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"
29 #include "environment.h"
34 #include "object-store.h"
35 #include "promisor-remote.h"
38 #include "submodule-config.h"
39 #include "submodule.h"
42 #include "unpack-trees.h"
43 #include "xdiff-interface.h"
46 * We have many arrays of size 3. Whenever we have such an array, the
47 * indices refer to one of the sides of the three-way merge. This is so
48 * pervasive that the constants 0, 1, and 2 are used in many places in the
49 * code (especially in arithmetic operations to find the other side's index
50 * or to compute a relevant mask), but sometimes these enum names are used
51 * to aid code clarity.
53 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
54 * referred to there is one of these three sides.
62 static unsigned RESULT_INITIALIZED
= 0x1abe11ed; /* unlikely accidental value */
64 struct traversal_callback_data
{
66 unsigned long dirmask
;
67 struct name_entry names
[3];
70 struct deferred_traversal_data
{
72 * possible_trivial_merges: directories to be explored only when needed
74 * possible_trivial_merges is a map of directory names to
75 * dir_rename_mask. When we detect that a directory is unchanged on
76 * one side, we can sometimes resolve the directory without recursing
77 * into it. Renames are the only things that can prevent such an
78 * optimization. However, for rename sources:
79 * - If no parent directory needed directory rename detection, then
80 * no path under such a directory can be a relevant_source.
81 * and for rename destinations:
82 * - If no cached rename has a target path under the directory AND
83 * - If there are no unpaired relevant_sources elsewhere in the
85 * then we don't need any path under this directory for a rename
86 * destination. The only way to know the last item above is to defer
87 * handling such directories until the end of collect_merge_info(),
88 * in handle_deferred_entries().
90 * For each we store dir_rename_mask, since that's the only bit of
91 * information we need, other than the path, to resume the recursive
94 struct strintmap possible_trivial_merges
;
97 * trivial_merges_okay: if trivial directory merges are okay
99 * See possible_trivial_merges above. The "no unpaired
100 * relevant_sources elsewhere in the repository" is a single boolean
101 * per merge side, which we store here. Note that while 0 means no,
102 * 1 only means "maybe" rather than "yes"; we optimistically set it
103 * to 1 initially and only clear when we determine it is unsafe to
104 * do trivial directory merges.
106 unsigned trivial_merges_okay
;
109 * target_dirs: ancestor directories of rename targets
111 * target_dirs contains all directory names that are an ancestor of
112 * any rename destination.
114 struct strset target_dirs
;
119 * All variables that are arrays of size 3 correspond to data tracked
120 * for the sides in enum merge_side. Index 0 is almost always unused
121 * because we often only need to track information for MERGE_SIDE1 and
122 * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames
123 * are determined relative to what changed since the MERGE_BASE).
127 * pairs: pairing of filenames from diffcore_rename()
129 struct diff_queue_struct pairs
[3];
132 * dirs_removed: directories removed on a given side of history.
134 * The keys of dirs_removed[side] are the directories that were removed
135 * on the given side of history. The value of the strintmap for each
136 * directory is a value from enum dir_rename_relevance.
138 struct strintmap dirs_removed
[3];
141 * dir_rename_count: tracking where parts of a directory were renamed to
143 * When files in a directory are renamed, they may not all go to the
144 * same location. Each strmap here tracks:
145 * old_dir => {new_dir => int}
146 * That is, dir_rename_count[side] is a strmap to a strintmap.
148 struct strmap dir_rename_count
[3];
151 * dir_renames: computed directory renames
153 * This is a map of old_dir => new_dir and is derived in part from
156 struct strmap dir_renames
[3];
159 * relevant_sources: deleted paths wanted in rename detection, and why
161 * relevant_sources is a set of deleted paths on each side of
162 * history for which we need rename detection. If a path is deleted
163 * on one side of history, we need to detect if it is part of a
165 * * the file is modified/deleted on the other side of history
166 * * we need to detect renames for an ancestor directory
167 * If neither of those are true, we can skip rename detection for
168 * that path. The reason is stored as a value from enum
169 * file_rename_relevance, as the reason can inform the algorithm in
170 * diffcore_rename_extended().
172 struct strintmap relevant_sources
[3];
174 struct deferred_traversal_data deferred
[3];
178 * 0: optimization removing unmodified potential rename source okay
179 * 2 or 4: optimization okay, but must check for files added to dir
180 * 7: optimization forbidden; need rename source in case of dir rename
182 unsigned dir_rename_mask
:3;
185 * callback_data_*: supporting data structures for alternate traversal
187 * We sometimes need to be able to traverse through all the files
188 * in a given tree before all immediate subdirectories within that
189 * tree. Since traverse_trees() doesn't do that naturally, we have
190 * a traverse_trees_wrapper() that stores any immediate
191 * subdirectories while traversing files, then traverses the
192 * immediate subdirectories later. These callback_data* variables
193 * store the information for the subdirectories so that we can do
194 * that traversal order.
196 struct traversal_callback_data
*callback_data
;
197 int callback_data_nr
, callback_data_alloc
;
198 char *callback_data_traverse_path
;
201 * merge_trees: trees passed to the merge algorithm for the merge
203 * merge_trees records the trees passed to the merge algorithm. But,
204 * this data also is stored in merge_result->priv. If a sequence of
205 * merges are being done (such as when cherry-picking or rebasing),
206 * the next merge can look at this and re-use information from
207 * previous merges under certain circumstances.
209 * See also all the cached_* variables.
211 struct tree
*merge_trees
[3];
214 * cached_pairs_valid_side: which side's cached info can be reused
216 * See the description for merge_trees. For repeated merges, at most
217 * only one side's cached information can be used. Valid values:
218 * MERGE_SIDE2: cached data from side2 can be reused
219 * MERGE_SIDE1: cached data from side1 can be reused
220 * 0: no cached data can be reused
221 * -1: See redo_after_renames; both sides can be reused.
223 int cached_pairs_valid_side
;
226 * cached_pairs: Caching of renames and deletions.
228 * These are mappings recording renames and deletions of individual
229 * files (not directories). They are thus a map from an old
230 * filename to either NULL (for deletions) or a new filename (for
233 struct strmap cached_pairs
[3];
236 * cached_target_names: just the destinations from cached_pairs
238 * We sometimes want a fast lookup to determine if a given filename
239 * is one of the destinations in cached_pairs. cached_target_names
240 * is thus duplicative information, but it provides a fast lookup.
242 struct strset cached_target_names
[3];
245 * cached_irrelevant: Caching of rename_sources that aren't relevant.
247 * If we try to detect a rename for a source path and succeed, it's
248 * part of a rename. If we try to detect a rename for a source path
249 * and fail, then it's a delete. If we do not try to detect a rename
250 * for a path, then we don't know if it's a rename or a delete. If
251 * merge-ort doesn't think the path is relevant, then we just won't
252 * cache anything for that path. But there's a slight problem in
253 * that merge-ort can think a path is RELEVANT_LOCATION, but due to
254 * commit 9bd342137e ("diffcore-rename: determine which
255 * relevant_sources are no longer relevant", 2021-03-13),
256 * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To
257 * avoid excessive calls to diffcore_rename_extended() we still need
258 * to cache such paths, though we cannot record them as either
259 * renames or deletes. So we cache them here as a "turned out to be
260 * irrelevant *for this commit*" as they are often also irrelevant
261 * for subsequent commits, though we will have to do some extra
262 * checking to see whether such paths become relevant for rename
263 * detection when cherry-picking/rebasing subsequent commits.
265 struct strset cached_irrelevant
[3];
268 * redo_after_renames: optimization flag for "restarting" the merge
270 * Sometimes it pays to detect renames, cache them, and then
271 * restart the merge operation from the beginning. The reason for
272 * this is that when we know where all the renames are, we know
273 * whether a certain directory has any paths under it affected --
274 * and if a directory is not affected then it permits us to do
275 * trivial tree merging in more cases. Doing trivial tree merging
276 * prevents the need to run process_entry() on every path
277 * underneath trees that can be trivially merged, and
278 * process_entry() is more expensive than collect_merge_info() --
279 * plus, the second collect_merge_info() will be much faster since
280 * it doesn't have to recurse into the relevant trees.
282 * Values for this flag:
283 * 0 = don't bother, not worth it (or conditions not yet checked)
284 * 1 = conditions for optimization met, optimization worthwhile
285 * 2 = we already did it (don't restart merge yet again)
287 unsigned redo_after_renames
;
290 * needed_limit: value needed for inexact rename detection to run
292 * If the current rename limit wasn't high enough for inexact
293 * rename detection to run, this records the limit needed. Otherwise,
294 * this value remains 0.
299 struct merge_options_internal
{
301 * paths: primary data structure in all of merge ort.
304 * * are full relative paths from the toplevel of the repository
305 * (e.g. "drivers/firmware/raspberrypi.c").
306 * * store all relevant paths in the repo, both directories and
307 * files (e.g. drivers, drivers/firmware would also be included)
308 * * these keys serve to intern all the path strings, which allows
309 * us to do pointer comparison on directory names instead of
310 * strcmp; we just have to be careful to use the interned strings.
312 * The values of paths:
313 * * either a pointer to a merged_info, or a conflict_info struct
314 * * merged_info contains all relevant information for a
315 * non-conflicted entry.
316 * * conflict_info contains a merged_info, plus any additional
317 * information about a conflict such as the higher orders stages
318 * involved and the names of the paths those came from (handy
319 * once renames get involved).
320 * * a path may start "conflicted" (i.e. point to a conflict_info)
321 * and then a later step (e.g. three-way content merge) determines
322 * it can be cleanly merged, at which point it'll be marked clean
323 * and the algorithm will ignore any data outside the contained
324 * merged_info for that entry
325 * * If an entry remains conflicted, the merged_info portion of a
326 * conflict_info will later be filled with whatever version of
327 * the file should be placed in the working directory (e.g. an
328 * as-merged-as-possible variation that contains conflict markers).
333 * conflicted: a subset of keys->values from "paths"
335 * conflicted is basically an optimization between process_entries()
336 * and record_conflicted_index_entries(); the latter could loop over
337 * ALL the entries in paths AGAIN and look for the ones that are
338 * still conflicted, but since process_entries() has to loop over
339 * all of them, it saves the ones it couldn't resolve in this strmap
340 * so that record_conflicted_index_entries() can iterate just the
343 struct strmap conflicted
;
346 * pool: memory pool for fast allocation/deallocation
348 * We allocate room for lots of filenames and auxiliary data
349 * structures in merge_options_internal, and it tends to all be
350 * freed together too. Using a memory pool for these provides a
353 struct mem_pool pool
;
356 * conflicts: logical conflicts and messages stored by _primary_ path
358 * This is a map of pathnames (a subset of the keys in "paths" above)
359 * to struct string_list, with each item's `util` containing a
360 * `struct logical_conflict_info`. Note, though, that for each path,
361 * it only stores the logical conflicts for which that path is the
362 * primary path; the path might be part of additional conflicts.
364 struct strmap conflicts
;
367 * renames: various data relating to rename detection
369 struct rename_info renames
;
372 * attr_index: hacky minimal index used for renormalization
374 * renormalization code _requires_ an index, though it only needs to
375 * find a .gitattributes file within the index. So, when
376 * renormalization is important, we create a special index with just
379 struct index_state attr_index
;
382 * current_dir_name, toplevel_dir: temporary vars
384 * These are used in collect_merge_info_callback(), and will set the
385 * various merged_info.directory_name for the various paths we get;
386 * see documentation for that variable and the requirements placed on
389 const char *current_dir_name
;
390 const char *toplevel_dir
;
392 /* call_depth: recursion level counter for merging merge bases */
395 /* field that holds submodule conflict information */
396 struct string_list conflicted_submodules
;
399 struct conflicted_submodule_item
{
404 static void conflicted_submodule_item_free(void *util
, const char *str UNUSED
)
406 struct conflicted_submodule_item
*item
= util
;
412 struct version_info
{
413 struct object_id oid
;
418 /* if is_null, ignore result. otherwise result has oid & mode */
419 struct version_info result
;
423 * clean: whether the path in question is cleanly merged.
425 * see conflict_info.merged for more details.
430 * basename_offset: offset of basename of path.
432 * perf optimization to avoid recomputing offset of final '/'
433 * character in pathname (0 if no '/' in pathname).
435 size_t basename_offset
;
438 * directory_name: containing directory name.
440 * Note that we assume directory_name is constructed such that
441 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
442 * i.e. string equality is equivalent to pointer equality. For this
443 * to hold, we have to be careful setting directory_name.
