 * identical, though renames can affect which path we got one of
2044 * the three blobs to merge on various sides of history.
2046 * extra_marker_size is the amount to extend conflict markers in
2047 * ll_merge; this is neeed if we have content merges of content
2048 * merges, which happens for example with rename/rename(2to1) and
2049 * rename/add conflicts.
2054 * handle_content_merge() needs both files to be of the same type, i.e.
2055 * both files OR both submodules OR both symlinks. Conflicting types
2056 * needs to be handled elsewhere.
2058 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
2061 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
2062 result
->mode
= b
->mode
;
2064 /* must be the 100644/100755 case */
2065 assert(S_ISREG(a
->mode
));
2066 result
->mode
= a
->mode
;
2067 clean
= (b
->mode
== o
->mode
);
2069 * FIXME: If opt->priv->call_depth && !clean, then we really
2070 * should not make result->mode match either a->mode or
2071 * b->mode; that causes t6036 "check conflicting mode for
2072 * regular file" to fail. It would be best to use some other
2073 * mode, but we'll confuse all kinds of stuff if we use one
2074 * where S_ISREG(result->mode) isn't true, and if we use
2075 * something like 0100666, then tree-walk.c's calls to
2076 * canon_mode() will just normalize that to 100644 for us and
2077 * thus not solve anything.
2079 * Figure out if there's some kind of way we can work around
2085 * Trivial oid merge.
2087 * Note: While one might assume that the next four lines would
2088 * be unnecessary due to the fact that match_mask is often
2089 * setup and already handled, renames don't always take care
2092 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
2093 oidcpy(&result
->oid
, &b
->oid
);
2094 else if (oideq(&b
->oid
, &o
->oid
))
2095 oidcpy(&result
->oid
, &a
->oid
);
2097 /* Remaining rules depend on file vs. submodule vs. symlink. */
2098 else if (S_ISREG(a
->mode
)) {
2099 mmbuffer_t result_buf
;
2100 int ret
= 0, merge_status
;
2104 * If 'o' is different type, treat it as null so we do a
2107 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2109 merge_status
= merge_3way(opt
, path
,
2110 two_way
? null_oid() : &o
->oid
,
2112 pathnames
, extra_marker_size
,
2115 if ((merge_status
< 0) || !result_buf
.ptr
)
2116 ret
= err(opt
, _("Failed to execute internal merge"));
2119 write_object_file(result_buf
.ptr
, result_buf
.size
,
2120 OBJ_BLOB
, &result
->oid
))
2121 ret
= err(opt
, _("Unable to add %s to database"),
2124 free(result_buf
.ptr
);
2127 clean
&= (merge_status
== 0);
2128 path_msg(opt
, INFO_AUTO_MERGING
, 1, path
, NULL
, NULL
, NULL
,
2129 _("Auto-merging %s"), path
);
2130 } else if (S_ISGITLINK(a
->mode
)) {
2131 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2132 clean
= merge_submodule(opt
, pathnames
[0],
2133 two_way
? null_oid() : &o
->oid
,
2134 &a
->oid
, &b
->oid
, &result
->oid
);
2135 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
2136 result
->mode
= o
->mode
;
2137 oidcpy(&result
->oid
, &o
->oid
);
2139 } else if (S_ISLNK(a
->mode
)) {
2140 if (opt
->priv
->call_depth
) {
2142 result
->mode
= o
->mode
;
2143 oidcpy(&result
->oid
, &o
->oid
);
2145 switch (opt
->recursive_variant
) {
2146 case MERGE_VARIANT_NORMAL
:
2148 oidcpy(&result
->oid
, &a
->oid
);
2150 case MERGE_VARIANT_OURS
:
2151 oidcpy(&result
->oid
, &a
->oid
);
2153 case MERGE_VARIANT_THEIRS
:
2154 oidcpy(&result
->oid
, &b
->oid
);
2159 BUG("unsupported object type in the tree: %06o for %s",
2165 /*** Function Grouping: functions related to detect_and_process_renames(), ***
2166 *** which are split into directory and regular rename detection sections. ***/
2168 /*** Function Grouping: functions related to directory rename detection ***/
2170 struct collision_info
{
2171 struct string_list source_files
;
2172 unsigned reported_already
:1;
2176 * Return a new string that replaces the beginning portion (which matches
2177 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
2178 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
2180 * Caller must ensure that old_path starts with rename_info->key + '/'.
2182 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
2183 const char *old_path
)
2185 struct strbuf new_path
= STRBUF_INIT
;
2186 const char *old_dir
= rename_info
->key
;
2187 const char *new_dir
= rename_info
->value
;
2188 int oldlen
, newlen
, new_dir_len
;
2190 oldlen
= strlen(old_dir
);
2191 if (*new_dir
== '\0')
2193 * If someone renamed/merged a subdirectory into the root
2194 * directory (e.g. 'some/subdir' -> ''), then we want to
2197 * as the rename; we need to make old_path + oldlen advance
2198 * past the '/' character.
2201 new_dir_len
= strlen(new_dir
);
2202 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
2203 strbuf_grow(&new_path
, newlen
);
2204 strbuf_add(&new_path
, new_dir
, new_dir_len
);
2205 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2207 return strbuf_detach(&new_path
, NULL
);
2210 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2212 struct merged_info
*mi
= strmap_get(paths
, path
);
2213 struct conflict_info
*ci
;
2216 INITIALIZE_CI(ci
, mi
);
2217 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2221 * See if there is a directory rename for path, and if there are any file
2222 * level conflicts on the given side for the renamed location. If there is
2223 * a rename and there are no conflicts, return the new name. Otherwise,
2226 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2228 unsigned side_index
,
2229 struct strmap_entry
*rename_info
,
2230 struct strmap
*collisions
)
2232 char *new_path
= NULL
;
2233 struct collision_info
*c_info
;
2235 struct strbuf collision_paths
= STRBUF_INIT
;
2238 * entry has the mapping of old directory name to new directory name
2239 * that we want to apply to path.
2241 new_path
= apply_dir_rename(rename_info
, path
);
2243 BUG("Failed to apply directory rename!");
2246 * The caller needs to have ensured that it has pre-populated
2247 * collisions with all paths that map to new_path. Do a quick check
2248 * to ensure that's the case.
2250 c_info
= strmap_get(collisions
, new_path
);
2252 BUG("c_info is NULL");
2255 * Check for one-sided add/add/.../add conflicts, i.e.
2256 * where implicit renames from the other side doing
2257 * directory rename(s) can affect this side of history
2258 * to put multiple paths into the same location. Warn
2259 * and bail on directory renames for such paths.
2261 if (c_info
->reported_already
) {
2263 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2264 c_info
->reported_already
= 1;
2265 strbuf_add_separated_string_list(&collision_paths
, ", ",
2266 &c_info
->source_files
);
2267 path_msg(opt
, CONFLICT_DIR_RENAME_FILE_IN_WAY
, 0,
2268 new_path
, NULL
, NULL
, &c_info
->source_files
,
2269 _("CONFLICT (implicit dir rename): Existing "
2270 "file/dir at %s in the way of implicit "
2271 "directory rename(s) putting the following "
2272 "path(s) there: %s."),
2273 new_path
, collision_paths
.buf
);
2275 } else if (c_info
->source_files
.nr
> 1) {
2276 c_info
->reported_already
= 1;
2277 strbuf_add_separated_string_list(&collision_paths
, ", ",
2278 &c_info
->source_files
);
2279 path_msg(opt
, CONFLICT_DIR_RENAME_COLLISION
, 0,
2280 new_path
, NULL
, NULL
, &c_info
->source_files
,
2281 _("CONFLICT (implicit dir rename): Cannot map "
2282 "more than one path to %s; implicit directory "
2283 "renames tried to put these paths there: %s"),
2284 new_path
, collision_paths
.buf
);
2288 /* Free memory we no longer need */
2289 strbuf_release(&collision_paths
);
2290 if (!clean
&& new_path
) {
2298 static void get_provisional_directory_renames(struct merge_options
*opt
,
2302 struct hashmap_iter iter
;
2303 struct strmap_entry
*entry
;
2304 struct rename_info
*renames
= &opt
->priv
->renames
;
2308 * dir_rename_count: old_directory -> {new_directory -> count}
2310 * dir_renames: old_directory -> best_new_directory
2311 * where best_new_directory is the one with the unique highest count.
2313 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2314 const char *source_dir
= entry
->key
;
2315 struct strintmap
*counts
= entry
->value
;
2316 struct hashmap_iter count_iter
;
2317 struct strmap_entry
*count_entry
;
2320 const char *best
= NULL
;
2322 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2323 const char *target_dir
= count_entry
->key
;
2324 intptr_t count
= (intptr_t)count_entry
->value
;
2328 else if (count
> max
) {
2337 if (bad_max
== max
) {
2338 path_msg(opt
, CONFLICT_DIR_RENAME_SPLIT
, 0,
2339 source_dir
, NULL
, NULL
, NULL
,
2340 _("CONFLICT (directory rename split): "
2341 "Unclear where to rename %s to; it was "
2342 "renamed to multiple other directories, "
2343 "with no destination getting a majority of "
2348 strmap_put(&renames
->dir_renames
[side
],
2349 source_dir
, (void*)best
);
2354 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2356 struct hashmap_iter iter
;
2357 struct strmap_entry
*entry
;
2358 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2359 struct rename_info
*renames
= &opt
->priv
->renames
;
2360 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2361 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2364 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2365 if (strmap_contains(side2_dir_renames
, entry
->key
))
2366 string_list_append(&duplicated
, entry
->key
);
2369 for (i
= 0; i
< duplicated
.nr
; i
++) {
2370 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2371 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2373 string_list_clear(&duplicated
, 0);
2376 static struct strmap_entry
*check_dir_renamed(const char *path
,
2377 struct strmap
*dir_renames
)
2379 char *temp
= xstrdup(path
);
2381 struct strmap_entry
*e
= NULL
;
2383 while ((end
= strrchr(temp
, '/'))) {
2385 e
= strmap_get_entry(dir_renames
, temp
);
2393 static void compute_collisions(struct strmap
*collisions
,
2394 struct strmap
*dir_renames
,
2395 struct diff_queue_struct
*pairs
)
2399 strmap_init_with_options(collisions
, NULL
, 0);
2400 if (strmap_empty(dir_renames
))
2404 * Multiple files can be mapped to the same path due to directory
2405 * renames done by the other side of history. Since that other
2406 * side of history could have merged multiple directories into one,
2407 * if our side of history added the same file basename to each of
2408 * those directories, then all N of them would get implicitly
2409 * renamed by the directory rename detection into the same path,
2410 * and we'd get an add/add/.../add conflict, and all those adds
2411 * from *this* side of history. This is not representable in the
2412 * index, and users aren't going to easily be able to make sense of
2413 * it. So we need to provide a good warning about what's
2414 * happening, and fall back to no-directory-rename detection
2415 * behavior for those paths.
2417 * See testcases 9e and all of section 5 from t6043 for examples.
2419 for (i
= 0; i
< pairs
->nr
; ++i
) {
2420 struct strmap_entry
*rename_info
;
2421 struct collision_info
*collision_info
;
2423 struct diff_filepair
*pair
= pairs
->queue
[i
];
2425 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2427 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2431 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2433 collision_info
= strmap_get(collisions
, new_path
);
2434 if (collision_info
) {
2437 CALLOC_ARRAY(collision_info
, 1);
2438 string_list_init_nodup(&collision_info
->source_files
);
2439 strmap_put(collisions
, new_path
, collision_info
);
2441 string_list_insert(&collision_info
->source_files
,
2446 static void free_collisions(struct strmap
*collisions
)
2448 struct hashmap_iter iter
;
2449 struct strmap_entry
*entry
;
2451 /* Free each value in the collisions map */
2452 strmap_for_each_entry(collisions
, &iter
, entry
) {
2453 struct collision_info
*info
= entry
->value
;
2454 string_list_clear(&info
->source_files
, 0);
2457 * In compute_collisions(), we set collisions.strdup_strings to 0
2458 * so that we wouldn't have to make another copy of the new_path
2459 * allocated by apply_dir_rename(). But now that we've used them
2460 * and have no other references to these strings, it is time to
2463 free_strmap_strings(collisions
);
2464 strmap_clear(collisions
, 1);
2467 static char *check_for_directory_rename(struct merge_options
*opt
,
2469 unsigned side_index
,
2470 struct strmap
*dir_renames
,
2471 struct strmap
*dir_rename_exclusions
,
2472 struct strmap
*collisions
,
2476 struct strmap_entry
*rename_info
;
2477 struct strmap_entry
*otherinfo
;
2478 const char *new_dir
;
2479 int other_side
= 3 - side_index
;
2482 * Cases where we don't have or don't want a directory rename for
2485 if (strmap_empty(dir_renames
))
2487 if (strmap_get(&collisions
[other_side
], path
))
2489 rename_info
= check_dir_renamed(path
, dir_renames
);
2494 * This next part is a little weird. We do not want to do an
2495 * implicit rename into a directory we renamed on our side, because
2496 * that will result in a spurious rename/rename(1to2) conflict. An
2498 * Base commit: dumbdir/afile, otherdir/bfile
2499 * Side 1: smrtdir/afile, otherdir/bfile
2500 * Side 2: dumbdir/afile, dumbdir/bfile
2501 * Here, while working on Side 1, we could notice that otherdir was
2502 * renamed/merged to dumbdir, and change the diff_filepair for
2503 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2504 * 2 will notice the rename from dumbdir to smrtdir, and do the
2505 * transitive rename to move it from dumbdir/bfile to
2506 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2507 * smrtdir, a rename/rename(1to2) conflict. We really just want
2508 * the file to end up in smrtdir. And the way to achieve that is
2509 * to not let Side1 do the rename to dumbdir, since we know that is
2510 * the source of one of our directory renames.
2512 * That's why otherinfo and dir_rename_exclusions is here.
2514 * As it turns out, this also prevents N-way transient rename
2515 * confusion; See testcases 9c and 9d of t6043.
2517 new_dir
= rename_info
->value
; /* old_dir = rename_info->key; */
2518 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2520 path_msg(opt
, INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
, 1,
2521 rename_info
->key
, path
, new_dir
, NULL
,
2522 _("WARNING: Avoiding applying %s -> %s rename "
2523 "to %s, because %s itself was renamed."),
2524 rename_info
->key
, new_dir
, path
, new_dir
);
2528 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2530 &collisions
[side_index
]);
2531 *clean_merge
&= (new_path
!= NULL
);
2536 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2537 struct diff_filepair
*pair
,
2541 * The basic idea is to get the conflict_info from opt->priv->paths
2542 * at old path, and insert it into new_path; basically just this:
2543 * ci = strmap_get(&opt->priv->paths, old_path);
2544 * strmap_remove(&opt->priv->paths, old_path, 0);
2545 * strmap_put(&opt->priv->paths, new_path, ci);
2546 * However, there are some factors complicating this:
2547 * - opt->priv->paths may already have an entry at new_path
2548 * - Each ci tracks its containing directory, so we need to
2550 * - If another ci has the same containing directory, then
2551 * the two char*'s MUST point to the same location. See the
2552 * comment in struct merged_info. strcmp equality is not
2553 * enough; we need pointer equality.
2554 * - opt->priv->paths must hold the parent directories of any
2555 * entries that are added. So, if this directory rename
2556 * causes entirely new directories, we must recursively add
2557 * parent directories.
2558 * - For each parent directory added to opt->priv->paths, we
2559 * also need to get its parent directory stored in its
2560 * conflict_info->merged.directory_name with all the same
2561 * requirements about pointer equality.
2563 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2564 struct conflict_info
*ci
, *new_ci
;
2565 struct strmap_entry
*entry
;
2566 const char *branch_with_new_path
, *branch_with_dir_rename
;
2567 const char *old_path
= pair
->two
->path
;
2568 const char *parent_name
;
2569 const char *cur_path
;
2572 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2573 old_path
= entry
->key
;
2577 /* Find parent directories missing from opt->priv->paths */
2578 cur_path
= mem_pool_strdup(&opt
->priv
->pool
, new_path
);
2579 free((char*)new_path
);
2580 new_path
= (char *)cur_path
;
2583 /* Find the parent directory of cur_path */
2584 char *last_slash
= strrchr(cur_path
, '/');
2586 parent_name
= mem_pool_strndup(&opt
->priv
->pool
,
2588 last_slash
- cur_path
);
2590 parent_name
= opt
->priv
->toplevel_dir
;
2594 /* Look it up in opt->priv->paths */
2595 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2597 parent_name
= entry
->key
; /* reuse known pointer */
2601 /* Record this is one of the directories we need to insert */
2602 string_list_append(&dirs_to_insert
, parent_name
);
2603 cur_path
= parent_name
;
2606 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2607 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2608 struct conflict_info
*dir_ci
;
2609 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2611 CALLOC_ARRAY(dir_ci
, 1);
2613 dir_ci
->merged
.directory_name
= parent_name
;
2614 len
= strlen(parent_name
);
2615 /* len+1 because of trailing '/' character */
2616 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2617 dir_ci
->dirmask
= ci
->filemask
;
2618 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2620 parent_name
= cur_dir
;
2623 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2624 assert(ci
->dirmask
== 0 || ci
->dirmask
== 1);
2625 if (ci
->dirmask
== 0)
2626 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2629 * This file exists on one side, but we still had a directory
2630 * at the old location that we can't remove until after
2631 * processing all paths below it. So, make a copy of ci in
2632 * new_ci and only put the file information into it.
2634 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
2635 memcpy(new_ci
, ci
, sizeof(*ci
));
2636 assert(!new_ci
->match_mask
);
2637 new_ci
->dirmask
= 0;
2638 new_ci
->stages
[1].mode
= 0;
2639 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
2642 * Now that we have the file information in new_ci, make sure
2643 * ci only has the directory information.
2646 ci
->merged
.clean
= 1;
2647 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
2648 if (ci
->dirmask
& (1 << i
))
2650 /* zero out any entries related to files */
2651 ci
->stages
[i
].mode
= 0;
2652 oidcpy(&ci
->stages
[i
].oid
, null_oid());
2655 // Now we want to focus on new_ci, so reassign ci to it
2659 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2660 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2662 /* Now, finally update ci and stick it into opt->priv->paths */
2663 ci
->merged
.directory_name
= parent_name
;
2664 len
= strlen(parent_name
);
2665 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2666 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2668 /* Place ci back into opt->priv->paths, but at new_path */
2669 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2673 /* A few sanity checks */
2675 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2676 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2677 assert(!new_ci
->merged
.clean
);
2679 /* Copy stuff from ci into new_ci */
2680 new_ci
->filemask
|= ci
->filemask
;
2681 if (new_ci
->dirmask
)
2682 new_ci
->df_conflict
= 1;
2683 index
= (ci
->filemask
>> 1);
2684 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2685 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2686 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2691 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2692 /* Notify user of updated path */
2693 if (pair
->status
== 'A')
2694 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2695 new_path
, old_path
, NULL
, NULL
,
2696 _("Path updated: %s added in %s inside a "
2697 "directory that was renamed in %s; moving "
2699 old_path
, branch_with_new_path
,
2700 branch_with_dir_rename
, new_path
);
2702 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2703 new_path
, old_path
, NULL
, NULL
,
2704 _("Path updated: %s renamed to %s in %s, "
2705 "inside a directory that was renamed in %s; "
2706 "moving it to %s."),
2707 pair
->one
->path
, old_path
, branch_with_new_path
,
2708 branch_with_dir_rename
, new_path
);
2711 * opt->detect_directory_renames has the value
2712 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2714 ci
->path_conflict
= 1;
2715 if (pair
->status
== 'A')
2716 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2717 new_path
, old_path
, NULL
, NULL
,
2718 _("CONFLICT (file location): %s added in %s "
2719 "inside a directory that was renamed in %s, "
2720 "suggesting it should perhaps be moved to "
2722 old_path
, branch_with_new_path
,
2723 branch_with_dir_rename
, new_path
);
2725 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2726 new_path
, old_path
, NULL
, NULL
,
2727 _("CONFLICT (file location): %s renamed to %s "
2728 "in %s, inside a directory that was renamed "
2729 "in %s, suggesting it should perhaps be "
2731 pair
->one
->path
, old_path
, branch_with_new_path
,
2732 branch_with_dir_rename
, new_path
);
2736 * Finally, record the new location.
2738 pair
->two
->path
= new_path
;
2741 /*** Function Grouping: functions related to regular rename detection ***/
2743 static int process_renames(struct merge_options
*opt
,
2744 struct diff_queue_struct
*renames
)
2746 int clean_merge
= 1, i
;
2748 for (i
= 0; i
< renames
->nr
; ++i
) {
2749 const char *oldpath
= NULL
, *newpath
;
2750 struct diff_filepair
*pair
= renames
->queue
[i
];
2751 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2752 struct strmap_entry
*old_ent
, *new_ent
;
2753 unsigned int old_sidemask
;
2754 int target_index
, other_source_index
;
2755 int source_deleted
, collision
, type_changed
;
2756 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2758 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2759 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2761 oldpath
= old_ent
->key
;
2762 oldinfo
= old_ent
->value
;
2764 newpath
= pair
->two
->path
;
2766 newpath
= new_ent
->key
;
2767 newinfo
= new_ent
->value
;
2771 * If pair->one->path isn't in opt->priv->paths, that means
2772 * that either directory rename detection removed that
2773 * path, or a parent directory of oldpath was resolved and
2774 * we don't even need the rename; in either case, we can
2775 * skip it. If oldinfo->merged.clean, then the other side
2776 * of history had no changes to oldpath and we don't need
2777 * the rename and can skip it.
2779 if (!oldinfo
|| oldinfo
->merged
.clean
)
2783 * diff_filepairs have copies of pathnames, thus we have to
2784 * use standard 'strcmp()' (negated) instead of '=='.