445 const char *directory_name
;
448 struct conflict_info
{
450 * merged: the version of the path that will be written to working tree
452 * WARNING: It is critical to check merged.clean and ensure it is 0
453 * before reading any conflict_info fields outside of merged.
454 * Allocated merge_info structs will always have clean set to 1.
455 * Allocated conflict_info structs will have merged.clean set to 0
456 * initially. The merged.clean field is how we know if it is safe
457 * to access other parts of conflict_info besides merged; if a
458 * conflict_info's merged.clean is changed to 1, the rest of the
459 * algorithm is not allowed to look at anything outside of the
460 * merged member anymore.
462 struct merged_info merged
;
464 /* oids & modes from each of the three trees for this path */
465 struct version_info stages
[3];
467 /* pathnames for each stage; may differ due to rename detection */
468 const char *pathnames
[3];
470 /* Whether this path is/was involved in a directory/file conflict */
471 unsigned df_conflict
:1;
474 * Whether this path is/was involved in a non-content conflict other
475 * than a directory/file conflict (e.g. rename/rename, rename/delete,
476 * file location based on possible directory rename).
478 unsigned path_conflict
:1;
481 * For filemask and dirmask, the ith bit corresponds to whether the
482 * ith entry is a file (filemask) or a directory (dirmask). Thus,
483 * filemask & dirmask is always zero, and filemask | dirmask is at
484 * most 7 but can be less when a path does not appear as either a
485 * file or a directory on at least one side of history.
487 * Note that these masks are related to enum merge_side, as the ith
488 * entry corresponds to side i.
490 * These values come from a traverse_trees() call; more info may be
491 * found looking at tree-walk.h's struct traverse_info,
492 * particularly the documentation above the "fn" member (note that
493 * filemask = mask & ~dirmask from that documentation).
499 * Optimization to track which stages match, to avoid the need to
500 * recompute it in multiple steps. Either 0 or at least 2 bits are
501 * set; if at least 2 bits are set, their corresponding stages match.
503 unsigned match_mask
:3;
506 enum conflict_and_info_types
{
507 /* "Simple" conflicts and informational messages */
508 INFO_AUTO_MERGING
= 0,
509 CONFLICT_CONTENTS
, /* text file that failed to merge */
511 CONFLICT_FILE_DIRECTORY
,
512 CONFLICT_DISTINCT_MODES
,
513 CONFLICT_MODIFY_DELETE
,
516 CONFLICT_RENAME_RENAME
, /* same file renamed differently */
517 CONFLICT_RENAME_COLLIDES
, /* rename/add or two files renamed to 1 */
518 CONFLICT_RENAME_DELETE
,
520 /* Basic directory rename */
521 CONFLICT_DIR_RENAME_SUGGESTED
,
522 INFO_DIR_RENAME_APPLIED
,
524 /* Special directory rename cases */
525 INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
,
526 CONFLICT_DIR_RENAME_FILE_IN_WAY
,
527 CONFLICT_DIR_RENAME_COLLISION
,
528 CONFLICT_DIR_RENAME_SPLIT
,
530 /* Basic submodule */
531 INFO_SUBMODULE_FAST_FORWARDING
,
532 CONFLICT_SUBMODULE_FAILED_TO_MERGE
,
534 /* Special submodule cases broken out from FAILED_TO_MERGE */
535 CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
,
536 CONFLICT_SUBMODULE_NOT_INITIALIZED
,
537 CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
,
538 CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
,
539 CONFLICT_SUBMODULE_NULL_MERGE_BASE
,
541 /* Keep this entry _last_ in the list */
546 * Short description of conflict type, relied upon by external tools.
548 * We can add more entries, but DO NOT change any of these strings. Also,
549 * Order MUST match conflict_info_and_types.
551 static const char *type_short_descriptions
[] = {
552 /*** "Simple" conflicts and informational messages ***/
553 [INFO_AUTO_MERGING
] = "Auto-merging",
554 [CONFLICT_CONTENTS
] = "CONFLICT (contents)",
555 [CONFLICT_BINARY
] = "CONFLICT (binary)",
556 [CONFLICT_FILE_DIRECTORY
] = "CONFLICT (file/directory)",
557 [CONFLICT_DISTINCT_MODES
] = "CONFLICT (distinct modes)",
558 [CONFLICT_MODIFY_DELETE
] = "CONFLICT (modify/delete)",
560 /*** Regular rename ***/
561 [CONFLICT_RENAME_RENAME
] = "CONFLICT (rename/rename)",
562 [CONFLICT_RENAME_COLLIDES
] = "CONFLICT (rename involved in collision)",
563 [CONFLICT_RENAME_DELETE
] = "CONFLICT (rename/delete)",
565 /*** Basic directory rename ***/
566 [CONFLICT_DIR_RENAME_SUGGESTED
] =
567 "CONFLICT (directory rename suggested)",
568 [INFO_DIR_RENAME_APPLIED
] = "Path updated due to directory rename",
570 /*** Special directory rename cases ***/
571 [INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
] =
572 "Directory rename skipped since directory was renamed on both sides",
573 [CONFLICT_DIR_RENAME_FILE_IN_WAY
] =
574 "CONFLICT (file in way of directory rename)",
575 [CONFLICT_DIR_RENAME_COLLISION
] = "CONFLICT(directory rename collision)",
576 [CONFLICT_DIR_RENAME_SPLIT
] = "CONFLICT(directory rename unclear split)",
578 /*** Basic submodule ***/
579 [INFO_SUBMODULE_FAST_FORWARDING
] = "Fast forwarding submodule",
580 [CONFLICT_SUBMODULE_FAILED_TO_MERGE
] = "CONFLICT (submodule)",
582 /*** Special submodule cases broken out from FAILED_TO_MERGE ***/
583 [CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
] =
584 "CONFLICT (submodule with possible resolution)",
585 [CONFLICT_SUBMODULE_NOT_INITIALIZED
] =
586 "CONFLICT (submodule not initialized)",
587 [CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
] =
588 "CONFLICT (submodule history not available)",
589 [CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
] =
590 "CONFLICT (submodule may have rewinds)",
591 [CONFLICT_SUBMODULE_NULL_MERGE_BASE
] =
592 "CONFLICT (submodule lacks merge base)"
595 struct logical_conflict_info
{
596 enum conflict_and_info_types type
;
600 /*** Function Grouping: various utility functions ***/
603 * For the next three macros, see warning for conflict_info.merged.
605 * In each of the below, mi is a struct merged_info*, and ci was defined
606 * as a struct conflict_info* (but we need to verify ci isn't actually
607 * pointed at a struct merged_info*).
609 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
610 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
611 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
613 #define INITIALIZE_CI(ci, mi) do { \
614 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
616 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
617 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
618 (ci) = (struct conflict_info *)(mi); \
619 assert((ci) && !(mi)->clean); \
622 static void free_strmap_strings(struct strmap
*map
)
624 struct hashmap_iter iter
;
625 struct strmap_entry
*entry
;
627 strmap_for_each_entry(map
, &iter
, entry
) {
628 free((char*)entry
->key
);
632 static void clear_or_reinit_internal_opts(struct merge_options_internal
*opti
,
635 struct rename_info
*renames
= &opti
->renames
;
637 void (*strmap_clear_func
)(struct strmap
*, int) =
638 reinitialize
? strmap_partial_clear
: strmap_clear
;
639 void (*strintmap_clear_func
)(struct strintmap
*) =
640 reinitialize
? strintmap_partial_clear
: strintmap_clear
;
641 void (*strset_clear_func
)(struct strset
*) =
642 reinitialize
? strset_partial_clear
: strset_clear
;
644 strmap_clear_func(&opti
->paths
, 0);
647 * All keys and values in opti->conflicted are a subset of those in
648 * opti->paths. We don't want to deallocate anything twice, so we
649 * don't free the keys and we pass 0 for free_values.
651 strmap_clear_func(&opti
->conflicted
, 0);
653 if (opti
->attr_index
.cache_nr
) /* true iff opt->renormalize */
654 discard_index(&opti
->attr_index
);
656 /* Free memory used by various renames maps */
657 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
658 strintmap_clear_func(&renames
->dirs_removed
[i
]);
659 strmap_clear_func(&renames
->dir_renames
[i
], 0);
660 strintmap_clear_func(&renames
->relevant_sources
[i
]);
662 assert(renames
->cached_pairs_valid_side
== 0);
663 if (i
!= renames
->cached_pairs_valid_side
&&
664 -1 != renames
->cached_pairs_valid_side
) {
665 strset_clear_func(&renames
->cached_target_names
[i
]);
666 strmap_clear_func(&renames
->cached_pairs
[i
], 1);
667 strset_clear_func(&renames
->cached_irrelevant
[i
]);
668 partial_clear_dir_rename_count(&renames
->dir_rename_count
[i
]);
670 strmap_clear(&renames
->dir_rename_count
[i
], 1);
673 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; ++i
) {
674 strintmap_clear_func(&renames
->deferred
[i
].possible_trivial_merges
);
675 strset_clear_func(&renames
->deferred
[i
].target_dirs
);
676 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
678 renames
->cached_pairs_valid_side
= 0;
679 renames
->dir_rename_mask
= 0;
682 struct hashmap_iter iter
;
683 struct strmap_entry
*e
;
685 /* Release and free each strbuf found in output */
686 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
687 struct string_list
*list
= e
->value
;
688 for (int i
= 0; i
< list
->nr
; i
++) {
689 struct logical_conflict_info
*info
=
691 strvec_clear(&info
->paths
);
694 * While strictly speaking we don't need to
695 * free(conflicts) here because we could pass
696 * free_values=1 when calling strmap_clear() on
697 * opti->conflicts, that would require strmap_clear
698 * to do another strmap_for_each_entry() loop, so we
699 * just free it while we're iterating anyway.
701 string_list_clear(list
, 1);
704 strmap_clear(&opti
->conflicts
, 0);
707 mem_pool_discard(&opti
->pool
, 0);
709 string_list_clear_func(&opti
->conflicted_submodules
,
710 conflicted_submodule_item_free
);
712 /* Clean out callback_data as well. */
713 FREE_AND_NULL(renames
->callback_data
);
714 renames
->callback_data_nr
= renames
->callback_data_alloc
= 0;
717 __attribute__((format (printf
, 2, 3)))
718 static int err(struct merge_options
*opt
, const char *err
, ...)
721 struct strbuf sb
= STRBUF_INIT
;
723 strbuf_addstr(&sb
, "error: ");
724 va_start(params
, err
);
725 strbuf_vaddf(&sb
, err
, params
);
734 static void format_commit(struct strbuf
*sb
,
736 struct repository
*repo
,
737 struct commit
*commit
)
739 struct merge_remote_desc
*desc
;
740 struct pretty_print_context ctx
= {0};
741 ctx
.abbrev
= DEFAULT_ABBREV
;
743 strbuf_addchars(sb
, ' ', indent
);
744 desc
= merge_remote_util(commit
);
746 strbuf_addf(sb
, "virtual %s\n", desc
->name
);
750 repo_format_commit_message(repo
, commit
, "%h %s", sb
, &ctx
);
751 strbuf_addch(sb
, '\n');
754 __attribute__((format (printf
, 8, 9)))
755 static void path_msg(struct merge_options
*opt
,
756 enum conflict_and_info_types type
,
757 int omittable_hint
, /* skippable under --remerge-diff */
758 const char *primary_path
,
759 const char *other_path_1
, /* may be NULL */
760 const char *other_path_2
, /* may be NULL */
761 struct string_list
*other_paths
, /* may be NULL */
762 const char *fmt
, ...)