2786 if (i
+ 1 < renames
->nr
&&
2787 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2788 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2789 const char *pathnames
[3];
2790 struct version_info merged
;
2791 struct conflict_info
*base
, *side1
, *side2
;
2792 unsigned was_binary_blob
= 0;
2794 pathnames
[0] = oldpath
;
2795 pathnames
[1] = newpath
;
2796 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2798 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2799 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2800 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2806 if (!strcmp(pathnames
[1], pathnames
[2])) {
2807 struct rename_info
*ri
= &opt
->priv
->renames
;
2810 /* Both sides renamed the same way */
2811 assert(side1
== side2
);
2812 memcpy(&side1
->stages
[0], &base
->stages
[0],
2814 side1
->filemask
|= (1 << MERGE_BASE
);
2815 /* Mark base as resolved by removal */
2816 base
->merged
.is_null
= 1;
2817 base
->merged
.clean
= 1;
2820 * Disable remembering renames optimization;
2821 * rename/rename(1to1) is incredibly rare, and
2822 * just disabling the optimization is easier
2823 * than purging cached_pairs,
2824 * cached_target_names, and dir_rename_counts.
2826 for (j
= 0; j
< 3; j
++)
2827 ri
->merge_trees
[j
] = NULL
;
2829 /* We handled both renames, i.e. i+1 handled */
2831 /* Move to next rename */
2835 /* This is a rename/rename(1to2) */
2836 clean_merge
= handle_content_merge(opt
,
2842 1 + 2 * opt
->priv
->call_depth
,
2844 if (clean_merge
< 0)
2847 merged
.mode
== side1
->stages
[1].mode
&&
2848 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2849 was_binary_blob
= 1;
2850 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2851 if (was_binary_blob
) {
2853 * Getting here means we were attempting to
2854 * merge a binary blob.
2856 * Since we can't merge binaries,
2857 * handle_content_merge() just takes one
2858 * side. But we don't want to copy the
2859 * contents of one side to both paths. We
2860 * used the contents of side1 above for
2861 * side1->stages, let's use the contents of
2862 * side2 for side2->stages below.
2864 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2865 merged
.mode
= side2
->stages
[2].mode
;
2867 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2869 side1
->path_conflict
= 1;
2870 side2
->path_conflict
= 1;
2872 * TODO: For renames we normally remove the path at the
2873 * old name. It would thus seem consistent to do the
2874 * same for rename/rename(1to2) cases, but we haven't
2875 * done so traditionally and a number of the regression
2876 * tests now encode an expectation that the file is
2877 * left there at stage 1. If we ever decide to change
2878 * this, add the following two lines here:
2879 * base->merged.is_null = 1;
2880 * base->merged.clean = 1;
2881 * and remove the setting of base->path_conflict to 1.
2883 base
->path_conflict
= 1;
2884 path_msg(opt
, CONFLICT_RENAME_RENAME
, 0,
2885 pathnames
[0], pathnames
[1], pathnames
[2], NULL
,
2886 _("CONFLICT (rename/rename): %s renamed to "
2887 "%s in %s and to %s in %s."),
2889 pathnames
[1], opt
->branch1
,
2890 pathnames
[2], opt
->branch2
);
2892 i
++; /* We handled both renames, i.e. i+1 handled */
2898 target_index
= pair
->score
; /* from collect_renames() */
2899 assert(target_index
== 1 || target_index
== 2);
2900 other_source_index
= 3 - target_index
;
2901 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2902 source_deleted
= (oldinfo
->filemask
== 1);
2903 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2904 type_changed
= !source_deleted
&&
2905 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2906 S_ISREG(newinfo
->stages
[target_index
].mode
));
2907 if (type_changed
&& collision
) {
2909 * special handling so later blocks can handle this...
2911 * if type_changed && collision are both true, then this
2912 * was really a double rename, but one side wasn't
2913 * detected due to lack of break detection. I.e.
2915 * orig: has normal file 'foo'
2916 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2917 * side2: renames 'foo' to 'bar'
2918 * In this case, the foo->bar rename on side1 won't be
2919 * detected because the new symlink named 'foo' is
2920 * there and we don't do break detection. But we detect
2921 * this here because we don't want to merge the content
2922 * of the foo symlink with the foo->bar file, so we
2923 * have some logic to handle this special case. The
2924 * easiest way to do that is make 'bar' on side1 not
2925 * be considered a colliding file but the other part
2926 * of a normal rename. If the file is very different,
2927 * well we're going to get content merge conflicts
2928 * anyway so it doesn't hurt. And if the colliding
2929 * file also has a different type, that'll be handled
2930 * by the content merge logic in process_entry() too.
2932 * See also t6430, 'rename vs. rename/symlink'
2936 if (source_deleted
) {
2937 if (target_index
== 1) {
2938 rename_branch
= opt
->branch1
;
2939 delete_branch
= opt
->branch2
;
2941 rename_branch
= opt
->branch2
;
2942 delete_branch
= opt
->branch1
;
2946 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2948 /* Need to check for special types of rename conflicts... */
2949 if (collision
&& !source_deleted
) {
2950 /* collision: rename/add or rename/rename(2to1) */
2951 const char *pathnames
[3];
2952 struct version_info merged
;
2954 struct conflict_info
*base
, *side1
, *side2
;
2957 pathnames
[0] = oldpath
;
2958 pathnames
[other_source_index
] = oldpath
;
2959 pathnames
[target_index
] = newpath
;
2961 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2962 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2963 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2969 clean
= handle_content_merge(opt
, pair
->one
->path
,
2974 1 + 2 * opt
->priv
->call_depth
,
2979 memcpy(&newinfo
->stages
[target_index
], &merged
,
2982 path_msg(opt
, CONFLICT_RENAME_COLLIDES
, 0,
2983 newpath
, oldpath
, NULL
, NULL
,
2984 _("CONFLICT (rename involved in "
2985 "collision): rename of %s -> %s has "
2986 "content conflicts AND collides "
2987 "with another path; this may result "
2988 "in nested conflict markers."),
2991 } else if (collision
&& source_deleted
) {
2993 * rename/add/delete or rename/rename(2to1)/delete:
2994 * since oldpath was deleted on the side that didn't
2995 * do the rename, there's not much of a content merge
2996 * we can do for the rename. oldinfo->merged.is_null
2997 * was already set, so we just leave things as-is so
2998 * they look like an add/add conflict.
3001 newinfo
->path_conflict
= 1;
3002 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
3003 newpath
, oldpath
, NULL
, NULL
,
3004 _("CONFLICT (rename/delete): %s renamed "
3005 "to %s in %s, but deleted in %s."),
3006 oldpath
, newpath
, rename_branch
, delete_branch
);
3009 * a few different cases...start by copying the
3010 * existing stage(s) from oldinfo over the newinfo
3011 * and update the pathname(s).
3013 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
3014 sizeof(newinfo
->stages
[0]));
3015 newinfo
->filemask
|= (1 << MERGE_BASE
);
3016 newinfo
->pathnames
[0] = oldpath
;
3018 /* rename vs. typechange */
3019 /* Mark the original as resolved by removal */
3020 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
3021 sizeof(oldinfo
->stages
[0].oid
));
3022 oldinfo
->stages
[0].mode
= 0;
3023 oldinfo
->filemask
&= 0x06;
3024 } else if (source_deleted
) {
3026 newinfo
->path_conflict
= 1;
3027 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
3028 newpath
, oldpath
, NULL
, NULL
,
3029 _("CONFLICT (rename/delete): %s renamed"
3030 " to %s in %s, but deleted in %s."),
3032 rename_branch
, delete_branch
);
3035 memcpy(&newinfo
->stages
[other_source_index
],
3036 &oldinfo
->stages
[other_source_index
],
3037 sizeof(newinfo
->stages
[0]));
3038 newinfo
->filemask
|= (1 << other_source_index
);
3039 newinfo
->pathnames
[other_source_index
] = oldpath
;
3043 if (!type_changed
) {
3044 /* Mark the original as resolved by removal */
3045 oldinfo
->merged
.is_null
= 1;
3046 oldinfo
->merged
.clean
= 1;
3054 static inline int possible_side_renames(struct rename_info
*renames
,
3055 unsigned side_index
)
3057 return renames
->pairs
[side_index
].nr
> 0 &&
3058 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
3061 static inline int possible_renames(struct rename_info
*renames
)
3063 return possible_side_renames(renames
, 1) ||
3064 possible_side_renames(renames
, 2) ||
3065 !strmap_empty(&renames
->cached_pairs
[1]) ||
3066 !strmap_empty(&renames
->cached_pairs
[2]);
3069 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
3072 * A simplified version of diff_resolve_rename_copy(); would probably
3073 * just use that function but it's static...
3076 struct diff_filepair
*p
;
3078 for (i
= 0; i
< q
->nr
; ++i
) {
3080 p
->status
= 0; /* undecided */
3081 if (!DIFF_FILE_VALID(p
->one
))
3082 p
->status
= DIFF_STATUS_ADDED
;
3083 else if (!DIFF_FILE_VALID(p
->two
))
3084 p
->status
= DIFF_STATUS_DELETED
;
3085 else if (DIFF_PAIR_RENAME(p
))
3086 p
->status
= DIFF_STATUS_RENAMED
;
3090 static void prune_cached_from_relevant(struct rename_info
*renames
,
3093 /* Reason for this function described in add_pair() */
3094 struct hashmap_iter iter
;
3095 struct strmap_entry
*entry
;
3097 /* Remove from relevant_sources all entries in cached_pairs[side] */
3098 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
3099 strintmap_remove(&renames
->relevant_sources
[side
],
3102 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
3103 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
3104 strintmap_remove(&renames
->relevant_sources
[side
],
3109 static void use_cached_pairs(struct merge_options
*opt
,
3110 struct strmap
*cached_pairs
,
3111 struct diff_queue_struct
*pairs
)
3113 struct hashmap_iter iter
;
3114 struct strmap_entry
*entry
;
3117 * Add to side_pairs all entries from renames->cached_pairs[side_index].
3118 * (Info in cached_irrelevant[side_index] is not relevant here.)
3120 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
3121 struct diff_filespec
*one
, *two
;
3122 const char *old_name
= entry
->key
;
3123 const char *new_name
= entry
->value
;
3125 new_name
= old_name
;
3128 * cached_pairs has *copies* of old_name and new_name,
3129 * because it has to persist across merges. Since
3130 * pool_alloc_filespec() will just re-use the existing
3131 * filenames, which will also get re-used by
3132 * opt->priv->paths if they become renames, and then
3133 * get freed at the end of the merge, that would leave
3134 * the copy in cached_pairs dangling. Avoid this by
3135 * making a copy here.
3137 old_name
= mem_pool_strdup(&opt
->priv
->pool
, old_name
);
3138 new_name
= mem_pool_strdup(&opt
->priv
->pool
, new_name
);
3140 /* We don't care about oid/mode, only filenames and status */
3141 one
= pool_alloc_filespec(&opt
->priv
->pool
, old_name
);
3142 two
= pool_alloc_filespec(&opt
->priv
->pool
, new_name
);
3143 pool_diff_queue(&opt
->priv
->pool
, pairs
, one
, two
);
3144 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
3148 static void cache_new_pair(struct rename_info
*renames
,
3155 new_path
= xstrdup(new_path
);
3156 old_value
= strmap_put(&renames
->cached_pairs
[side
],
3157 old_path
, new_path
);
3158 strset_add(&renames
->cached_target_names
[side
], new_path
);
3165 static void possibly_cache_new_pair(struct rename_info
*renames
,
3166 struct diff_filepair
*p
,
3170 int dir_renamed_side
= 0;
3174 * Directory renames happen on the other side of history from
3175 * the side that adds new files to the old directory.
3177 dir_renamed_side
= 3 - side
;
3179 int val
= strintmap_get(&renames
->relevant_sources
[side
],
3181 if (val
== RELEVANT_NO_MORE
) {
3182 assert(p
->status
== 'D');
3183 strset_add(&renames
->cached_irrelevant
[side
],
3190 if (p
->status
== 'D') {
3192 * If we already had this delete, we'll just set it's value
3193 * to NULL again, so no harm.
3195 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
3196 } else if (p
->status
== 'R') {
3198 new_path
= p
->two
->path
;
3200 cache_new_pair(renames
, dir_renamed_side
,
3201 p
->two
->path
, new_path
, 0);
3202 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
3203 } else if (p
->status
== 'A' && new_path
) {
3204 cache_new_pair(renames
, dir_renamed_side
,
3205 p
->two
->path
, new_path
, 0);
3209 static int compare_pairs(const void *a_
, const void *b_
)
3211 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
3212 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
3214 return strcmp(a
->one
->path
, b
->one
->path
);
3217 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
3218 static int detect_regular_renames(struct merge_options
*opt
,
3219 unsigned side_index
)
3221 struct diff_options diff_opts
;
3222 struct rename_info
*renames
= &opt
->priv
->renames
;
3224 prune_cached_from_relevant(renames
, side_index
);
3225 if (!possible_side_renames(renames
, side_index
)) {
3227 * No rename detection needed for this side, but we still need
3228 * to make sure 'adds' are marked correctly in case the other
3229 * side had directory renames.
3231 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
3235 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
3236 repo_diff_setup(opt
->repo
, &diff_opts
);
3237 diff_opts
.flags
.recursive
= 1;
3238 diff_opts
.flags
.rename_empty
= 0;
3239 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
3240 diff_opts
.rename_limit
= opt
->rename_limit
;
3241 if (opt
->rename_limit
<= 0)
3242 diff_opts
.rename_limit
= 7000;
3243 diff_opts
.rename_score
= opt
->rename_score
;
3244 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
3245 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3246 diff_setup_done(&diff_opts
);
3248 diff_queued_diff
= renames
->pairs
[side_index
];
3249 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
3250 diffcore_rename_extended(&diff_opts
,
3252 &renames
->relevant_sources
[side_index
],
3253 &renames
->dirs_removed
[side_index
],
3254 &renames
->dir_rename_count
[side_index
],
3255 &renames
->cached_pairs
[side_index
]);
3256 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
3257 resolve_diffpair_statuses(&diff_queued_diff
);
3259 if (diff_opts
.needed_rename_limit
> 0)
3260 renames
->redo_after_renames
= 0;
3261 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
3262 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
3264 renames
->pairs
[side_index
] = diff_queued_diff
;
3266 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3267 diff_queued_diff
.nr
= 0;
3268 diff_queued_diff
.queue
= NULL
;
3269 diff_flush(&diff_opts
);
3275 * Get information of all renames which occurred in 'side_pairs', making use
3276 * of any implicit directory renames in side_dir_renames (also making use of
3277 * implicit directory renames rename_exclusions as needed by
3278 * check_for_directory_rename()). Add all (updated) renames into result.
3280 static int collect_renames(struct merge_options
*opt
,
3281 struct diff_queue_struct
*result
,
3282 unsigned side_index
,
3283 struct strmap
*collisions
,
3284 struct strmap
*dir_renames_for_side
,
3285 struct strmap
*rename_exclusions
)
3288 struct diff_queue_struct
*side_pairs
;
3289 struct rename_info
*renames
= &opt
->priv
->renames
;
3291 side_pairs
= &renames
->pairs
[side_index
];
3293 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3294 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3295 char *new_path
; /* non-NULL only with directory renames */
3297 if (p
->status
!= 'A' && p
->status
!= 'R') {
3298 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3299 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3303 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3305 dir_renames_for_side
,
3310 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3311 if (p
->status
!= 'R' && !new_path
) {
3312 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3317 apply_directory_rename_modifications(opt
, p
, new_path
);
3320 * p->score comes back from diffcore_rename_extended() with
3321 * the similarity of the renamed file. The similarity is
3322 * was used to determine that the two files were related
3323 * and are a rename, which we have already used, but beyond
3324 * that we have no use for the similarity. So p->score is
3325 * now irrelevant. However, process_renames() will need to
3326 * know which side of the merge this rename was associated
3327 * with, so overwrite p->score with that value.
3329 p
->score
= side_index
;
3330 result
->queue
[result
->nr
++] = p
;
3336 static int detect_and_process_renames(struct merge_options
*opt
,
3337 struct tree
*merge_base
,
3341 struct diff_queue_struct combined
= { 0 };
3342 struct rename_info
*renames
= &opt
->priv
->renames
;
3343 struct strmap collisions
[3];
3344 int need_dir_renames
, s
, i
, clean
= 1;
3345 unsigned detection_run
= 0;
3347 if (!possible_renames(renames
))
3350 trace2_region_enter("merge", "regular renames", opt
->repo
);
3351 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3352 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3353 if (renames
->needed_limit
) {
3354 renames
->cached_pairs_valid_side
= 0;
3355 renames
->redo_after_renames
= 0;
3357 if (renames
->redo_after_renames
&& detection_run
) {
3359 struct diff_filepair
*p
;
3361 /* Cache the renames, we found */
3362 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3363 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3364 p
= renames
->pairs
[side
].queue
[i
];
3365 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3369 /* Restart the merge with the cached renames */
3370 renames
->redo_after_renames
= 2;
3371 trace2_region_leave("merge", "regular renames", opt
->repo
);
3374 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3375 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3376 trace2_region_leave("merge", "regular renames", opt
->repo
);
3378 trace2_region_enter("merge", "directory renames", opt
->repo
);
3380 !opt
->priv
->call_depth
&&
3381 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3382 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3384 if (need_dir_renames
) {
3385 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3386 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3387 handle_directory_level_conflicts(opt
);
3390 ALLOC_GROW(combined
.queue
,
3391 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3393 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
3394 int other_side
= 3 - i
;
3395 compute_collisions(&collisions
[i
],
3396 &renames
->dir_renames
[other_side
],
3397 &renames
->pairs
[i
]);
3399 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3401 &renames
->dir_renames
[2],
3402 &renames
->dir_renames
[1]);
3403 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3405 &renames
->dir_renames
[1],
3406 &renames
->dir_renames
[2]);
3407 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++)
3408 free_collisions(&collisions
[i
]);
3409 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3410 trace2_region_leave("merge", "directory renames", opt
->repo
);
3412 trace2_region_enter("merge", "process renames", opt
->repo
);
3413 clean
&= process_renames(opt
, &combined
);
3414 trace2_region_leave("merge", "process renames", opt
->repo
);
3416 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3420 * Free now unneeded filepairs, which would have been handled
3421 * in collect_renames() normally but we skipped that code.
3423 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3424 struct diff_queue_struct
*side_pairs
;
3427 side_pairs
= &renames
->pairs
[s
];
3428 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3429 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3430 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3435 /* Free memory for renames->pairs[] and combined */
3436 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3437 free(renames
->pairs
[s
].queue
);
3438 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3440 for (i
= 0; i
< combined
.nr
; i
++)
3441 pool_diff_free_filepair(&opt
->priv
->pool
, combined
.queue
[i
]);
3442 free(combined
.queue
);
3447 /*** Function Grouping: functions related to process_entries() ***/
3449 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3451 unsigned char c1
, c2
;
3454 * Here we only care that entries for directories appear adjacent
3455 * to and before files underneath the directory. We can achieve
3456 * that by pretending to add a trailing slash to every file and
3457 * then sorting. In other words, we do not want the natural
3462 * Instead, we want "foo" to sort as though it were "foo/", so that
3467 * To achieve this, we basically implement our own strcmp, except that
3468 * if we get to the end of either string instead of comparing NUL to
3469 * another character, we compare '/' to it.
3471 * If this unusual "sort as though '/' were appended" perplexes
3472 * you, perhaps it will help to note that this is not the final
3473 * sort. write_tree() will sort again without the trailing slash
3474 * magic, but just on paths immediately under a given tree.
3476 * The reason to not use df_name_compare directly was that it was
3477 * just too expensive (we don't have the string lengths handy), so
3478 * it was reimplemented.
3482 * NOTE: This function will never be called with two equal strings,
3483 * because it is used to sort the keys of a strmap, and strmaps have
3484 * unique keys by construction. That simplifies our c1==c2 handling
3488 while (*one
&& (*one
== *two
)) {
3493 c1
= *one
? *one
: '/';
3494 c2
= *two
? *two
: '/';
3497 /* Getting here means one is a leading directory of the other */
3498 return (*one
) ? 1 : -1;
3503 static int read_oid_strbuf(struct merge_options
*opt
,
3504 const struct object_id
*oid
,
3508 enum object_type type
;
3510 buf
= read_object_file(oid
, &type
, &size
);
3512 return err(opt
, _("cannot read object %s"), oid_to_hex(oid
));
3513 if (type
!= OBJ_BLOB
) {
3515 return err(opt
, _("object %s is not a blob"), oid_to_hex(oid
));
3517 strbuf_attach(dst
, buf
, size
, size
+ 1);
3521 static int blob_unchanged(struct merge_options
*opt
,
3522 const struct version_info
*base
,
3523 const struct version_info
*side
,
3526 struct strbuf basebuf
= STRBUF_INIT
;
3527 struct strbuf sidebuf
= STRBUF_INIT
;
3528 int ret
= 0; /* assume changed for safety */
3529 struct index_state
*idx
= &opt
->priv
->attr_index
;
3531 if (!idx
->initialized
)
3532 initialize_attr_index(opt
);
3534 if (base
->mode
!= side
->mode
)
3536 if (oideq(&base
->oid
, &side
->oid
))
3539 if (read_oid_strbuf(opt
, &base
->oid
, &basebuf
) ||
3540 read_oid_strbuf(opt
, &side
->oid
, &sidebuf
))
3543 * Note: binary | is used so that both renormalizations are
3544 * performed. Comparison can be skipped if both files are
3545 * unchanged since their sha1s have already been compared.
3547 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3548 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3549 ret
= (basebuf
.len
== sidebuf
.len
&&
3550 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3553 strbuf_release(&basebuf
);
3554 strbuf_release(&sidebuf
);
3558 struct directory_versions
{
3560 * versions: list of (basename -> version_info)
3562 * The basenames are in reverse lexicographic order of full pathnames,
3563 * as processed in process_entries(). This puts all entries within
3564 * a directory together, and covers the directory itself after
3565 * everything within it, allowing us to write subtrees before needing
3566 * to record information for the tree itself.