765 struct string_list
*path_conflicts
;
766 struct logical_conflict_info
*info
;
767 struct strbuf buf
= STRBUF_INIT
;
769 struct strbuf tmp
= STRBUF_INIT
;
772 assert(omittable_hint
==
773 !starts_with(type_short_descriptions
[type
], "CONFLICT") ||
774 type
== CONFLICT_DIR_RENAME_SUGGESTED
);
775 if (opt
->record_conflict_msgs_as_headers
&& omittable_hint
)
776 return; /* Do not record mere hints in headers */
777 if (opt
->priv
->call_depth
&& opt
->verbosity
< 5)
778 return; /* Ignore messages from inner merges */
780 /* Ensure path_conflicts (ptr to array of logical_conflict) allocated */
781 path_conflicts
= strmap_get(&opt
->priv
->conflicts
, primary_path
);
782 if (!path_conflicts
) {
783 path_conflicts
= xmalloc(sizeof(*path_conflicts
));
784 string_list_init_dup(path_conflicts
);
785 strmap_put(&opt
->priv
->conflicts
, primary_path
, path_conflicts
);
788 /* Add a logical_conflict at the end to store info from this call */
789 info
= xcalloc(1, sizeof(*info
));
791 strvec_init(&info
->paths
);
793 /* Handle the list of paths */
794 strvec_push(&info
->paths
, primary_path
);
796 strvec_push(&info
->paths
, other_path_1
);
798 strvec_push(&info
->paths
, other_path_2
);
800 for (int i
= 0; i
< other_paths
->nr
; i
++)
801 strvec_push(&info
->paths
, other_paths
->items
[i
].string
);
803 /* Handle message and its format, in normal case */
804 dest
= (opt
->record_conflict_msgs_as_headers
? &tmp
: &buf
);
807 if (opt
->priv
->call_depth
) {
808 strbuf_addchars(dest
, ' ', 2);
809 strbuf_addstr(dest
, "From inner merge:");
810 strbuf_addchars(dest
, ' ', opt
->priv
->call_depth
* 2);
812 strbuf_vaddf(dest
, fmt
, ap
);
815 /* Handle specialized formatting of message under --remerge-diff */
816 if (opt
->record_conflict_msgs_as_headers
) {
817 int i_sb
= 0, i_tmp
= 0;
819 /* Start with the specified prefix */
820 if (opt
->msg_header_prefix
)
821 strbuf_addf(&buf
, "%s ", opt
->msg_header_prefix
);
823 /* Copy tmp to sb, adding spaces after newlines */
824 strbuf_grow(&buf
, buf
.len
+ 2*tmp
.len
); /* more than sufficient */
825 for (; i_tmp
< tmp
.len
; i_tmp
++, i_sb
++) {
826 /* Copy next character from tmp to sb */
827 buf
.buf
[buf
.len
+ i_sb
] = tmp
.buf
[i_tmp
];
829 /* If we copied a newline, add a space */
830 if (tmp
.buf
[i_tmp
] == '\n')
831 buf
.buf
[++i_sb
] = ' ';
833 /* Update length and ensure it's NUL-terminated */
835 buf
.buf
[buf
.len
] = '\0';
837 strbuf_release(&tmp
);
839 string_list_append_nodup(path_conflicts
, strbuf_detach(&buf
, NULL
))
843 static struct diff_filespec
*pool_alloc_filespec(struct mem_pool
*pool
,
846 /* Similar to alloc_filespec(), but allocate from pool and reuse path */
847 struct diff_filespec
*spec
;
849 spec
= mem_pool_calloc(pool
, 1, sizeof(*spec
));
850 spec
->path
= (char*)path
; /* spec won't modify it */
853 spec
->is_binary
= -1;
857 static struct diff_filepair
*pool_diff_queue(struct mem_pool
*pool
,
858 struct diff_queue_struct
*queue
,
859 struct diff_filespec
*one
,
860 struct diff_filespec
*two
)
862 /* Same code as diff_queue(), except allocate from pool */
863 struct diff_filepair
*dp
;
865 dp
= mem_pool_calloc(pool
, 1, sizeof(*dp
));
873 /* add a string to a strbuf, but converting "/" to "_" */
874 static void add_flattened_path(struct strbuf
*out
, const char *s
)
877 strbuf_addstr(out
, s
);
878 for (; i
< out
->len
; i
++)
879 if (out
->buf
[i
] == '/')
883 static char *unique_path(struct merge_options
*opt
,
888 struct strbuf newpath
= STRBUF_INIT
;
891 struct strmap
*existing_paths
= &opt
->priv
->paths
;
893 strbuf_addf(&newpath
, "%s~", path
);
894 add_flattened_path(&newpath
, branch
);
896 base_len
= newpath
.len
;
897 while (strmap_contains(existing_paths
, newpath
.buf
)) {
898 strbuf_setlen(&newpath
, base_len
);
899 strbuf_addf(&newpath
, "_%d", suffix
++);
902 /* Track the new path in our memory pool */
903 ret
= mem_pool_alloc(&opt
->priv
->pool
, newpath
.len
+ 1);
904 memcpy(ret
, newpath
.buf
, newpath
.len
+ 1);
905 strbuf_release(&newpath
);
909 /*** Function Grouping: functions related to collect_merge_info() ***/
911 static int traverse_trees_wrapper_callback(int n
,
913 unsigned long dirmask
,
914 struct name_entry
*names
,
915 struct traverse_info
*info
)
917 struct merge_options
*opt
= info
->data
;
918 struct rename_info
*renames
= &opt
->priv
->renames
;
919 unsigned filemask
= mask
& ~dirmask
;
923 if (!renames
->callback_data_traverse_path
)
924 renames
->callback_data_traverse_path
= xstrdup(info
->traverse_path
);
926 if (filemask
&& filemask
== renames
->dir_rename_mask
)
927 renames
->dir_rename_mask
= 0x07;
929 ALLOC_GROW(renames
->callback_data
, renames
->callback_data_nr
+ 1,
930 renames
->callback_data_alloc
);
931 renames
->callback_data
[renames
->callback_data_nr
].mask
= mask
;
932 renames
->callback_data
[renames
->callback_data_nr
].dirmask
= dirmask
;
933 COPY_ARRAY(renames
->callback_data
[renames
->callback_data_nr
].names
,
935 renames
->callback_data_nr
++;
941 * Much like traverse_trees(), BUT:
942 * - read all the tree entries FIRST, saving them
943 * - note that the above step provides an opportunity to compute necessary
944 * additional details before the "real" traversal
945 * - loop through the saved entries and call the original callback on them
947 static int traverse_trees_wrapper(struct index_state
*istate
,
950 struct traverse_info
*info
)
952 int ret
, i
, old_offset
;
953 traverse_callback_t old_fn
;
954 char *old_callback_data_traverse_path
;
955 struct merge_options
*opt
= info
->data
;
956 struct rename_info
*renames
= &opt
->priv
->renames
;
958 assert(renames
->dir_rename_mask
== 2 || renames
->dir_rename_mask
== 4);
960 old_callback_data_traverse_path
= renames
->callback_data_traverse_path
;
962 old_offset
= renames
->callback_data_nr
;
964 renames
->callback_data_traverse_path
= NULL
;
965 info
->fn
= traverse_trees_wrapper_callback
;
966 ret
= traverse_trees(istate
, n
, t
, info
);
970 info
->traverse_path
= renames
->callback_data_traverse_path
;
972 for (i
= old_offset
; i
< renames
->callback_data_nr
; ++i
) {
974 renames
->callback_data
[i
].mask
,
975 renames
->callback_data
[i
].dirmask
,
976 renames
->callback_data
[i
].names
,
980 renames
->callback_data_nr
= old_offset
;
981 free(renames
->callback_data_traverse_path
);
982 renames
->callback_data_traverse_path
= old_callback_data_traverse_path
;
983 info
->traverse_path
= NULL
;
987 static void setup_path_info(struct merge_options
*opt
,
988 struct string_list_item
*result
,
989 const char *current_dir_name
,
990 int current_dir_name_len
,
991 char *fullpath
, /* we'll take over ownership */
992 struct name_entry
*names
,
993 struct name_entry
*merged_version
,
994 unsigned is_null
, /* boolean */
995 unsigned df_conflict
, /* boolean */
998 int resolved
/* boolean */)
1000 /* result->util is void*, so mi is a convenience typed variable */
1001 struct merged_info
*mi
;
1003 assert(!is_null
|| resolved
);
1004 assert(!df_conflict
|| !resolved
); /* df_conflict implies !resolved */
1005 assert(resolved
== (merged_version
!= NULL
));
1007 mi
= mem_pool_calloc(&opt
->priv
->pool
, 1,
1008 resolved
? sizeof(struct merged_info
) :
1009 sizeof(struct conflict_info
));
1010 mi
->directory_name
= current_dir_name
;
1011 mi
->basename_offset
= current_dir_name_len
;
1012 mi
->clean
= !!resolved
;
1014 mi
->result
.mode
= merged_version
->mode
;
1015 oidcpy(&mi
->result
.oid
, &merged_version
->oid
);
1016 mi
->is_null
= !!is_null
;
1019 struct conflict_info
*ci
;
1021 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1022 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1023 ci
->pathnames
[i
] = fullpath
;
1024 ci
->stages
[i
].mode
= names
[i
].mode
;
1025 oidcpy(&ci
->stages
[i
].oid
, &names
[i
].oid
);
1027 ci
->filemask
= filemask
;
1028 ci
->dirmask
= dirmask
;
1029 ci
->df_conflict
= !!df_conflict
;
1032 * Assume is_null for now, but if we have entries
1033 * under the directory then when it is complete in
1034 * write_completed_directory() it'll update this.
1035 * Also, for D/F conflicts, we have to handle the
1036 * directory first, then clear this bit and process
1037 * the file to see how it is handled -- that occurs
1038 * near the top of process_entry().
1042 strmap_put(&opt
->priv
->paths
, fullpath
, mi
);
1043 result
->string
= fullpath
;
1047 static void add_pair(struct merge_options
*opt
,
1048 struct name_entry
*names
,
1049 const char *pathname
,
1051 unsigned is_add
/* if false, is_delete */,
1052 unsigned match_mask
,
1053 unsigned dir_rename_mask
)
1055 struct diff_filespec
*one
, *two
;
1056 struct rename_info
*renames
= &opt
->priv
->renames
;
1057 int names_idx
= is_add
? side
: 0;
1060 assert(match_mask
== 0 || match_mask
== 6);
1061 if (strset_contains(&renames
->cached_target_names
[side
],
1065 unsigned content_relevant
= (match_mask
== 0);
1066 unsigned location_relevant
= (dir_rename_mask
== 0x07);
1068 assert(match_mask
== 0 || match_mask
== 3 || match_mask
== 5);
1071 * If pathname is found in cached_irrelevant[side] due to
1072 * previous pick but for this commit content is relevant,
1073 * then we need to remove it from cached_irrelevant.
1075 if (content_relevant
)
1076 /* strset_remove is no-op if strset doesn't have key */
1077 strset_remove(&renames
->cached_irrelevant
[side
],
1081 * We do not need to re-detect renames for paths that we already
1082 * know the pairing, i.e. for cached_pairs (or
1083 * cached_irrelevant). However, handle_deferred_entries() needs
1084 * to loop over the union of keys from relevant_sources[side] and
1085 * cached_pairs[side], so for simplicity we set relevant_sources
1086 * for all the cached_pairs too and then strip them back out in
1087 * prune_cached_from_relevant() at the beginning of
1088 * detect_regular_renames().
1090 if (content_relevant
|| location_relevant
) {
1091 /* content_relevant trumps location_relevant */
1092 strintmap_set(&renames
->relevant_sources
[side
], pathname
,
1093 content_relevant
? RELEVANT_CONTENT
: RELEVANT_LOCATION
);
1097 * Avoid creating pair if we've already cached rename results.
1098 * Note that we do this after setting relevant_sources[side]
1099 * as noted in the comment above.
1101 if (strmap_contains(&renames
->cached_pairs
[side
], pathname
) ||
1102 strset_contains(&renames
->cached_irrelevant
[side
], pathname
))
1106 one
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1107 two
= pool_alloc_filespec(&opt
->priv
->pool
, pathname
);
1108 fill_filespec(is_add
? two
: one
,
1109 &names
[names_idx
].oid
, 1, names
[names_idx
].mode
);
1110 pool_diff_queue(&opt
->priv
->pool
, &renames
->pairs
[side
], one
, two
);
1113 static void collect_rename_info(struct merge_options
*opt
,
1114 struct name_entry
*names
,
1115 const char *dirname
,
1116 const char *fullname
,
1119 unsigned match_mask
)
1121 struct rename_info
*renames
= &opt
->priv
->renames
;
1125 * Update dir_rename_mask (determines ignore-rename-source validity)
1127 * dir_rename_mask helps us keep track of when directory rename
1128 * detection may be relevant. Basically, whenver a directory is
1129 * removed on one side of history, and a file is added to that
1130 * directory on the other side of history, directory rename
1131 * detection is relevant (meaning we have to detect renames for all
1132 * files within that directory to deduce where the directory
1133 * moved). Also, whenever a directory needs directory rename
1134 * detection, due to the "majority rules" choice for where to move
1135 * it (see t6423 testcase 1f), we also need to detect renames for
1136 * all files within subdirectories of that directory as well.
1138 * Here we haven't looked at files within the directory yet, we are
1139 * just looking at the directory itself. So, if we aren't yet in
1140 * a case where a parent directory needed directory rename detection
1141 * (i.e. dir_rename_mask != 0x07), and if the directory was removed
1142 * on one side of history, record the mask of the other side of
1143 * history in dir_rename_mask.