3568 struct string_list versions
;
3571 * offsets: list of (full relative path directories -> integer offsets)
3573 * Since versions contains basenames from files in multiple different
3574 * directories, we need to know which entries in versions correspond
3575 * to which directories. Values of e.g.
3579 * Would mean that entries 0-1 of versions are files in the toplevel
3580 * directory, entries 2-4 are files under src/, and the remaining
3581 * entries starting at index 5 are files under src/moduleA/.
3583 struct string_list offsets
;
3586 * last_directory: directory that previously processed file found in
3588 * last_directory starts NULL, but records the directory in which the
3589 * previous file was found within. As soon as
3590 * directory(current_file) != last_directory
3591 * then we need to start updating accounting in versions & offsets.
3592 * Note that last_directory is always the last path in "offsets" (or
3593 * NULL if "offsets" is empty) so this exists just for quick access.
3595 const char *last_directory
;
3597 /* last_directory_len: cached computation of strlen(last_directory) */
3598 unsigned last_directory_len
;
3601 static int tree_entry_order(const void *a_
, const void *b_
)
3603 const struct string_list_item
*a
= a_
;
3604 const struct string_list_item
*b
= b_
;
3606 const struct merged_info
*ami
= a
->util
;
3607 const struct merged_info
*bmi
= b
->util
;
3608 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3609 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3612 static int write_tree(struct object_id
*result_oid
,
3613 struct string_list
*versions
,
3614 unsigned int offset
,
3617 size_t maxlen
= 0, extra
;
3619 struct strbuf buf
= STRBUF_INIT
;
3622 assert(offset
<= versions
->nr
);
3623 nr
= versions
->nr
- offset
;
3625 /* No need for STABLE_QSORT -- filenames must be unique */
3626 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3628 /* Pre-allocate some space in buf */
3629 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3630 for (i
= 0; i
< nr
; i
++) {
3631 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3633 strbuf_grow(&buf
, maxlen
);
3635 /* Write each entry out to buf */
3636 for (i
= 0; i
< nr
; i
++) {
3637 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3638 struct version_info
*ri
= &mi
->result
;
3639 strbuf_addf(&buf
, "%o %s%c",
3641 versions
->items
[offset
+i
].string
, '\0');
3642 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3645 /* Write this object file out, and record in result_oid */
3646 if (write_object_file(buf
.buf
, buf
.len
, OBJ_TREE
, result_oid
))
3648 strbuf_release(&buf
);
3652 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3654 struct merged_info
*mi
)
3656 const char *basename
;
3659 /* nothing to record */
3662 basename
= path
+ mi
->basename_offset
;
3663 assert(strchr(basename
, '/') == NULL
);
3664 string_list_append(&dir_metadata
->versions
,
3665 basename
)->util
= &mi
->result
;
3668 static int write_completed_directory(struct merge_options
*opt
,
3669 const char *new_directory_name
,
3670 struct directory_versions
*info
)
3672 const char *prev_dir
;
3673 struct merged_info
*dir_info
= NULL
;
3674 unsigned int offset
, ret
= 0;
3677 * Some explanation of info->versions and info->offsets...
3679 * process_entries() iterates over all relevant files AND
3680 * directories in reverse lexicographic order, and calls this
3681 * function. Thus, an example of the paths that process_entries()
3682 * could operate on (along with the directories for those paths
3687 * src/moduleB/umm.c src/moduleB
3688 * src/moduleB/stuff.h src/moduleB
3689 * src/moduleB/baz.c src/moduleB
3691 * src/moduleA/foo.c src/moduleA
3692 * src/moduleA/bar.c src/moduleA
3699 * always contains the unprocessed entries and their
3700 * version_info information. For example, after the first five
3701 * entries above, info->versions would be:
3703 * xtract.c <xtract.c's version_info>
3704 * token.txt <token.txt's version_info>
3705 * umm.c <src/moduleB/umm.c's version_info>
3706 * stuff.h <src/moduleB/stuff.h's version_info>
3707 * baz.c <src/moduleB/baz.c's version_info>
3709 * Once a subdirectory is completed we remove the entries in
3710 * that subdirectory from info->versions, writing it as a tree
3711 * (write_tree()). Thus, as soon as we get to src/moduleB,
3712 * info->versions would be updated to
3714 * xtract.c <xtract.c's version_info>
3715 * token.txt <token.txt's version_info>
3716 * moduleB <src/moduleB's version_info>
3720 * helps us track which entries in info->versions correspond to
3721 * which directories. When we are N directories deep (e.g. 4
3722 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3723 * directories (+1 because of toplevel dir). Corresponding to
3724 * the info->versions example above, after processing five entries
3725 * info->offsets will be:
3730 * which is used to know that xtract.c & token.txt are from the
3731 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3732 * src/moduleB directory. Again, following the example above,
3733 * once we need to process src/moduleB, then info->offsets is
3739 * which says that moduleB (and only moduleB so far) is in the
3742 * One unique thing to note about info->offsets here is that
3743 * "src" was not added to info->offsets until there was a path
3744 * (a file OR directory) immediately below src/ that got
3747 * Since process_entry() just appends new entries to info->versions,
3748 * write_completed_directory() only needs to do work if the next path
3749 * is in a directory that is different than the last directory found
3754 * If we are working with the same directory as the last entry, there
3755 * is no work to do. (See comments above the directory_name member of
3756 * struct merged_info for why we can use pointer comparison instead of
3759 if (new_directory_name
== info
->last_directory
)
3763 * If we are just starting (last_directory is NULL), or last_directory
3764 * is a prefix of the current directory, then we can just update
3765 * info->offsets to record the offset where we started this directory
3766 * and update last_directory to have quick access to it.
3768 if (info
->last_directory
== NULL
||
3769 !strncmp(new_directory_name
, info
->last_directory
,
3770 info
->last_directory_len
)) {
3771 uintptr_t offset
= info
->versions
.nr
;
3773 info
->last_directory
= new_directory_name
;
3774 info
->last_directory_len
= strlen(info
->last_directory
);
3776 * Record the offset into info->versions where we will
3777 * start recording basenames of paths found within
3778 * new_directory_name.
3780 string_list_append(&info
->offsets
,
3781 info
->last_directory
)->util
= (void*)offset
;
3786 * The next entry that will be processed will be within
3787 * new_directory_name. Since at this point we know that
3788 * new_directory_name is within a different directory than
3789 * info->last_directory, we have all entries for info->last_directory
3790 * in info->versions and we need to create a tree object for them.
3792 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3794 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3795 if (offset
== info
->versions
.nr
) {
3797 * Actually, we don't need to create a tree object in this
3798 * case. Whenever all files within a directory disappear
3799 * during the merge (e.g. unmodified on one side and
3800 * deleted on the other, or files were renamed elsewhere),
3801 * then we get here and the directory itself needs to be
3802 * omitted from its parent tree as well.
3804 dir_info
->is_null
= 1;
3807 * Write out the tree to the git object directory, and also
3808 * record the mode and oid in dir_info->result.
3810 dir_info
->is_null
= 0;
3811 dir_info
->result
.mode
= S_IFDIR
;
3812 if (write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3813 opt
->repo
->hash_algo
->rawsz
) < 0)
3818 * We've now used several entries from info->versions and one entry
3819 * from info->offsets, so we get rid of those values.
3822 info
->versions
.nr
= offset
;
3825 * Now we've taken care of the completed directory, but we need to
3826 * prepare things since future entries will be in
3827 * new_directory_name. (In particular, process_entry() will be
3828 * appending new entries to info->versions.) So, we need to make
3829 * sure new_directory_name is the last entry in info->offsets.
3831 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3832 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3833 if (new_directory_name
!= prev_dir
) {
3834 uintptr_t c
= info
->versions
.nr
;
3835 string_list_append(&info
->offsets
,
3836 new_directory_name
)->util
= (void*)c
;
3839 /* And, of course, we need to update last_directory to match. */
3840 info
->last_directory
= new_directory_name
;
3841 info
->last_directory_len
= strlen(info
->last_directory
);
3846 /* Per entry merge function */
3847 static int process_entry(struct merge_options
*opt
,
3849 struct conflict_info
*ci
,
3850 struct directory_versions
*dir_metadata
)
3852 int df_file_index
= 0;
3855 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3856 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3857 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3858 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3861 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3862 if (ci
->filemask
== 0)
3863 /* nothing else to handle */
3865 assert(ci
->df_conflict
);
3868 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3872 * directory no longer in the way, but we do have a file we
3873 * need to place here so we need to clean away the "directory
3874 * merges to nothing" result.
3876 ci
->df_conflict
= 0;
3877 assert(ci
->filemask
!= 0);
3878 ci
->merged
.clean
= 0;
3879 ci
->merged
.is_null
= 0;
3880 /* and we want to zero out any directory-related entries */
3881 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3883 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3884 if (ci
->filemask
& (1 << i
))
3886 ci
->stages
[i
].mode
= 0;
3887 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3889 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3891 * This started out as a D/F conflict, and the entries in
3892 * the competing directory were not removed by the merge as
3893 * evidenced by write_completed_directory() writing a value
3894 * to ci->merged.result.mode.
3896 struct conflict_info
*new_ci
;
3898 const char *old_path
= path
;
3901 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3904 * If filemask is 1, we can just ignore the file as having
3905 * been deleted on both sides. We do not want to overwrite
3906 * ci->merged.result, since it stores the tree for all the
3909 if (ci
->filemask
== 1) {
3915 * This file still exists on at least one side, and we want
3916 * the directory to remain here, so we need to move this
3917 * path to some new location.
3919 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
3921 /* We don't really want new_ci->merged.result copied, but it'll
3922 * be overwritten below so it doesn't matter. We also don't
3923 * want any directory mode/oid values copied, but we'll zero
3924 * those out immediately. We do want the rest of ci copied.
3926 memcpy(new_ci
, ci
, sizeof(*ci
));
3927 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3928 new_ci
->dirmask
= 0;
3929 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3930 if (new_ci
->filemask
& (1 << i
))
3932 /* zero out any entries related to directories */
3933 new_ci
->stages
[i
].mode
= 0;
3934 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3938 * Find out which side this file came from; note that we
3939 * cannot just use ci->filemask, because renames could cause
3940 * the filemask to go back to 7. So we use dirmask, then
3941 * pick the opposite side's index.
3943 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3944 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3945 path
= unique_path(opt
, path
, branch
);
3946 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3948 path_msg(opt
, CONFLICT_FILE_DIRECTORY
, 0,
3949 path
, old_path
, NULL
, NULL
,
3950 _("CONFLICT (file/directory): directory in the way "
3951 "of %s from %s; moving it to %s instead."),
3952 old_path
, branch
, path
);
3955 * Zero out the filemask for the old ci. At this point, ci
3956 * was just an entry for a directory, so we don't need to
3957 * do anything more with it.