1145 if (renames
->dir_rename_mask
!= 0x07 &&
1146 (dirmask
== 3 || dirmask
== 5)) {
1147 /* simple sanity check */
1148 assert(renames
->dir_rename_mask
== 0 ||
1149 renames
->dir_rename_mask
== (dirmask
& ~1));
1150 /* update dir_rename_mask; have it record mask of new side */
1151 renames
->dir_rename_mask
= (dirmask
& ~1);
1154 /* Update dirs_removed, as needed */
1155 if (dirmask
== 1 || dirmask
== 3 || dirmask
== 5) {
1156 /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */
1157 unsigned sides
= (0x07 - dirmask
)/2;
1158 unsigned relevance
= (renames
->dir_rename_mask
== 0x07) ?
1159 RELEVANT_FOR_ANCESTOR
: NOT_RELEVANT
;
1161 * Record relevance of this directory. However, note that
1162 * when collect_merge_info_callback() recurses into this
1163 * directory and calls collect_rename_info() on paths
1164 * within that directory, if we find a path that was added
1165 * to this directory on the other side of history, we will
1166 * upgrade this value to RELEVANT_FOR_SELF; see below.
1169 strintmap_set(&renames
->dirs_removed
[1], fullname
,
1172 strintmap_set(&renames
->dirs_removed
[2], fullname
,
1177 * Here's the block that potentially upgrades to RELEVANT_FOR_SELF.
1178 * When we run across a file added to a directory. In such a case,
1179 * find the directory of the file and upgrade its relevance.
1181 if (renames
->dir_rename_mask
== 0x07 &&
1182 (filemask
== 2 || filemask
== 4)) {
1184 * Need directory rename for parent directory on other side
1185 * of history from added file. Thus
1186 * side = (~filemask & 0x06) >> 1
1188 * side = 3 - (filemask/2).
1190 unsigned side
= 3 - (filemask
>> 1);
1191 strintmap_set(&renames
->dirs_removed
[side
], dirname
,
1195 if (filemask
== 0 || filemask
== 7)
1198 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; ++side
) {
1199 unsigned side_mask
= (1 << side
);
1201 /* Check for deletion on side */
1202 if ((filemask
& 1) && !(filemask
& side_mask
))
1203 add_pair(opt
, names
, fullname
, side
, 0 /* delete */,
1204 match_mask
& filemask
,
1205 renames
->dir_rename_mask
);
1207 /* Check for addition on side */
1208 if (!(filemask
& 1) && (filemask
& side_mask
))
1209 add_pair(opt
, names
, fullname
, side
, 1 /* add */,
1210 match_mask
& filemask
,
1211 renames
->dir_rename_mask
);
1215 static int collect_merge_info_callback(int n
,
1217 unsigned long dirmask
,
1218 struct name_entry
*names
,
1219 struct traverse_info
*info
)
1223 * common ancestor (mbase) has mask 1, and stored in index 0 of names
1224 * head of side 1 (side1) has mask 2, and stored in index 1 of names
1225 * head of side 2 (side2) has mask 4, and stored in index 2 of names
1227 struct merge_options
*opt
= info
->data
;
1228 struct merge_options_internal
*opti
= opt
->priv
;
1229 struct rename_info
*renames
= &opt
->priv
->renames
;
1230 struct string_list_item pi
; /* Path Info */
1231 struct conflict_info
*ci
; /* typed alias to pi.util (which is void*) */
1232 struct name_entry
*p
;
1235 const char *dirname
= opti
->current_dir_name
;
1236 unsigned prev_dir_rename_mask
= renames
->dir_rename_mask
;
1237 unsigned filemask
= mask
& ~dirmask
;
1238 unsigned match_mask
= 0; /* will be updated below */
1239 unsigned mbase_null
= !(mask
& 1);
1240 unsigned side1_null
= !(mask
& 2);
1241 unsigned side2_null
= !(mask
& 4);
1242 unsigned side1_matches_mbase
= (!side1_null
&& !mbase_null
&&
1243 names
[0].mode
== names
[1].mode
&&
1244 oideq(&names
[0].oid
, &names
[1].oid
));
1245 unsigned side2_matches_mbase
= (!side2_null
&& !mbase_null
&&
1246 names
[0].mode
== names
[2].mode
&&
1247 oideq(&names
[0].oid
, &names
[2].oid
));
1248 unsigned sides_match
= (!side1_null
&& !side2_null
&&
1249 names
[1].mode
== names
[2].mode
&&
1250 oideq(&names
[1].oid
, &names
[2].oid
));
1253 * Note: When a path is a file on one side of history and a directory
1254 * in another, we have a directory/file conflict. In such cases, if
1255 * the conflict doesn't resolve from renames and deletions, then we
1256 * always leave directories where they are and move files out of the
1257 * way. Thus, while struct conflict_info has a df_conflict field to
1258 * track such conflicts, we ignore that field for any directories at
1259 * a path and only pay attention to it for files at the given path.
1260 * The fact that we leave directories were they are also means that
1261 * we do not need to worry about getting additional df_conflict
1262 * information propagated from parent directories down to children
1263 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
1264 * sets a newinfo.df_conflicts field specifically to propagate it).
1266 unsigned df_conflict
= (filemask
!= 0) && (dirmask
!= 0);
1268 /* n = 3 is a fundamental assumption. */
1270 BUG("Called collect_merge_info_callback wrong");
1273 * A bunch of sanity checks verifying that traverse_trees() calls
1274 * us the way I expect. Could just remove these at some point,
1275 * though maybe they are helpful to future code readers.
1277 assert(mbase_null
== is_null_oid(&names
[0].oid
));
1278 assert(side1_null
== is_null_oid(&names
[1].oid
));
1279 assert(side2_null
== is_null_oid(&names
[2].oid
));
1280 assert(!mbase_null
|| !side1_null
|| !side2_null
);
1281 assert(mask
> 0 && mask
< 8);
1283 /* Determine match_mask */
1284 if (side1_matches_mbase
)
1285 match_mask
= (side2_matches_mbase
? 7 : 3);
1286 else if (side2_matches_mbase
)
1288 else if (sides_match
)
1292 * Get the name of the relevant filepath, which we'll pass to
1293 * setup_path_info() for tracking.
1298 len
= traverse_path_len(info
, p
->pathlen
);
1300 /* +1 in both of the following lines to include the NUL byte */
1301 fullpath
= mem_pool_alloc(&opt
->priv
->pool
, len
+ 1);
1302 make_traverse_path(fullpath
, len
+ 1, info
, p
->path
, p
->pathlen
);
1305 * If mbase, side1, and side2 all match, we can resolve early. Even
1306 * if these are trees, there will be no renames or anything
1309 if (side1_matches_mbase
&& side2_matches_mbase
) {
1310 /* mbase, side1, & side2 all match; use mbase as resolution */
1311 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1312 names
, names
+0, mbase_null
, 0 /* df_conflict */,
1313 filemask
, dirmask
, 1 /* resolved */);
1318 * If the sides match, and all three paths are present and are
1319 * files, then we can take either as the resolution. We can't do
1320 * this with trees, because there may be rename sources from the
1323 if (sides_match
&& filemask
== 0x07) {
1324 /* use side1 (== side2) version as resolution */
1325 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1326 names
, names
+1, side1_null
, 0,
1327 filemask
, dirmask
, 1);
1332 * If side1 matches mbase and all three paths are present and are
1333 * files, then we can use side2 as the resolution. We cannot
1334 * necessarily do so this for trees, because there may be rename
1335 * destinations within side2.
1337 if (side1_matches_mbase
&& filemask
== 0x07) {
1338 /* use side2 version as resolution */
1339 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1340 names
, names
+2, side2_null
, 0,
1341 filemask
, dirmask
, 1);
1345 /* Similar to above but swapping sides 1 and 2 */
1346 if (side2_matches_mbase
&& filemask
== 0x07) {
1347 /* use side1 version as resolution */
1348 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1349 names
, names
+1, side1_null
, 0,
1350 filemask
, dirmask
, 1);
1355 * Sometimes we can tell that a source path need not be included in
1356 * rename detection -- namely, whenever either
1357 * side1_matches_mbase && side2_null
1359 * side2_matches_mbase && side1_null
1360 * However, we call collect_rename_info() even in those cases,
1361 * because exact renames are cheap and would let us remove both a
1362 * source and destination path. We'll cull the unneeded sources
1365 collect_rename_info(opt
, names
, dirname
, fullpath
,
1366 filemask
, dirmask
, match_mask
);
1369 * None of the special cases above matched, so we have a
1370 * provisional conflict. (Rename detection might allow us to
1371 * unconflict some more cases, but that comes later so all we can
1372 * do now is record the different non-null file hashes.)
1374 setup_path_info(opt
, &pi
, dirname
, info
->pathlen
, fullpath
,
1375 names
, NULL
, 0, df_conflict
, filemask
, dirmask
, 0);
1379 ci
->match_mask
= match_mask
;
1381 /* If dirmask, recurse into subdirectories */
1383 struct traverse_info newinfo
;
1384 struct tree_desc t
[3];
1385 void *buf
[3] = {NULL
, NULL
, NULL
};
1386 const char *original_dir_name
;
1390 * Check for whether we can avoid recursing due to one side
1391 * matching the merge base. The side that does NOT match is
1392 * the one that might have a rename destination we need.
1394 assert(!side1_matches_mbase
|| !side2_matches_mbase
);
1395 side
= side1_matches_mbase
? MERGE_SIDE2
:
1396 side2_matches_mbase
? MERGE_SIDE1
: MERGE_BASE
;
1397 if (filemask
== 0 && (dirmask
== 2 || dirmask
== 4)) {
1399 * Also defer recursing into new directories; set up a
1400 * few variables to let us do so.
1402 ci
->match_mask
= (7 - dirmask
);
1405 if (renames
->dir_rename_mask
!= 0x07 &&
1406 side
!= MERGE_BASE
&&
1407 renames
->deferred
[side
].trivial_merges_okay
&&
1408 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1410 strintmap_set(&renames
->deferred
[side
].possible_trivial_merges
,
1411 pi
.string
, renames
->dir_rename_mask
);
1412 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1416 /* We need to recurse */
1417 ci
->match_mask
&= filemask
;
1419 newinfo
.prev
= info
;
1420 newinfo
.name
= p
->path
;
1421 newinfo
.namelen
= p
->pathlen
;
1422 newinfo
.pathlen
= st_add3(newinfo
.pathlen
, p
->pathlen
, 1);
1424 * If this directory we are about to recurse into cared about
1425 * its parent directory (the current directory) having a D/F
1426 * conflict, then we'd propagate the masks in this way:
1427 * newinfo.df_conflicts |= (mask & ~dirmask);
1428 * But we don't worry about propagating D/F conflicts. (See
1429 * comment near setting of local df_conflict variable near
1430 * the beginning of this function).
1433 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
1434 if (i
== 1 && side1_matches_mbase
)
1436 else if (i
== 2 && side2_matches_mbase
)
1438 else if (i
== 2 && sides_match
)
1441 const struct object_id
*oid
= NULL
;
1443 oid
= &names
[i
].oid
;
1444 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1450 original_dir_name
= opti
->current_dir_name
;
1451 opti
->current_dir_name
= pi
.string
;
1452 if (renames
->dir_rename_mask
== 0 ||
1453 renames
->dir_rename_mask
== 0x07)
1454 ret
= traverse_trees(NULL
, 3, t
, &newinfo
);
1456 ret
= traverse_trees_wrapper(NULL
, 3, t
, &newinfo
);
1457 opti
->current_dir_name
= original_dir_name
;
1458 renames
->dir_rename_mask
= prev_dir_rename_mask
;
1460 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1470 static void resolve_trivial_directory_merge(struct conflict_info
*ci
, int side
)
1473 assert((side
== 1 && ci
->match_mask
== 5) ||
1474 (side
== 2 && ci
->match_mask
== 3));
1475 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
1476 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
1477 ci
->merged
.is_null
= is_null_oid(&ci
->stages
[side
].oid
);
1479 ci
->merged
.clean
= 1; /* (ci->filemask == 0); */
1482 static int handle_deferred_entries(struct merge_options
*opt
,
1483 struct traverse_info
*info
)
1485 struct rename_info
*renames
= &opt
->priv
->renames
;
1486 struct hashmap_iter iter
;
1487 struct strmap_entry
*entry
;
1489 int path_count_before
, path_count_after
= 0;
1491 path_count_before
= strmap_get_size(&opt
->priv
->paths
);
1492 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
1493 unsigned optimization_okay
= 1;
1494 struct strintmap copy
;
1496 /* Loop over the set of paths we need to know rename info for */
1497 strset_for_each_entry(&renames
->relevant_sources
[side
],
1499 char *rename_target
, *dir
, *dir_marker
;
1500 struct strmap_entry
*e
;
1503 * If we don't know delete/rename info for this path,
1504 * then we need to recurse into all trees to get all
1505 * adds to make sure we have it.