3962 * Now note that we're working on the new entry (path was
3969 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3970 * which the code goes through even for the df_conflict cases
3973 if (ci
->match_mask
) {
3974 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3975 if (ci
->match_mask
== 6) {
3976 /* stages[1] == stages[2] */
3977 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3978 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3980 /* determine the mask of the side that didn't match */
3981 unsigned int othermask
= 7 & ~ci
->match_mask
;
3982 int side
= (othermask
== 4) ? 2 : 1;
3984 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3985 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3986 if (ci
->merged
.is_null
)
3987 ci
->merged
.clean
= 1;
3988 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3990 assert(othermask
== 2 || othermask
== 4);
3991 assert(ci
->merged
.is_null
==
3992 (ci
->filemask
== ci
->match_mask
));
3994 } else if (ci
->filemask
>= 6 &&
3995 (S_IFMT
& ci
->stages
[1].mode
) !=
3996 (S_IFMT
& ci
->stages
[2].mode
)) {
3997 /* Two different items from (file/submodule/symlink) */
3998 if (opt
->priv
->call_depth
) {
3999 /* Just use the version from the merge base */
4000 ci
->merged
.clean
= 0;
4001 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
4002 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
4003 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
4005 /* Handle by renaming one or both to separate paths. */
4006 unsigned o_mode
= ci
->stages
[0].mode
;
4007 unsigned a_mode
= ci
->stages
[1].mode
;
4008 unsigned b_mode
= ci
->stages
[2].mode
;
4009 struct conflict_info
*new_ci
;
4010 const char *a_path
= NULL
, *b_path
= NULL
;
4011 int rename_a
= 0, rename_b
= 0;
4013 new_ci
= mem_pool_alloc(&opt
->priv
->pool
,
4016 if (S_ISREG(a_mode
))
4018 else if (S_ISREG(b_mode
))
4026 a_path
= unique_path(opt
, path
, opt
->branch1
);
4028 b_path
= unique_path(opt
, path
, opt
->branch2
);
4030 if (rename_a
&& rename_b
) {
4031 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4032 path
, a_path
, b_path
, NULL
,
4033 _("CONFLICT (distinct types): %s had "
4034 "different types on each side; "
4035 "renamed both of them so each can "
4036 "be recorded somewhere."),
4039 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4040 path
, rename_a
? a_path
: b_path
,
4042 _("CONFLICT (distinct types): %s had "
4043 "different types on each side; "
4044 "renamed one of them so each can be "
4045 "recorded somewhere."),
4049 ci
->merged
.clean
= 0;
4050 memcpy(new_ci
, ci
, sizeof(*new_ci
));
4052 /* Put b into new_ci, removing a from stages */
4053 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
4054 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
4055 new_ci
->stages
[1].mode
= 0;
4056 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
4057 new_ci
->filemask
= 5;
4058 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
4059 new_ci
->stages
[0].mode
= 0;
4060 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
4061 new_ci
->filemask
= 4;
4064 /* Leave only a in ci, fixing stages. */
4065 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
4066 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
4067 ci
->stages
[2].mode
= 0;
4068 oidcpy(&ci
->stages
[2].oid
, null_oid());
4070 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
4071 ci
->stages
[0].mode
= 0;
4072 oidcpy(&ci
->stages
[0].oid
, null_oid());
4076 /* Insert entries into opt->priv_paths */
4077 assert(rename_a
|| rename_b
);
4079 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
4083 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
4085 if (rename_a
&& rename_b
)
4086 strmap_remove(&opt
->priv
->paths
, path
, 0);
4089 * Do special handling for b_path since process_entry()
4090 * won't be called on it specially.
4092 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
4093 record_entry_for_tree(dir_metadata
, b_path
,
4097 * Remaining code for processing this entry should
4098 * think in terms of processing a_path.
4103 } else if (ci
->filemask
>= 6) {
4104 /* Need a two-way or three-way content merge */
4105 struct version_info merged_file
;
4107 struct version_info
*o
= &ci
->stages
[0];
4108 struct version_info
*a
= &ci
->stages
[1];
4109 struct version_info
*b
= &ci
->stages
[2];
4111 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
4113 opt
->priv
->call_depth
* 2,
4115 if (clean_merge
< 0)
4117 ci
->merged
.clean
= clean_merge
&&
4118 !ci
->df_conflict
&& !ci
->path_conflict
;
4119 ci
->merged
.result
.mode
= merged_file
.mode
;
4120 ci
->merged
.is_null
= (merged_file
.mode
== 0);
4121 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
4122 if (clean_merge
&& ci
->df_conflict
) {
4123 assert(df_file_index
== 1 || df_file_index
== 2);
4124 ci
->filemask
= 1 << df_file_index
;
4125 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
4126 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
4129 const char *reason
= _("content");
4130 if (ci
->filemask
== 6)
4131 reason
= _("add/add");
4132 if (S_ISGITLINK(merged_file
.mode
))
4133 reason
= _("submodule");
4134 path_msg(opt
, CONFLICT_CONTENTS
, 0,
4135 path
, NULL
, NULL
, NULL
,
4136 _("CONFLICT (%s): Merge conflict in %s"),
4139 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
4141 const char *modify_branch
, *delete_branch
;
4142 int side
= (ci
->filemask
== 5) ? 2 : 1;
4143 int index
= opt
->priv
->call_depth
? 0 : side
;
4145 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
4146 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
4147 ci
->merged
.clean
= 0;
4149 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
4150 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
4152 if (opt
->renormalize
&&
4153 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
4155 if (!ci
->path_conflict
) {
4157 * Blob unchanged after renormalization, so
4158 * there's no modify/delete conflict after all;
4159 * we can just remove the file.
4161 ci
->merged
.is_null
= 1;
4162 ci
->merged
.clean
= 1;
4164 * file goes away => even if there was a
4165 * directory/file conflict there isn't one now.
4167 ci
->df_conflict
= 0;
4169 /* rename/delete, so conflict remains */
4171 } else if (ci
->path_conflict
&&
4172 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
4174 * This came from a rename/delete; no action to take,
4175 * but avoid printing "modify/delete" conflict notice
4176 * since the contents were not modified.
4179 path_msg(opt
, CONFLICT_MODIFY_DELETE
, 0,
4180 path
, NULL
, NULL
, NULL
,
4181 _("CONFLICT (modify/delete): %s deleted in %s "
4182 "and modified in %s. Version %s of %s left "
4184 path
, delete_branch
, modify_branch
,
4185 modify_branch
, path
);
4187 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
4188 /* Added on one side */
4189 int side
= (ci
->filemask
== 4) ? 2 : 1;
4190 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
4191 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
4192 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
4193 } else if (ci
->filemask
== 1) {
4194 /* Deleted on both sides */
4195 ci
->merged
.is_null
= 1;
4196 ci
->merged
.result
.mode
= 0;
4197 oidcpy(&ci
->merged
.result
.oid
, null_oid());
4198 assert(!ci
->df_conflict
);
4199 ci
->merged
.clean
= !ci
->path_conflict
;
4203 * If still conflicted, record it separately. This allows us to later
4204 * iterate over just conflicted entries when updating the index instead
4205 * of iterating over all entries.
4207 if (!ci
->merged
.clean
)
4208 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
4210 /* Record metadata for ci->merged in dir_metadata */
4211 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
4215 static void prefetch_for_content_merges(struct merge_options
*opt
,
4216 struct string_list
*plist
)
4218 struct string_list_item
*e
;
4219 struct oid_array to_fetch
= OID_ARRAY_INIT
;
4221 if (opt
->repo
!= the_repository
|| !has_promisor_remote())
4224 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
4225 /* char *path = e->string; */
4226 struct conflict_info
*ci
= e
->util
;
4229 /* Ignore clean entries */
4230 if (ci
->merged
.clean
)
4233 /* Ignore entries that don't need a content merge */
4234 if (ci
->match_mask
|| ci
->filemask
< 6 ||
4235 !S_ISREG(ci
->stages
[1].mode
) ||
4236 !S_ISREG(ci
->stages
[2].mode
) ||
4237 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
4240 /* Also don't need content merge if base matches either side */
4241 if (ci
->filemask
== 7 &&
4242 S_ISREG(ci
->stages
[0].mode
) &&
4243 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
4244 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
4247 for (i
= 0; i
< 3; i
++) {
4248 unsigned side_mask
= (1 << i
);
4249 struct version_info
*vi
= &ci
->stages
[i
];
4251 if ((ci
->filemask
& side_mask
) &&
4252 S_ISREG(vi
->mode
) &&
4253 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
4254 OBJECT_INFO_FOR_PREFETCH
))
4255 oid_array_append(&to_fetch
, &vi
->oid
);
4259 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
4260 oid_array_clear(&to_fetch
);
4263 static int process_entries(struct merge_options
*opt
,
4264 struct object_id
*result_oid
)
4266 struct hashmap_iter iter
;
4267 struct strmap_entry
*e
;
4268 struct string_list plist
= STRING_LIST_INIT_NODUP
;
4269 struct string_list_item
*entry
;
4270 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
4271 STRING_LIST_INIT_NODUP
,
4275 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
4276 if (strmap_empty(&opt
->priv
->paths
)) {
4277 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
4281 /* Hack to pre-allocate plist to the desired size */
4282 trace2_region_enter("merge", "plist grow", opt
->repo
);
4283 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
4284 trace2_region_leave("merge", "plist grow", opt
->repo
);
4286 /* Put every entry from paths into plist, then sort */
4287 trace2_region_enter("merge", "plist copy", opt
->repo
);
4288 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
4289 string_list_append(&plist
, e
->key
)->util
= e
->value
;
4291 trace2_region_leave("merge", "plist copy", opt
->repo
);
4293 trace2_region_enter("merge", "plist special sort", opt
->repo
);
4294 plist
.cmp
= sort_dirs_next_to_their_children
;
4295 string_list_sort(&plist
);
4296 trace2_region_leave("merge", "plist special sort", opt
->repo
);
4298 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
4301 * Iterate over the items in reverse order, so we can handle paths
4302 * below a directory before needing to handle the directory itself.
4304 * This allows us to write subtrees before we need to write trees,
4305 * and it also enables sane handling of directory/file conflicts
4306 * (because it allows us to know whether the directory is still in
4307 * the way when it is time to process the file at the same path).
4309 trace2_region_enter("merge", "processing", opt
->repo
);
4310 prefetch_for_content_merges(opt
, &plist
);
4311 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4312 char *path
= entry
->string
;
4314 * NOTE: mi may actually be a pointer to a conflict_info, but
4315 * we have to check mi->clean first to see if it's safe to
4316 * reassign to such a pointer type.
4318 struct merged_info
*mi
= entry
->util
;
4320 if (write_completed_directory(opt
, mi
->directory_name
,
4321 &dir_metadata
) < 0) {
4326 record_entry_for_tree(&dir_metadata
, path
, mi
);
4328 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4329 if (process_entry(opt
, path
, ci
, &dir_metadata
) < 0) {
4335 trace2_region_leave("merge", "processing", opt
->repo
);
4337 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4338 if (dir_metadata
.offsets
.nr
!= 1 ||
4339 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4340 printf("dir_metadata.offsets.nr = %"PRIuMAX
" (should be 1)\n",
4341 (uintmax_t)dir_metadata
.offsets
.nr
);
4342 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4343 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4345 BUG("dir_metadata accounting completely off; shouldn't happen");
4347 if (write_tree(result_oid
, &dir_metadata
.versions
, 0,
4348 opt
->repo
->hash_algo
->rawsz
) < 0)
4351 string_list_clear(&plist
, 0);
4352 string_list_clear(&dir_metadata
.versions
, 0);
4353 string_list_clear(&dir_metadata
.offsets
, 0);
4354 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4359 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4361 static int checkout(struct merge_options
*opt
,
4365 /* Switch the index/working copy from old to new */
4367 struct tree_desc trees
[2];
4368 struct unpack_trees_options unpack_opts
;
4370 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4371 unpack_opts
.head_idx
= -1;
4372 unpack_opts
.src_index
= opt
->repo
->index
;
4373 unpack_opts
.dst_index
= opt
->repo
->index
;
4375 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4378 * NOTE: if this were just "git checkout" code, we would probably
4379 * read or refresh the cache and check for a conflicted index, but
4380 * builtin/merge.c or sequencer.c really needs to read the index
4381 * and check for conflicted entries before starting merging for a
4382 * good user experience (no sense waiting for merges/rebases before
4383 * erroring out), so there's no reason to duplicate that work here.
4386 /* 2-way merge to the new branch */
4387 unpack_opts
.update
= 1;
4388 unpack_opts
.merge
= 1;
4389 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4390 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4391 unpack_opts
.fn
= twoway_merge
;
4392 unpack_opts
.preserve_ignored
= 0; /* FIXME: !opts->overwrite_ignore */
4394 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4396 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4398 ret
= unpack_trees(2, trees
, &unpack_opts
);
4399 clear_unpack_trees_porcelain(&unpack_opts
);
4403 static int record_conflicted_index_entries(struct merge_options
*opt
)
4405 struct hashmap_iter iter
;
4406 struct strmap_entry
*e
;
4407 struct index_state
*index
= opt
->repo
->index
;
4408 struct checkout state
= CHECKOUT_INIT
;
4410 int original_cache_nr
;
4412 if (strmap_empty(&opt
->priv
->conflicted
))
4416 * We are in a conflicted state. These conflicts might be inside
4417 * sparse-directory entries, so check if any entries are outside
4418 * of the sparse-checkout cone preemptively.
4420 * We set original_cache_nr below, but that might change if
4421 * index_name_pos() calls ask for paths within sparse directories.
4423 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4424 if (!path_in_sparse_checkout(e
->key
, index
)) {
4425 ensure_full_index(index
);
4430 /* If any entries have skip_worktree set, we'll have to check 'em out */
4433 state
.refresh_cache
= 1;
4434 state
.istate
= index
;
4435 original_cache_nr
= index
->cache_nr
;
4437 /* Append every entry from conflicted into index, then sort */
4438 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4439 const char *path
= e
->key
;
4440 struct conflict_info
*ci
= e
->value
;
4442 struct cache_entry
*ce
;
4448 * The index will already have a stage=0 entry for this path,
4449 * because we created an as-merged-as-possible version of the
4450 * file and checkout() moved the working copy and index over
4453 * However, previous iterations through this loop will have
4454 * added unstaged entries to the end of the cache which
4455 * ignore the standard alphabetical ordering of cache
4456 * entries and break invariants needed for index_name_pos()
4457 * to work. However, we know the entry we want is before
4458 * those appended cache entries, so do a temporary swap on
4459 * cache_nr to only look through entries of interest.
4461 SWAP(index
->cache_nr
, original_cache_nr
);
4462 pos
= index_name_pos(index
, path
, strlen(path
));
4463 SWAP(index
->cache_nr
, original_cache_nr
);
4465 if (ci
->filemask
!= 1)
4466 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4467 cache_tree_invalidate_path(index
, path
);
4469 ce
= index
->cache
[pos
];
4472 * Clean paths with CE_SKIP_WORKTREE set will not be
4473 * written to the working tree by the unpack_trees()
4474 * call in checkout(). Our conflicted entries would
4475 * have appeared clean to that code since we ignored
4476 * the higher order stages. Thus, we need override
4477 * the CE_SKIP_WORKTREE bit and manually write those
4478 * files to the working disk here.
4480 if (ce_skip_worktree(ce
))
4481 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4484 * Mark this cache entry for removal and instead add
4485 * new stage>0 entries corresponding to the
4486 * conflicts. If there are many conflicted entries, we
4487 * want to avoid memmove'ing O(NM) entries by
4488 * inserting the new entries one at a time. So,
4489 * instead, we just add the new cache entries to the
4490 * end (ignoring normal index requirements on sort
4491 * order) and sort the index once we're all done.
4493 ce
->ce_flags
|= CE_REMOVE
;
4496 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4497 struct version_info
*vi
;
4498 if (!(ci
->filemask
& (1ul << i
)))
4500 vi
= &ci
->stages
[i
];
4501 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4503 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4508 * Remove the unused cache entries (and invalidate the relevant
4509 * cache-trees), then sort the index entries to get the conflicted
4510 * entries we added to the end into their right locations.
4512 remove_marked_cache_entries(index
, 1);
4514 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4515 * on filename and secondarily on stage, and (name, stage #) are a
4518 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4523 static void print_submodule_conflict_suggestion(struct string_list
*csub
) {
4524 struct string_list_item
*item
;
4525 struct strbuf msg
= STRBUF_INIT
;
4526 struct strbuf tmp
= STRBUF_INIT
;
4527 struct strbuf subs
= STRBUF_INIT
;
4532 strbuf_add_separated_string_list(&subs
, " ", csub
);
4533 for_each_string_list_item(item
, csub
) {
4534 struct conflicted_submodule_item
*util
= item
->util
;
4537 * NEEDSWORK: The steps to resolve these errors deserve a more
4538 * detailed explanation than what is currently printed below.
4540 if (util
->flag
== CONFLICT_SUBMODULE_NOT_INITIALIZED
||
4541 util
->flag
== CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
)
4545 * TRANSLATORS: This is a line of advice to resolve a merge
4546 * conflict in a submodule. The first argument is the submodule
4547 * name, and the second argument is the abbreviated id of the
4548 * commit that needs to be merged. For example:
4549 * - go to submodule (mysubmodule), and either merge commit abc1234"
4551 strbuf_addf(&tmp
, _(" - go to submodule (%s), and either merge commit %s\n"
4552 " or update to an existing commit which has merged those changes\n"),
4553 item
->string
, util
->abbrev
);
4557 * TRANSLATORS: This is a detailed message for resolving submodule
4558 * conflicts. The first argument is string containing one step per
4559 * submodule. The second is a space-separated list of submodule names.
4562 _("Recursive merging with submodules currently only supports trivial cases.\n"
4563 "Please manually handle the merging of each conflicted submodule.\n"
4564 "This can be accomplished with the following steps:\n"
4566 " - come back to superproject and run:\n\n"
4568 " to record the above merge or update\n"
4569 " - resolve any other conflicts in the superproject\n"
4570 " - commit the resulting index in the superproject\n"),
4573 printf("%s", msg
.buf
);
4575 strbuf_release(&subs
);
4576 strbuf_release(&tmp
);
4577 strbuf_release(&msg
);
4580 void merge_display_update_messages(struct merge_options
*opt
,
4582 struct merge_result
*result
)
4584 struct merge_options_internal
*opti
= result
->priv
;
4585 struct hashmap_iter iter
;
4586 struct strmap_entry
*e
;
4587 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4589 if (opt
->record_conflict_msgs_as_headers
)
4590 BUG("Either display conflict messages or record them as headers, not both");
4592 trace2_region_enter("merge", "display messages", opt
->repo
);
4594 /* Hack to pre-allocate olist to the desired size */
4595 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->conflicts
),
4598 /* Put every entry from output into olist, then sort */
4599 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
4600 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4602 string_list_sort(&olist
);
4604 /* Iterate over the items, printing them */
4605 for (int path_nr
= 0; path_nr
< olist
.nr
; ++path_nr
) {
4606 struct string_list
*conflicts
= olist
.items
[path_nr
].util
;
4607 for (int i
= 0; i
< conflicts
->nr
; i
++) {
4608 struct logical_conflict_info
*info
=
4609 conflicts
->items
[i
].util
;
4612 printf("%lu", (unsigned long)info
->paths
.nr
);
4614 for (int n
= 0; n
< info
->paths
.nr
; n
++) {
4615 fputs(info
->paths
.v
[n
], stdout
);
4618 fputs(type_short_descriptions
[info
->type
],
4622 puts(conflicts
->items
[i
].string
);
4627 string_list_clear(&olist
, 0);
4629 print_submodule_conflict_suggestion(&opti
->conflicted_submodules
);
4631 /* Also include needed rename limit adjustment now */
4632 diff_warn_rename_limit("merge.renamelimit",
4633 opti
->renames
.needed_limit
, 0);
4635 trace2_region_leave("merge", "display messages", opt
->repo
);
4638 void merge_get_conflicted_files(struct merge_result
*result
,
4639 struct string_list
*conflicted_files
)
4641 struct hashmap_iter iter
;
4642 struct strmap_entry
*e
;
4643 struct merge_options_internal
*opti
= result
->priv
;
4645 strmap_for_each_entry(&opti
->conflicted
, &iter
, e
) {
4646 const char *path
= e
->key
;
4647 struct conflict_info
*ci
= e
->value
;
4652 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4653 struct stage_info
*si
;
4655 if (!(ci
->filemask
& (1ul << i
)))
4658 si
= xmalloc(sizeof(*si
));
4660 si
->mode
= ci
->stages
[i
].mode
;
4661 oidcpy(&si
->oid
, &ci
->stages
[i
].oid
);
4662 string_list_append(conflicted_files
, path
)->util
= si
;
4665 /* string_list_sort() uses a stable sort, so we're good */
4666 string_list_sort(conflicted_files
);
4669 void merge_switch_to_result(struct merge_options
*opt
,
4671 struct merge_result
*result
,
4672 int update_worktree_and_index
,
4673 int display_update_msgs
)
4675 assert(opt
->priv
== NULL
);
4676 if (result
->clean
>= 0 && update_worktree_and_index
) {
4677 const char *filename
;
4680 trace2_region_enter("merge", "checkout", opt
->repo
);
4681 if (checkout(opt
, head
, result
->tree
)) {
4682 /* failure to function */
4684 merge_finalize(opt
, result
);
4685 trace2_region_leave("merge", "checkout", opt
->repo
);
4688 trace2_region_leave("merge", "checkout", opt
->repo
);
4690 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4691 opt
->priv
= result
->priv
;
4692 if (record_conflicted_index_entries(opt
)) {
4693 /* failure to function */
4696 merge_finalize(opt
, result
);
4697 trace2_region_leave("merge", "record_conflicted",
4702 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4704 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4705 filename
= git_path_auto_merge(opt
->repo
);
4706 fp
= xfopen(filename
, "w");
4707 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4709 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4711 if (display_update_msgs
)
4712 merge_display_update_messages(opt
, /* detailed */ 0, result
);
4714 merge_finalize(opt
, result
);
4717 void merge_finalize(struct merge_options
*opt
,
4718 struct merge_result
*result
)
4720 struct merge_options_internal
*opti
= result
->priv
;
4722 if (opt
->renormalize
)
4723 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4724 assert(opt
->priv
== NULL
);
4726 clear_or_reinit_internal_opts(opti
, 0);
4727 FREE_AND_NULL(opti
);
4730 /*** Function Grouping: helper functions for merge_incore_*() ***/
4732 static struct tree
*shift_tree_object(struct repository
*repo
,
4733 struct tree
*one
, struct tree
*two
,
4734 const char *subtree_shift
)
4736 struct object_id shifted
;
4738 if (!*subtree_shift
) {
4739 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4741 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4744 if (oideq(&two
->object
.oid
, &shifted
))
4746 return lookup_tree(repo
, &shifted
);
4749 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4754 static struct commit
*make_virtual_commit(struct repository
*repo
,
4756 const char *comment
)
4758 struct commit
*commit
= alloc_commit_node(repo
);
4760 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4761 set_commit_tree(commit
, tree
);
4762 commit
->object
.parsed
= 1;
4766 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4768 struct rename_info
*renames
;
4770 struct mem_pool
*pool
= NULL
;
4772 /* Sanity checks on opt */
4773 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4776 assert(opt
->branch1
&& opt
->branch2
);
4778 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4779 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4780 assert(opt
->rename_limit
>= -1);
4781 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4782 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4784 assert(opt
->xdl_opts
>= 0);
4785 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4786 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4788 if (opt
->msg_header_prefix
)
4789 assert(opt
->record_conflict_msgs_as_headers
);
4792 * detect_renames, verbosity, buffer_output, and obuf are ignored
4793 * fields that were used by "recursive" rather than "ort" -- but
4794 * sanity check them anyway.