1507 if (strset_contains(&renames
->cached_irrelevant
[side
],
1510 e
= strmap_get_entry(&renames
->cached_pairs
[side
],
1513 optimization_okay
= 0;
1517 /* If this is a delete, we have enough info already */
1518 rename_target
= e
->value
;
1522 /* If we already walked the rename target, we're good */
1523 if (strmap_contains(&opt
->priv
->paths
, rename_target
))
1527 * Otherwise, we need to get a list of directories that
1528 * will need to be recursed into to get this
1531 dir
= xstrdup(rename_target
);
1532 while ((dir_marker
= strrchr(dir
, '/'))) {
1534 if (strset_contains(&renames
->deferred
[side
].target_dirs
,
1537 strset_add(&renames
->deferred
[side
].target_dirs
,
1542 renames
->deferred
[side
].trivial_merges_okay
= optimization_okay
;
1544 * We need to recurse into any directories in
1545 * possible_trivial_merges[side] found in target_dirs[side].
1546 * But when we recurse, we may need to queue up some of the
1547 * subdirectories for possible_trivial_merges[side]. Since
1548 * we can't safely iterate through a hashmap while also adding
1549 * entries, move the entries into 'copy', iterate over 'copy',
1550 * and then we'll also iterate anything added into
1551 * possible_trivial_merges[side] once this loop is done.
1553 copy
= renames
->deferred
[side
].possible_trivial_merges
;
1554 strintmap_init_with_options(&renames
->deferred
[side
].possible_trivial_merges
,
1558 strintmap_for_each_entry(©
, &iter
, entry
) {
1559 const char *path
= entry
->key
;
1560 unsigned dir_rename_mask
= (intptr_t)entry
->value
;
1561 struct conflict_info
*ci
;
1563 struct tree_desc t
[3];
1564 void *buf
[3] = {NULL
,};
1567 ci
= strmap_get(&opt
->priv
->paths
, path
);
1569 dirmask
= ci
->dirmask
;
1571 if (optimization_okay
&&
1572 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1574 resolve_trivial_directory_merge(ci
, side
);
1579 info
->namelen
= strlen(path
);
1580 info
->pathlen
= info
->namelen
+ 1;
1582 for (i
= 0; i
< 3; i
++, dirmask
>>= 1) {
1583 if (i
== 1 && ci
->match_mask
== 3)
1585 else if (i
== 2 && ci
->match_mask
== 5)
1587 else if (i
== 2 && ci
->match_mask
== 6)
1590 const struct object_id
*oid
= NULL
;
1592 oid
= &ci
->stages
[i
].oid
;
1593 buf
[i
] = fill_tree_descriptor(opt
->repo
,
1598 ci
->match_mask
&= ci
->filemask
;
1599 opt
->priv
->current_dir_name
= path
;
1600 renames
->dir_rename_mask
= dir_rename_mask
;
1601 if (renames
->dir_rename_mask
== 0 ||
1602 renames
->dir_rename_mask
== 0x07)
1603 ret
= traverse_trees(NULL
, 3, t
, info
);
1605 ret
= traverse_trees_wrapper(NULL
, 3, t
, info
);
1607 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++)
1613 strintmap_clear(©
);
1614 strintmap_for_each_entry(&renames
->deferred
[side
].possible_trivial_merges
,
1616 const char *path
= entry
->key
;
1617 struct conflict_info
*ci
;
1619 ci
= strmap_get(&opt
->priv
->paths
, path
);
1622 assert(renames
->deferred
[side
].trivial_merges_okay
&&
1623 !strset_contains(&renames
->deferred
[side
].target_dirs
,
1625 resolve_trivial_directory_merge(ci
, side
);
1627 if (!optimization_okay
|| path_count_after
)
1628 path_count_after
= strmap_get_size(&opt
->priv
->paths
);
1630 if (path_count_after
) {
1632 * The choice of wanted_factor here does not affect
1633 * correctness, only performance. When the
1634 * path_count_after / path_count_before
1635 * ratio is high, redoing after renames is a big
1636 * performance boost. I suspect that redoing is a wash
1637 * somewhere near a value of 2, and below that redoing will
1638 * slow things down. I applied a fudge factor and picked
1639 * 3; see the commit message when this was introduced for
1640 * back of the envelope calculations for this ratio.
1642 const int wanted_factor
= 3;
1644 /* We should only redo collect_merge_info one time */
1645 assert(renames
->redo_after_renames
== 0);
1647 if (path_count_after
/ path_count_before
>= wanted_factor
) {
1648 renames
->redo_after_renames
= 1;
1649 renames
->cached_pairs_valid_side
= -1;
1651 } else if (renames
->redo_after_renames
== 2)
1652 renames
->redo_after_renames
= 0;
1656 static int collect_merge_info(struct merge_options
*opt
,
1657 struct tree
*merge_base
,
1662 struct tree_desc t
[3];
1663 struct traverse_info info
;
1665 opt
->priv
->toplevel_dir
= "";
1666 opt
->priv
->current_dir_name
= opt
->priv
->toplevel_dir
;
1667 setup_traverse_info(&info
, opt
->priv
->toplevel_dir
);
1668 info
.fn
= collect_merge_info_callback
;
1670 info
.show_all_errors
= 1;
1672 parse_tree(merge_base
);
1675 init_tree_desc(t
+ 0, merge_base
->buffer
, merge_base
->size
);
1676 init_tree_desc(t
+ 1, side1
->buffer
, side1
->size
);
1677 init_tree_desc(t
+ 2, side2
->buffer
, side2
->size
);
1679 trace2_region_enter("merge", "traverse_trees", opt
->repo
);
1680 ret
= traverse_trees(NULL
, 3, t
, &info
);
1682 ret
= handle_deferred_entries(opt
, &info
);
1683 trace2_region_leave("merge", "traverse_trees", opt
->repo
);
1688 /*** Function Grouping: functions related to threeway content merges ***/
1690 static int find_first_merges(struct repository
*repo
,
1694 struct object_array
*result
)
1697 struct object_array merges
= OBJECT_ARRAY_INIT
;
1698 struct commit
*commit
;
1699 int contains_another
;
1701 char merged_revision
[GIT_MAX_HEXSZ
+ 2];
1702 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1703 "--all", merged_revision
, NULL
};
1704 struct rev_info revs
;
1705 struct setup_revision_opt rev_opts
;
1707 memset(result
, 0, sizeof(struct object_array
));
1708 memset(&rev_opts
, 0, sizeof(rev_opts
));
1710 /* get all revisions that merge commit a */
1711 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1712 oid_to_hex(&a
->object
.oid
));
1713 repo_init_revisions(repo
, &revs
, NULL
);
1714 /* FIXME: can't handle linked worktrees in submodules yet */
1715 revs
.single_worktree
= path
!= NULL
;
1716 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1718 /* save all revisions from the above list that contain b */
1719 if (prepare_revision_walk(&revs
))
1720 die("revision walk setup failed");
1721 while ((commit
= get_revision(&revs
)) != NULL
) {
1722 struct object
*o
= &(commit
->object
);
1723 if (repo_in_merge_bases(repo
, b
, commit
))
1724 add_object_array(o
, NULL
, &merges
);
1726 reset_revision_walk();
1728 /* Now we've got all merges that contain a and b. Prune all
1729 * merges that contain another found merge and save them in
1732 for (i
= 0; i
< merges
.nr
; i
++) {
1733 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1735 contains_another
= 0;
1736 for (j
= 0; j
< merges
.nr
; j
++) {
1737 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1738 if (i
!= j
&& repo_in_merge_bases(repo
, m2
, m1
)) {
1739 contains_another
= 1;
1744 if (!contains_another
)
1745 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1748 object_array_clear(&merges
);
1749 release_revisions(&revs
);
1753 static int merge_submodule(struct merge_options
*opt
,
1755 const struct object_id
*o
,
1756 const struct object_id
*a
,
1757 const struct object_id
*b
,
1758 struct object_id
*result
)
1760 struct repository subrepo
;
1761 struct strbuf sb
= STRBUF_INIT
;
1763 struct commit
*commit_o
, *commit_a
, *commit_b
;
1765 struct object_array merges
;
1768 int search
= !opt
->priv
->call_depth
;
1769 int sub_not_initialized
= 1;
1770 int sub_flag
= CONFLICT_SUBMODULE_FAILED_TO_MERGE
;
1772 /* store fallback answer in result in case we fail */
1773 oidcpy(result
, opt
->priv
->call_depth
? o
: a
);
1775 /* we can not handle deletion conflicts */
1776 if (is_null_oid(a
) || is_null_oid(b
))
1777 BUG("submodule deleted on one side; this should be handled outside of merge_submodule()");
1779 if ((sub_not_initialized
= repo_submodule_init(&subrepo
,
1780 opt
->repo
, path
, null_oid()))) {
1781 path_msg(opt
, CONFLICT_SUBMODULE_NOT_INITIALIZED
, 0,
1782 path
, NULL
, NULL
, NULL
,
1783 _("Failed to merge submodule %s (not checked out)"),
1785 sub_flag
= CONFLICT_SUBMODULE_NOT_INITIALIZED
;
1789 if (is_null_oid(o
)) {
1790 path_msg(opt
, CONFLICT_SUBMODULE_NULL_MERGE_BASE
, 0,
1791 path
, NULL
, NULL
, NULL
,
1792 _("Failed to merge submodule %s (no merge base)"),
1797 if (!(commit_o
= lookup_commit_reference(&subrepo
, o
)) ||
1798 !(commit_a
= lookup_commit_reference(&subrepo
, a
)) ||
1799 !(commit_b
= lookup_commit_reference(&subrepo
, b
))) {
1800 path_msg(opt
, CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
, 0,
1801 path
, NULL
, NULL
, NULL
,
1802 _("Failed to merge submodule %s (commits not present)"),
1804 sub_flag
= CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
;
1808 /* check whether both changes are forward */
1809 if (!repo_in_merge_bases(&subrepo
, commit_o
, commit_a
) ||
1810 !repo_in_merge_bases(&subrepo
, commit_o
, commit_b
)) {
1811 path_msg(opt
, CONFLICT_SUBMODULE_MAY_HAVE_REWINDS
, 0,
1812 path
, NULL
, NULL
, NULL
,
1813 _("Failed to merge submodule %s "
1814 "(commits don't follow merge-base)"),
1819 /* Case #1: a is contained in b or vice versa */
1820 if (repo_in_merge_bases(&subrepo
, commit_a
, commit_b
)) {
1822 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1823 path
, NULL
, NULL
, NULL
,
1824 _("Note: Fast-forwarding submodule %s to %s"),
1825 path
, oid_to_hex(b
));
1829 if (repo_in_merge_bases(&subrepo
, commit_b
, commit_a
)) {
1831 path_msg(opt
, INFO_SUBMODULE_FAST_FORWARDING
, 1,
1832 path
, NULL
, NULL
, NULL
,
1833 _("Note: Fast-forwarding submodule %s to %s"),
1834 path
, oid_to_hex(a
));
1840 * Case #2: There are one or more merges that contain a and b in
1841 * the submodule. If there is only one, then present it as a
1842 * suggestion to the user, but leave it marked unmerged so the
1843 * user needs to confirm the resolution.
1846 /* Skip the search if makes no sense to the calling context. */
1850 /* find commit which merges them */
1851 parent_count
= find_first_merges(&subrepo
, path
, commit_a
, commit_b
,
1853 switch (parent_count
) {
1855 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE
, 0,
1856 path
, NULL
, NULL
, NULL
,
1857 _("Failed to merge submodule %s"), path
);
1861 format_commit(&sb
, 4, &subrepo
,
1862 (struct commit
*)merges
.objects
[0].item
);
1863 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1864 path
, NULL
, NULL
, NULL
,
1865 _("Failed to merge submodule %s, but a possible merge "
1866 "resolution exists: %s"),
1868 strbuf_release(&sb
);
1871 for (i
= 0; i
< merges
.nr
; i
++)
1872 format_commit(&sb
, 4, &subrepo
,
1873 (struct commit
*)merges
.objects
[i
].item
);
1874 path_msg(opt
, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION
, 0,
1875 path
, NULL
, NULL
, NULL
,
1876 _("Failed to merge submodule %s, but multiple "
1877 "possible merges exist:\n%s"), path
, sb
.buf
);
1878 strbuf_release(&sb
);
1881 object_array_clear(&merges
);
1883 if (!opt
->priv
->call_depth
&& !ret
) {
1884 struct string_list
*csub
= &opt
->priv
->conflicted_submodules
;
1885 struct conflicted_submodule_item
*util
;
1888 util
= xmalloc(sizeof(*util
));
1889 util
->flag
= sub_flag
;
1890 util
->abbrev
= NULL
;
1891 if (!sub_not_initialized
) {
1892 abbrev
= repo_find_unique_abbrev(&subrepo
, b
, DEFAULT_ABBREV
);
1893 util
->abbrev
= xstrdup(abbrev
);
1895 string_list_append(csub
, path
)->util
= util
;
1898 if (!sub_not_initialized
)
1899 repo_clear(&subrepo
);
1903 static void initialize_attr_index(struct merge_options
*opt
)
1906 * The renormalize_buffer() functions require attributes, and
1907 * annoyingly those can only be read from the working tree or from
1908 * an index_state. merge-ort doesn't have an index_state, so we
1909 * generate a fake one containing only attribute information.