4796 assert(opt
->detect_renames
>= -1 &&
4797 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4798 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4799 assert(opt
->buffer_output
<= 2);
4800 assert(opt
->obuf
.len
== 0);
4802 assert(opt
->priv
== NULL
);
4803 if (result
->_properly_initialized
!= 0 &&
4804 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4805 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4806 assert(!!result
->priv
== !!result
->_properly_initialized
);
4808 opt
->priv
= result
->priv
;
4809 result
->priv
= NULL
;
4811 * opt->priv non-NULL means we had results from a previous
4812 * run; do a few sanity checks that user didn't mess with
4813 * it in an obvious fashion.
4815 assert(opt
->priv
->call_depth
== 0);
4816 assert(!opt
->priv
->toplevel_dir
||
4817 0 == strlen(opt
->priv
->toplevel_dir
));
4819 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4821 /* Default to histogram diff. Actually, just hardcode it...for now. */
4822 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4824 /* Handle attr direction stuff for renormalization */
4825 if (opt
->renormalize
)
4826 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4828 /* Initialization of opt->priv, our internal merge data */
4829 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4831 clear_or_reinit_internal_opts(opt
->priv
, 1);
4832 string_list_init_nodup(&opt
->priv
->conflicted_submodules
);
4833 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4836 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4838 /* Initialization of various renames fields */
4839 renames
= &opt
->priv
->renames
;
4840 mem_pool_init(&opt
->priv
->pool
, 0);
4841 pool
= &opt
->priv
->pool
;
4842 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4843 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4844 NOT_RELEVANT
, pool
, 0);
4845 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4847 strmap_init_with_options(&renames
->dir_renames
[i
],
4850 * relevant_sources uses -1 for the default, because we need
4851 * to be able to distinguish not-in-strintmap from valid
4852 * relevant_source values from enum file_rename_relevance.
4853 * In particular, possibly_cache_new_pair() expects a negative
4854 * value for not-found entries.
4856 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4857 -1 /* explicitly invalid */,
4859 strmap_init_with_options(&renames
->cached_pairs
[i
],
4861 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4863 strset_init_with_options(&renames
->cached_target_names
[i
],
4866 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4867 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4869 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4871 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4875 * Although we initialize opt->priv->paths with strdup_strings=0,
4876 * that's just to avoid making yet another copy of an allocated
4877 * string. Putting the entry into paths means we are taking
4878 * ownership, so we will later free it.
4880 * In contrast, conflicted just has a subset of keys from paths, so
4881 * we don't want to free those (it'd be a duplicate free).
4883 strmap_init_with_options(&opt
->priv
->paths
, pool
, 0);
4884 strmap_init_with_options(&opt
->priv
->conflicted
, pool
, 0);
4887 * keys & string_lists in conflicts will sometimes need to outlive
4888 * "paths", so it will have a copy of relevant keys. It's probably
4889 * a small subset of the overall paths that have special output.
4891 strmap_init(&opt
->priv
->conflicts
);
4893 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4896 static void merge_check_renames_reusable(struct merge_options
*opt
,
4897 struct merge_result
*result
,
4898 struct tree
*merge_base
,
4902 struct rename_info
*renames
;
4903 struct tree
**merge_trees
;
4904 struct merge_options_internal
*opti
= result
->priv
;
4909 renames
= &opti
->renames
;
4910 merge_trees
= renames
->merge_trees
;
4913 * Handle case where previous merge operation did not want cache to
4914 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4916 if (!merge_trees
[0]) {
4917 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4918 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4923 * Handle other cases; note that merge_trees[0..2] will only
4924 * be NULL if opti is, or if all three were manually set to
4925 * NULL by e.g. rename/rename(1to1) handling.
4927 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4929 /* Check if we meet a condition for re-using cached_pairs */
4930 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4931 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4932 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4933 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4934 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4935 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4937 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4940 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4943 * Originally from merge_trees_internal(); heavily adapted, though.
4945 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4946 struct tree
*merge_base
,
4949 struct merge_result
*result
)
4951 struct object_id working_tree_oid
;
4953 if (opt
->subtree_shift
) {
4954 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4955 opt
->subtree_shift
);
4956 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4957 opt
->subtree_shift
);
4961 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4962 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4964 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4965 * base, and 2-3) the trees for the two trees we're merging.
4967 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
4968 oid_to_hex(&merge_base
->object
.oid
),
4969 oid_to_hex(&side1
->object
.oid
),
4970 oid_to_hex(&side2
->object
.oid
));
4974 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4976 trace2_region_enter("merge", "renames", opt
->repo
);
4977 result
->clean
= detect_and_process_renames(opt
, merge_base
,
4979 trace2_region_leave("merge", "renames", opt
->repo
);
4980 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4981 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4982 clear_or_reinit_internal_opts(opt
->priv
, 1);
4983 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4987 trace2_region_enter("merge", "process_entries", opt
->repo
);
4988 if (process_entries(opt
, &working_tree_oid
) < 0)
4990 trace2_region_leave("merge", "process_entries", opt
->repo
);
4992 /* Set return values */
4993 result
->path_messages
= &opt
->priv
->conflicts
;
4995 if (result
->clean
>= 0) {
4996 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4997 /* existence of conflicted entries implies unclean */
4998 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
5000 if (!opt
->priv
->call_depth
) {
5001 result
->priv
= opt
->priv
;
5002 result
->_properly_initialized
= RESULT_INITIALIZED
;
5008 * Originally from merge_recursive_internal(); somewhat adapted, though.
5010 static void merge_ort_internal(struct merge_options
*opt
,
5011 struct commit_list
*merge_bases
,
5014 struct merge_result
*result
)
5016 struct commit
*next
;
5017 struct commit
*merged_merge_bases
;
5018 const char *ancestor_name
;
5019 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
5022 merge_bases
= get_merge_bases(h1
, h2
);
5023 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
5024 merge_bases
= reverse_commit_list(merge_bases
);
5027 merged_merge_bases
= pop_commit(&merge_bases
);
5028 if (!merged_merge_bases
) {
5029 /* if there is no common ancestor, use an empty tree */
5032 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
5033 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
5035 ancestor_name
= "empty tree";
5036 } else if (merge_bases
) {
5037 ancestor_name
= "merged common ancestors";
5039 strbuf_add_unique_abbrev(&merge_base_abbrev
,
5040 &merged_merge_bases
->object
.oid
,
5042 ancestor_name
= merge_base_abbrev
.buf
;
5045 for (next
= pop_commit(&merge_bases
); next
;
5046 next
= pop_commit(&merge_bases
)) {
5047 const char *saved_b1
, *saved_b2
;
5048 struct commit
*prev
= merged_merge_bases
;
5050 opt
->priv
->call_depth
++;
5052 * When the merge fails, the result contains files
5053 * with conflict markers. The cleanness flag is
5054 * ignored (unless indicating an error), it was never
5055 * actually used, as result of merge_trees has always
5056 * overwritten it: the committed "conflicts" were
5059 saved_b1
= opt
->branch1
;
5060 saved_b2
= opt
->branch2
;
5061 opt
->branch1
= "Temporary merge branch 1";
5062 opt
->branch2
= "Temporary merge branch 2";
5063 merge_ort_internal(opt
, NULL
, prev
, next
, result
);
5064 if (result
->clean
< 0)
5066 opt
->branch1
= saved_b1
;
5067 opt
->branch2
= saved_b2
;
5068 opt
->priv
->call_depth
--;
5070 merged_merge_bases
= make_virtual_commit(opt
->repo
,
5073 commit_list_insert(prev
, &merged_merge_bases
->parents
);
5074 commit_list_insert(next
, &merged_merge_bases
->parents
->next
);
5076 clear_or_reinit_internal_opts(opt
->priv
, 1);
5079 opt
->ancestor
= ancestor_name
;
5080 merge_ort_nonrecursive_internal(opt
,
5081 repo_get_commit_tree(opt
->repo
,
5082 merged_merge_bases
),
5083 repo_get_commit_tree(opt
->repo
, h1
),
5084 repo_get_commit_tree(opt
->repo
, h2
),
5086 strbuf_release(&merge_base_abbrev
);
5087 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
5090 void merge_incore_nonrecursive(struct merge_options
*opt
,
5091 struct tree
*merge_base
,
5094 struct merge_result
*result
)
5096 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
5098 trace2_region_enter("merge", "merge_start", opt
->repo
);
5099 assert(opt
->ancestor
!= NULL
);
5100 merge_check_renames_reusable(opt
, result
, merge_base
, side1
, side2
);
5101 merge_start(opt
, result
);
5103 * Record the trees used in this merge, so if there's a next merge in
5104 * a cherry-pick or rebase sequence it might be able to take advantage
5105 * of the cached_pairs in that next merge.
5107 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
5108 opt
->priv
->renames
.merge_trees
[1] = side1
;
5109 opt
->priv
->renames
.merge_trees
[2] = side2
;
5110 trace2_region_leave("merge", "merge_start", opt
->repo
);
5112 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
5113 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
5116 void merge_incore_recursive(struct merge_options
*opt
,
5117 struct commit_list
*merge_bases
,
5118 struct commit
*side1
,
5119 struct commit
*side2
,
5120 struct merge_result
*result
)
5122 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
5124 /* We set the ancestor label based on the merge_bases */
5125 assert(opt
->ancestor
== NULL
);
5127 trace2_region_enter("merge", "merge_start", opt
->repo
);
5128 merge_start(opt
, result
);
5129 trace2_region_leave("merge", "merge_start", opt
->repo
);
5131 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
5132 trace2_region_leave("merge", "incore_recursive", opt
->repo
);