1911 struct merged_info
*mi
;
1912 struct index_state
*attr_index
= &opt
->priv
->attr_index
;
1913 struct cache_entry
*ce
;
1915 attr_index
->initialized
= 1;
1917 if (!opt
->renormalize
)
1920 mi
= strmap_get(&opt
->priv
->paths
, GITATTRIBUTES_FILE
);
1925 int len
= strlen(GITATTRIBUTES_FILE
);
1926 ce
= make_empty_cache_entry(attr_index
, len
);
1927 ce
->ce_mode
= create_ce_mode(mi
->result
.mode
);
1928 ce
->ce_flags
= create_ce_flags(0);
1929 ce
->ce_namelen
= len
;
1930 oidcpy(&ce
->oid
, &mi
->result
.oid
);
1931 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1932 add_index_entry(attr_index
, ce
,
1933 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1934 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
, &ce
->oid
);
1937 struct conflict_info
*ci
;
1939 ASSIGN_AND_VERIFY_CI(ci
, mi
);
1940 for (stage
= 0; stage
< 3; stage
++) {
1941 unsigned stage_mask
= (1 << stage
);
1943 if (!(ci
->filemask
& stage_mask
))
1945 len
= strlen(GITATTRIBUTES_FILE
);
1946 ce
= make_empty_cache_entry(attr_index
, len
);
1947 ce
->ce_mode
= create_ce_mode(ci
->stages
[stage
].mode
);
1948 ce
->ce_flags
= create_ce_flags(stage
);
1949 ce
->ce_namelen
= len
;
1950 oidcpy(&ce
->oid
, &ci
->stages
[stage
].oid
);
1951 memcpy(ce
->name
, GITATTRIBUTES_FILE
, len
);
1952 add_index_entry(attr_index
, ce
,
1953 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
1954 get_stream_filter(attr_index
, GITATTRIBUTES_FILE
,
1960 static int merge_3way(struct merge_options
*opt
,
1962 const struct object_id
*o
,
1963 const struct object_id
*a
,
1964 const struct object_id
*b
,
1965 const char *pathnames
[3],
1966 const int extra_marker_size
,
1967 mmbuffer_t
*result_buf
)
1969 mmfile_t orig
, src1
, src2
;
1970 struct ll_merge_options ll_opts
= {0};
1971 char *base
, *name1
, *name2
;
1972 enum ll_merge_result merge_status
;
1974 if (!opt
->priv
->attr_index
.initialized
)
1975 initialize_attr_index(opt
);
1977 ll_opts
.renormalize
= opt
->renormalize
;
1978 ll_opts
.extra_marker_size
= extra_marker_size
;
1979 ll_opts
.xdl_opts
= opt
->xdl_opts
;
1981 if (opt
->priv
->call_depth
) {
1982 ll_opts
.virtual_ancestor
= 1;
1983 ll_opts
.variant
= 0;
1985 switch (opt
->recursive_variant
) {
1986 case MERGE_VARIANT_OURS
:
1987 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1989 case MERGE_VARIANT_THEIRS
:
1990 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1993 ll_opts
.variant
= 0;
1998 assert(pathnames
[0] && pathnames
[1] && pathnames
[2] && opt
->ancestor
);
1999 if (pathnames
[0] == pathnames
[1] && pathnames
[1] == pathnames
[2]) {
2000 base
= mkpathdup("%s", opt
->ancestor
);
2001 name1
= mkpathdup("%s", opt
->branch1
);
2002 name2
= mkpathdup("%s", opt
->branch2
);
2004 base
= mkpathdup("%s:%s", opt
->ancestor
, pathnames
[0]);
2005 name1
= mkpathdup("%s:%s", opt
->branch1
, pathnames
[1]);
2006 name2
= mkpathdup("%s:%s", opt
->branch2
, pathnames
[2]);
2009 read_mmblob(&orig
, o
);
2010 read_mmblob(&src1
, a
);
2011 read_mmblob(&src2
, b
);
2013 merge_status
= ll_merge(result_buf
, path
, &orig
, base
,
2014 &src1
, name1
, &src2
, name2
,
2015 &opt
->priv
->attr_index
, &ll_opts
);
2016 if (merge_status
== LL_MERGE_BINARY_CONFLICT
)
2017 path_msg(opt
, CONFLICT_BINARY
, 0,
2018 path
, NULL
, NULL
, NULL
,
2019 "warning: Cannot merge binary files: %s (%s vs. %s)",
2020 path
, name1
, name2
);
2028 return merge_status
;
2031 static int handle_content_merge(struct merge_options
*opt
,
2033 const struct version_info
*o
,
2034 const struct version_info
*a
,
2035 const struct version_info
*b
,
2036 const char *pathnames
[3],
2037 const int extra_marker_size
,
2038 struct version_info
*result
)
2041 * path is the target location where we want to put the file, and
2042 * is used to determine any normalization rules in ll_merge.
2044 * The normal case is that path and all entries in pathnames are
2045 * identical, though renames can affect which path we got one of
2046 * the three blobs to merge on various sides of history.
2048 * extra_marker_size is the amount to extend conflict markers in
2049 * ll_merge; this is neeed if we have content merges of content
2050 * merges, which happens for example with rename/rename(2to1) and
2051 * rename/add conflicts.
2056 * handle_content_merge() needs both files to be of the same type, i.e.
2057 * both files OR both submodules OR both symlinks. Conflicting types
2058 * needs to be handled elsewhere.
2060 assert((S_IFMT
& a
->mode
) == (S_IFMT
& b
->mode
));
2063 if (a
->mode
== b
->mode
|| a
->mode
== o
->mode
)
2064 result
->mode
= b
->mode
;
2066 /* must be the 100644/100755 case */
2067 assert(S_ISREG(a
->mode
));
2068 result
->mode
= a
->mode
;
2069 clean
= (b
->mode
== o
->mode
);
2071 * FIXME: If opt->priv->call_depth && !clean, then we really
2072 * should not make result->mode match either a->mode or
2073 * b->mode; that causes t6036 "check conflicting mode for
2074 * regular file" to fail. It would be best to use some other
2075 * mode, but we'll confuse all kinds of stuff if we use one
2076 * where S_ISREG(result->mode) isn't true, and if we use
2077 * something like 0100666, then tree-walk.c's calls to
2078 * canon_mode() will just normalize that to 100644 for us and
2079 * thus not solve anything.
2081 * Figure out if there's some kind of way we can work around
2087 * Trivial oid merge.
2089 * Note: While one might assume that the next four lines would
2090 * be unnecessary due to the fact that match_mask is often
2091 * setup and already handled, renames don't always take care
2094 if (oideq(&a
->oid
, &b
->oid
) || oideq(&a
->oid
, &o
->oid
))
2095 oidcpy(&result
->oid
, &b
->oid
);
2096 else if (oideq(&b
->oid
, &o
->oid
))
2097 oidcpy(&result
->oid
, &a
->oid
);
2099 /* Remaining rules depend on file vs. submodule vs. symlink. */
2100 else if (S_ISREG(a
->mode
)) {
2101 mmbuffer_t result_buf
;
2102 int ret
= 0, merge_status
;
2106 * If 'o' is different type, treat it as null so we do a
2109 two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2111 merge_status
= merge_3way(opt
, path
,
2112 two_way
? null_oid() : &o
->oid
,
2114 pathnames
, extra_marker_size
,
2117 if ((merge_status
< 0) || !result_buf
.ptr
)
2118 ret
= err(opt
, _("Failed to execute internal merge"));
2121 write_object_file(result_buf
.ptr
, result_buf
.size
,
2122 OBJ_BLOB
, &result
->oid
))
2123 ret
= err(opt
, _("Unable to add %s to database"),
2126 free(result_buf
.ptr
);
2129 clean
&= (merge_status
== 0);
2130 path_msg(opt
, INFO_AUTO_MERGING
, 1, path
, NULL
, NULL
, NULL
,
2131 _("Auto-merging %s"), path
);
2132 } else if (S_ISGITLINK(a
->mode
)) {
2133 int two_way
= ((S_IFMT
& o
->mode
) != (S_IFMT
& a
->mode
));
2134 clean
= merge_submodule(opt
, pathnames
[0],
2135 two_way
? null_oid() : &o
->oid
,
2136 &a
->oid
, &b
->oid
, &result
->oid
);
2137 if (opt
->priv
->call_depth
&& two_way
&& !clean
) {
2138 result
->mode
= o
->mode
;
2139 oidcpy(&result
->oid
, &o
->oid
);
2141 } else if (S_ISLNK(a
->mode
)) {
2142 if (opt
->priv
->call_depth
) {
2144 result
->mode
= o
->mode
;
2145 oidcpy(&result
->oid
, &o
->oid
);
2147 switch (opt
->recursive_variant
) {
2148 case MERGE_VARIANT_NORMAL
:
2150 oidcpy(&result
->oid
, &a
->oid
);
2152 case MERGE_VARIANT_OURS
:
2153 oidcpy(&result
->oid
, &a
->oid
);
2155 case MERGE_VARIANT_THEIRS
:
2156 oidcpy(&result
->oid
, &b
->oid
);
2161 BUG("unsupported object type in the tree: %06o for %s",
2167 /*** Function Grouping: functions related to detect_and_process_renames(), ***
2168 *** which are split into directory and regular rename detection sections. ***/
2170 /*** Function Grouping: functions related to directory rename detection ***/
2172 struct collision_info
{
2173 struct string_list source_files
;
2174 unsigned reported_already
:1;
2178 * Return a new string that replaces the beginning portion (which matches
2179 * rename_info->key), with rename_info->util.new_dir. In perl-speak:
2180 * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/);
2182 * Caller must ensure that old_path starts with rename_info->key + '/'.
2184 static char *apply_dir_rename(struct strmap_entry
*rename_info
,
2185 const char *old_path
)
2187 struct strbuf new_path
= STRBUF_INIT
;
2188 const char *old_dir
= rename_info
->key
;
2189 const char *new_dir
= rename_info
->value
;
2190 int oldlen
, newlen
, new_dir_len
;
2192 oldlen
= strlen(old_dir
);
2193 if (*new_dir
== '\0')
2195 * If someone renamed/merged a subdirectory into the root
2196 * directory (e.g. 'some/subdir' -> ''), then we want to
2199 * as the rename; we need to make old_path + oldlen advance
2200 * past the '/' character.
2203 new_dir_len
= strlen(new_dir
);
2204 newlen
= new_dir_len
+ (strlen(old_path
) - oldlen
) + 1;
2205 strbuf_grow(&new_path
, newlen
);
2206 strbuf_add(&new_path
, new_dir
, new_dir_len
);
2207 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
2209 return strbuf_detach(&new_path
, NULL
);
2212 static int path_in_way(struct strmap
*paths
, const char *path
, unsigned side_mask
)
2214 struct merged_info
*mi
= strmap_get(paths
, path
);
2215 struct conflict_info
*ci
;
2218 INITIALIZE_CI(ci
, mi
);
2219 return mi
->clean
|| (side_mask
& (ci
->filemask
| ci
->dirmask
));
2223 * See if there is a directory rename for path, and if there are any file
2224 * level conflicts on the given side for the renamed location. If there is
2225 * a rename and there are no conflicts, return the new name. Otherwise,
2228 static char *handle_path_level_conflicts(struct merge_options
*opt
,
2230 unsigned side_index
,
2231 struct strmap_entry
*rename_info
,
2232 struct strmap
*collisions
)
2234 char *new_path
= NULL
;
2235 struct collision_info
*c_info
;
2237 struct strbuf collision_paths
= STRBUF_INIT
;
2240 * entry has the mapping of old directory name to new directory name
2241 * that we want to apply to path.
2243 new_path
= apply_dir_rename(rename_info
, path
);
2245 BUG("Failed to apply directory rename!");
2248 * The caller needs to have ensured that it has pre-populated
2249 * collisions with all paths that map to new_path. Do a quick check
2250 * to ensure that's the case.
2252 c_info
= strmap_get(collisions
, new_path
);
2254 BUG("c_info is NULL");
2257 * Check for one-sided add/add/.../add conflicts, i.e.
2258 * where implicit renames from the other side doing
2259 * directory rename(s) can affect this side of history
2260 * to put multiple paths into the same location. Warn
2261 * and bail on directory renames for such paths.
2263 if (c_info
->reported_already
) {
2265 } else if (path_in_way(&opt
->priv
->paths
, new_path
, 1 << side_index
)) {
2266 c_info
->reported_already
= 1;
2267 strbuf_add_separated_string_list(&collision_paths
, ", ",
2268 &c_info
->source_files
);
2269 path_msg(opt
, CONFLICT_DIR_RENAME_FILE_IN_WAY
, 0,
2270 new_path
, NULL
, NULL
, &c_info
->source_files
,
2271 _("CONFLICT (implicit dir rename): Existing "
2272 "file/dir at %s in the way of implicit "
2273 "directory rename(s) putting the following "
2274 "path(s) there: %s."),
2275 new_path
, collision_paths
.buf
);
2277 } else if (c_info
->source_files
.nr
> 1) {
2278 c_info
->reported_already
= 1;
2279 strbuf_add_separated_string_list(&collision_paths
, ", ",
2280 &c_info
->source_files
);
2281 path_msg(opt
, CONFLICT_DIR_RENAME_COLLISION
, 0,
2282 new_path
, NULL
, NULL
, &c_info
->source_files
,
2283 _("CONFLICT (implicit dir rename): Cannot map "
2284 "more than one path to %s; implicit directory "
2285 "renames tried to put these paths there: %s"),
2286 new_path
, collision_paths
.buf
);
2290 /* Free memory we no longer need */
2291 strbuf_release(&collision_paths
);
2292 if (!clean
&& new_path
) {
2300 static void get_provisional_directory_renames(struct merge_options
*opt
,
2304 struct hashmap_iter iter
;
2305 struct strmap_entry
*entry
;
2306 struct rename_info
*renames
= &opt
->priv
->renames
;
2310 * dir_rename_count: old_directory -> {new_directory -> count}
2312 * dir_renames: old_directory -> best_new_directory
2313 * where best_new_directory is the one with the unique highest count.
2315 strmap_for_each_entry(&renames
->dir_rename_count
[side
], &iter
, entry
) {
2316 const char *source_dir
= entry
->key
;
2317 struct strintmap
*counts
= entry
->value
;
2318 struct hashmap_iter count_iter
;
2319 struct strmap_entry
*count_entry
;
2322 const char *best
= NULL
;
2324 strintmap_for_each_entry(counts
, &count_iter
, count_entry
) {
2325 const char *target_dir
= count_entry
->key
;
2326 intptr_t count
= (intptr_t)count_entry
->value
;
2330 else if (count
> max
) {
2339 if (bad_max
== max
) {
2340 path_msg(opt
, CONFLICT_DIR_RENAME_SPLIT
, 0,
2341 source_dir
, NULL
, NULL
, NULL
,
2342 _("CONFLICT (directory rename split): "
2343 "Unclear where to rename %s to; it was "
2344 "renamed to multiple other directories, "
2345 "with no destination getting a majority of "
2350 strmap_put(&renames
->dir_renames
[side
],
2351 source_dir
, (void*)best
);
2356 static void handle_directory_level_conflicts(struct merge_options
*opt
)
2358 struct hashmap_iter iter
;
2359 struct strmap_entry
*entry
;
2360 struct string_list duplicated
= STRING_LIST_INIT_NODUP
;
2361 struct rename_info
*renames
= &opt
->priv
->renames
;
2362 struct strmap
*side1_dir_renames
= &renames
->dir_renames
[MERGE_SIDE1
];
2363 struct strmap
*side2_dir_renames
= &renames
->dir_renames
[MERGE_SIDE2
];
2366 strmap_for_each_entry(side1_dir_renames
, &iter
, entry
) {
2367 if (strmap_contains(side2_dir_renames
, entry
->key
))
2368 string_list_append(&duplicated
, entry
->key
);
2371 for (i
= 0; i
< duplicated
.nr
; i
++) {
2372 strmap_remove(side1_dir_renames
, duplicated
.items
[i
].string
, 0);
2373 strmap_remove(side2_dir_renames
, duplicated
.items
[i
].string
, 0);
2375 string_list_clear(&duplicated
, 0);
2378 static struct strmap_entry
*check_dir_renamed(const char *path
,
2379 struct strmap
*dir_renames
)
2381 char *temp
= xstrdup(path
);
2383 struct strmap_entry
*e
= NULL
;
2385 while ((end
= strrchr(temp
, '/'))) {
2387 e
= strmap_get_entry(dir_renames
, temp
);
2395 static void compute_collisions(struct strmap
*collisions
,
2396 struct strmap
*dir_renames
,
2397 struct diff_queue_struct
*pairs
)
2401 strmap_init_with_options(collisions
, NULL
, 0);
2402 if (strmap_empty(dir_renames
))
2406 * Multiple files can be mapped to the same path due to directory
2407 * renames done by the other side of history. Since that other
2408 * side of history could have merged multiple directories into one,
2409 * if our side of history added the same file basename to each of
2410 * those directories, then all N of them would get implicitly
2411 * renamed by the directory rename detection into the same path,
2412 * and we'd get an add/add/.../add conflict, and all those adds
2413 * from *this* side of history. This is not representable in the
2414 * index, and users aren't going to easily be able to make sense of
2415 * it. So we need to provide a good warning about what's
2416 * happening, and fall back to no-directory-rename detection
2417 * behavior for those paths.
2419 * See testcases 9e and all of section 5 from t6043 for examples.
2421 for (i
= 0; i
< pairs
->nr
; ++i
) {
2422 struct strmap_entry
*rename_info
;
2423 struct collision_info
*collision_info
;
2425 struct diff_filepair
*pair
= pairs
->queue
[i
];
2427 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2429 rename_info
= check_dir_renamed(pair
->two
->path
, dir_renames
);
2433 new_path
= apply_dir_rename(rename_info
, pair
->two
->path
);
2435 collision_info
= strmap_get(collisions
, new_path
);
2436 if (collision_info
) {
2439 CALLOC_ARRAY(collision_info
, 1);
2440 string_list_init_nodup(&collision_info
->source_files
);
2441 strmap_put(collisions
, new_path
, collision_info
);
2443 string_list_insert(&collision_info
->source_files
,
2448 static void free_collisions(struct strmap
*collisions
)
2450 struct hashmap_iter iter
;
2451 struct strmap_entry
*entry
;
2453 /* Free each value in the collisions map */
2454 strmap_for_each_entry(collisions
, &iter
, entry
) {
2455 struct collision_info
*info
= entry
->value
;
2456 string_list_clear(&info
->source_files
, 0);
2459 * In compute_collisions(), we set collisions.strdup_strings to 0
2460 * so that we wouldn't have to make another copy of the new_path
2461 * allocated by apply_dir_rename(). But now that we've used them
2462 * and have no other references to these strings, it is time to
2465 free_strmap_strings(collisions
);
2466 strmap_clear(collisions
, 1);
2469 static char *check_for_directory_rename(struct merge_options
*opt
,
2471 unsigned side_index
,
2472 struct strmap
*dir_renames
,
2473 struct strmap
*dir_rename_exclusions
,
2474 struct strmap
*collisions
,
2478 struct strmap_entry
*rename_info
;
2479 struct strmap_entry
*otherinfo
;
2480 const char *new_dir
;
2481 int other_side
= 3 - side_index
;
2484 * Cases where we don't have or don't want a directory rename for
2487 if (strmap_empty(dir_renames
))
2489 if (strmap_get(&collisions
[other_side
], path
))
2491 rename_info
= check_dir_renamed(path
, dir_renames
);
2496 * This next part is a little weird. We do not want to do an
2497 * implicit rename into a directory we renamed on our side, because
2498 * that will result in a spurious rename/rename(1to2) conflict. An
2500 * Base commit: dumbdir/afile, otherdir/bfile
2501 * Side 1: smrtdir/afile, otherdir/bfile
2502 * Side 2: dumbdir/afile, dumbdir/bfile
2503 * Here, while working on Side 1, we could notice that otherdir was
2504 * renamed/merged to dumbdir, and change the diff_filepair for
2505 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2506 * 2 will notice the rename from dumbdir to smrtdir, and do the
2507 * transitive rename to move it from dumbdir/bfile to
2508 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2509 * smrtdir, a rename/rename(1to2) conflict. We really just want
2510 * the file to end up in smrtdir. And the way to achieve that is
2511 * to not let Side1 do the rename to dumbdir, since we know that is
2512 * the source of one of our directory renames.
2514 * That's why otherinfo and dir_rename_exclusions is here.
2516 * As it turns out, this also prevents N-way transient rename
2517 * confusion; See testcases 9c and 9d of t6043.
2519 new_dir
= rename_info
->value
; /* old_dir = rename_info->key; */
2520 otherinfo
= strmap_get_entry(dir_rename_exclusions
, new_dir
);
2522 path_msg(opt
, INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME
, 1,
2523 rename_info
->key
, path
, new_dir
, NULL
,
2524 _("WARNING: Avoiding applying %s -> %s rename "
2525 "to %s, because %s itself was renamed."),
2526 rename_info
->key
, new_dir
, path
, new_dir
);
2530 new_path
= handle_path_level_conflicts(opt
, path
, side_index
,
2532 &collisions
[side_index
]);
2533 *clean_merge
&= (new_path
!= NULL
);
2538 static void apply_directory_rename_modifications(struct merge_options
*opt
,
2539 struct diff_filepair
*pair
,
2543 * The basic idea is to get the conflict_info from opt->priv->paths
2544 * at old path, and insert it into new_path; basically just this:
2545 * ci = strmap_get(&opt->priv->paths, old_path);
2546 * strmap_remove(&opt->priv->paths, old_path, 0);
2547 * strmap_put(&opt->priv->paths, new_path, ci);
2548 * However, there are some factors complicating this:
2549 * - opt->priv->paths may already have an entry at new_path
2550 * - Each ci tracks its containing directory, so we need to
2552 * - If another ci has the same containing directory, then
2553 * the two char*'s MUST point to the same location. See the
2554 * comment in struct merged_info. strcmp equality is not
2555 * enough; we need pointer equality.
2556 * - opt->priv->paths must hold the parent directories of any
2557 * entries that are added. So, if this directory rename
2558 * causes entirely new directories, we must recursively add
2559 * parent directories.
2560 * - For each parent directory added to opt->priv->paths, we
2561 * also need to get its parent directory stored in its
2562 * conflict_info->merged.directory_name with all the same
2563 * requirements about pointer equality.
2565 struct string_list dirs_to_insert
= STRING_LIST_INIT_NODUP
;
2566 struct conflict_info
*ci
, *new_ci
;
2567 struct strmap_entry
*entry
;
2568 const char *branch_with_new_path
, *branch_with_dir_rename
;
2569 const char *old_path
= pair
->two
->path
;
2570 const char *parent_name
;
2571 const char *cur_path
;
2574 entry
= strmap_get_entry(&opt
->priv
->paths
, old_path
);
2575 old_path
= entry
->key
;
2579 /* Find parent directories missing from opt->priv->paths */
2580 cur_path
= mem_pool_strdup(&opt
->priv
->pool
, new_path
);
2581 free((char*)new_path
);
2582 new_path
= (char *)cur_path
;
2585 /* Find the parent directory of cur_path */
2586 char *last_slash
= strrchr(cur_path
, '/');
2588 parent_name
= mem_pool_strndup(&opt
->priv
->pool
,
2590 last_slash
- cur_path
);
2592 parent_name
= opt
->priv
->toplevel_dir
;
2596 /* Look it up in opt->priv->paths */
2597 entry
= strmap_get_entry(&opt
->priv
->paths
, parent_name
);
2599 parent_name
= entry
->key
; /* reuse known pointer */
2603 /* Record this is one of the directories we need to insert */
2604 string_list_append(&dirs_to_insert
, parent_name
);
2605 cur_path
= parent_name
;
2608 /* Traverse dirs_to_insert and insert them into opt->priv->paths */
2609 for (i
= dirs_to_insert
.nr
-1; i
>= 0; --i
) {
2610 struct conflict_info
*dir_ci
;
2611 char *cur_dir
= dirs_to_insert
.items
[i
].string
;
2613 CALLOC_ARRAY(dir_ci
, 1);
2615 dir_ci
->merged
.directory_name
= parent_name
;
2616 len
= strlen(parent_name
);
2617 /* len+1 because of trailing '/' character */
2618 dir_ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2619 dir_ci
->dirmask
= ci
->filemask
;
2620 strmap_put(&opt
->priv
->paths
, cur_dir
, dir_ci
);
2622 parent_name
= cur_dir
;
2625 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2626 assert(ci
->dirmask
== 0 || ci
->dirmask
== 1);
2627 if (ci
->dirmask
== 0)
2628 strmap_remove(&opt
->priv
->paths
, old_path
, 0);
2631 * This file exists on one side, but we still had a directory
2632 * at the old location that we can't remove until after
2633 * processing all paths below it. So, make a copy of ci in
2634 * new_ci and only put the file information into it.
2636 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
2637 memcpy(new_ci
, ci
, sizeof(*ci
));
2638 assert(!new_ci
->match_mask
);
2639 new_ci
->dirmask
= 0;
2640 new_ci
->stages
[1].mode
= 0;
2641 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
2644 * Now that we have the file information in new_ci, make sure
2645 * ci only has the directory information.
2648 ci
->merged
.clean
= 1;
2649 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
2650 if (ci
->dirmask
& (1 << i
))
2652 /* zero out any entries related to files */
2653 ci
->stages
[i
].mode
= 0;
2654 oidcpy(&ci
->stages
[i
].oid
, null_oid());
2657 // Now we want to focus on new_ci, so reassign ci to it
2661 branch_with_new_path
= (ci
->filemask
== 2) ? opt
->branch1
: opt
->branch2
;
2662 branch_with_dir_rename
= (ci
->filemask
== 2) ? opt
->branch2
: opt
->branch1
;
2664 /* Now, finally update ci and stick it into opt->priv->paths */
2665 ci
->merged
.directory_name
= parent_name
;
2666 len
= strlen(parent_name
);
2667 ci
->merged
.basename_offset
= (len
> 0 ? len
+1 : len
);
2668 new_ci
= strmap_get(&opt
->priv
->paths
, new_path
);
2670 /* Place ci back into opt->priv->paths, but at new_path */
2671 strmap_put(&opt
->priv
->paths
, new_path
, ci
);
2675 /* A few sanity checks */
2677 assert(ci
->filemask
== 2 || ci
->filemask
== 4);
2678 assert((new_ci
->filemask
& ci
->filemask
) == 0);
2679 assert(!new_ci
->merged
.clean
);
2681 /* Copy stuff from ci into new_ci */
2682 new_ci
->filemask
|= ci
->filemask
;
2683 if (new_ci
->dirmask
)
2684 new_ci
->df_conflict
= 1;
2685 index
= (ci
->filemask
>> 1);
2686 new_ci
->pathnames
[index
] = ci
->pathnames
[index
];
2687 new_ci
->stages
[index
].mode
= ci
->stages
[index
].mode
;
2688 oidcpy(&new_ci
->stages
[index
].oid
, &ci
->stages
[index
].oid
);
2693 if (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
) {
2694 /* Notify user of updated path */
2695 if (pair
->status
== 'A')
2696 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2697 new_path
, old_path
, NULL
, NULL
,
2698 _("Path updated: %s added in %s inside a "
2699 "directory that was renamed in %s; moving "
2701 old_path
, branch_with_new_path
,
2702 branch_with_dir_rename
, new_path
);
2704 path_msg(opt
, INFO_DIR_RENAME_APPLIED
, 1,
2705 new_path
, old_path
, NULL
, NULL
,
2706 _("Path updated: %s renamed to %s in %s, "
2707 "inside a directory that was renamed in %s; "
2708 "moving it to %s."),
2709 pair
->one
->path
, old_path
, branch_with_new_path
,
2710 branch_with_dir_rename
, new_path
);
2713 * opt->detect_directory_renames has the value
2714 * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts.
2716 ci
->path_conflict
= 1;
2717 if (pair
->status
== 'A')
2718 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2719 new_path
, old_path
, NULL
, NULL
,
2720 _("CONFLICT (file location): %s added in %s "
2721 "inside a directory that was renamed in %s, "
2722 "suggesting it should perhaps be moved to "
2724 old_path
, branch_with_new_path
,
2725 branch_with_dir_rename
, new_path
);
2727 path_msg(opt
, CONFLICT_DIR_RENAME_SUGGESTED
, 1,
2728 new_path
, old_path
, NULL
, NULL
,
2729 _("CONFLICT (file location): %s renamed to %s "
2730 "in %s, inside a directory that was renamed "
2731 "in %s, suggesting it should perhaps be "
2733 pair
->one
->path
, old_path
, branch_with_new_path
,
2734 branch_with_dir_rename
, new_path
);
2738 * Finally, record the new location.
2740 pair
->two
->path
= new_path
;
2743 /*** Function Grouping: functions related to regular rename detection ***/
2745 static int process_renames(struct merge_options
*opt
,
2746 struct diff_queue_struct
*renames
)
2748 int clean_merge
= 1, i
;
2750 for (i
= 0; i
< renames
->nr
; ++i
) {
2751 const char *oldpath
= NULL
, *newpath
;
2752 struct diff_filepair
*pair
= renames
->queue
[i
];
2753 struct conflict_info
*oldinfo
= NULL
, *newinfo
= NULL
;
2754 struct strmap_entry
*old_ent
, *new_ent
;
2755 unsigned int old_sidemask
;
2756 int target_index
, other_source_index
;
2757 int source_deleted
, collision
, type_changed
;
2758 const char *rename_branch
= NULL
, *delete_branch
= NULL
;
2760 old_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->one
->path
);
2761 new_ent
= strmap_get_entry(&opt
->priv
->paths
, pair
->two
->path
);
2763 oldpath
= old_ent
->key
;
2764 oldinfo
= old_ent
->value
;
2766 newpath
= pair
->two
->path
;
2768 newpath
= new_ent
->key
;
2769 newinfo
= new_ent
->value
;
2773 * If pair->one->path isn't in opt->priv->paths, that means
2774 * that either directory rename detection removed that
2775 * path, or a parent directory of oldpath was resolved and
2776 * we don't even need the rename; in either case, we can
2777 * skip it. If oldinfo->merged.clean, then the other side
2778 * of history had no changes to oldpath and we don't need
2779 * the rename and can skip it.
2781 if (!oldinfo
|| oldinfo
->merged
.clean
)
2785 * diff_filepairs have copies of pathnames, thus we have to
2786 * use standard 'strcmp()' (negated) instead of '=='.
2788 if (i
+ 1 < renames
->nr
&&
2789 !strcmp(oldpath
, renames
->queue
[i
+1]->one
->path
)) {
2790 /* Handle rename/rename(1to2) or rename/rename(1to1) */
2791 const char *pathnames
[3];
2792 struct version_info merged
;
2793 struct conflict_info
*base
, *side1
, *side2
;
2794 unsigned was_binary_blob
= 0;
2796 pathnames
[0] = oldpath
;
2797 pathnames
[1] = newpath
;
2798 pathnames
[2] = renames
->queue
[i
+1]->two
->path
;
2800 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2801 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2802 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2808 if (!strcmp(pathnames
[1], pathnames
[2])) {
2809 struct rename_info
*ri
= &opt
->priv
->renames
;
2812 /* Both sides renamed the same way */
2813 assert(side1
== side2
);
2814 memcpy(&side1
->stages
[0], &base
->stages
[0],
2816 side1
->filemask
|= (1 << MERGE_BASE
);
2817 /* Mark base as resolved by removal */
2818 base
->merged
.is_null
= 1;
2819 base
->merged
.clean
= 1;
2822 * Disable remembering renames optimization;
2823 * rename/rename(1to1) is incredibly rare, and
2824 * just disabling the optimization is easier
2825 * than purging cached_pairs,
2826 * cached_target_names, and dir_rename_counts.
2828 for (j
= 0; j
< 3; j
++)
2829 ri
->merge_trees
[j
] = NULL
;
2831 /* We handled both renames, i.e. i+1 handled */
2833 /* Move to next rename */
2837 /* This is a rename/rename(1to2) */
2838 clean_merge
= handle_content_merge(opt
,
2844 1 + 2 * opt
->priv
->call_depth
,
2846 if (clean_merge
< 0)
2849 merged
.mode
== side1
->stages
[1].mode
&&
2850 oideq(&merged
.oid
, &side1
->stages
[1].oid
))
2851 was_binary_blob
= 1;
2852 memcpy(&side1
->stages
[1], &merged
, sizeof(merged
));
2853 if (was_binary_blob
) {
2855 * Getting here means we were attempting to
2856 * merge a binary blob.
2858 * Since we can't merge binaries,
2859 * handle_content_merge() just takes one
2860 * side. But we don't want to copy the
2861 * contents of one side to both paths. We
2862 * used the contents of side1 above for
2863 * side1->stages, let's use the contents of
2864 * side2 for side2->stages below.
2866 oidcpy(&merged
.oid
, &side2
->stages
[2].oid
);
2867 merged
.mode
= side2
->stages
[2].mode
;
2869 memcpy(&side2
->stages
[2], &merged
, sizeof(merged
));
2871 side1
->path_conflict
= 1;
2872 side2
->path_conflict
= 1;
2874 * TODO: For renames we normally remove the path at the
2875 * old name. It would thus seem consistent to do the
2876 * same for rename/rename(1to2) cases, but we haven't
2877 * done so traditionally and a number of the regression
2878 * tests now encode an expectation that the file is
2879 * left there at stage 1. If we ever decide to change
2880 * this, add the following two lines here:
2881 * base->merged.is_null = 1;
2882 * base->merged.clean = 1;
2883 * and remove the setting of base->path_conflict to 1.
2885 base
->path_conflict
= 1;
2886 path_msg(opt
, CONFLICT_RENAME_RENAME
, 0,
2887 pathnames
[0], pathnames
[1], pathnames
[2], NULL
,
2888 _("CONFLICT (rename/rename): %s renamed to "
2889 "%s in %s and to %s in %s."),
2891 pathnames
[1], opt
->branch1
,
2892 pathnames
[2], opt
->branch2
);
2894 i
++; /* We handled both renames, i.e. i+1 handled */
2900 target_index
= pair
->score
; /* from collect_renames() */
2901 assert(target_index
== 1 || target_index
== 2);
2902 other_source_index
= 3 - target_index
;
2903 old_sidemask
= (1 << other_source_index
); /* 2 or 4 */
2904 source_deleted
= (oldinfo
->filemask
== 1);
2905 collision
= ((newinfo
->filemask
& old_sidemask
) != 0);
2906 type_changed
= !source_deleted
&&
2907 (S_ISREG(oldinfo
->stages
[other_source_index
].mode
) !=
2908 S_ISREG(newinfo
->stages
[target_index
].mode
));
2909 if (type_changed
&& collision
) {
2911 * special handling so later blocks can handle this...
2913 * if type_changed && collision are both true, then this
2914 * was really a double rename, but one side wasn't
2915 * detected due to lack of break detection. I.e.
2917 * orig: has normal file 'foo'
2918 * side1: renames 'foo' to 'bar', adds 'foo' symlink
2919 * side2: renames 'foo' to 'bar'
2920 * In this case, the foo->bar rename on side1 won't be
2921 * detected because the new symlink named 'foo' is
2922 * there and we don't do break detection. But we detect
2923 * this here because we don't want to merge the content
2924 * of the foo symlink with the foo->bar file, so we
2925 * have some logic to handle this special case. The
2926 * easiest way to do that is make 'bar' on side1 not
2927 * be considered a colliding file but the other part
2928 * of a normal rename. If the file is very different,
2929 * well we're going to get content merge conflicts
2930 * anyway so it doesn't hurt. And if the colliding
2931 * file also has a different type, that'll be handled
2932 * by the content merge logic in process_entry() too.
2934 * See also t6430, 'rename vs. rename/symlink'
2938 if (source_deleted
) {
2939 if (target_index
== 1) {
2940 rename_branch
= opt
->branch1
;
2941 delete_branch
= opt
->branch2
;
2943 rename_branch
= opt
->branch2
;
2944 delete_branch
= opt
->branch1
;
2948 assert(source_deleted
|| oldinfo
->filemask
& old_sidemask
);
2950 /* Need to check for special types of rename conflicts... */
2951 if (collision
&& !source_deleted
) {
2952 /* collision: rename/add or rename/rename(2to1) */
2953 const char *pathnames
[3];
2954 struct version_info merged
;
2956 struct conflict_info
*base
, *side1
, *side2
;
2959 pathnames
[0] = oldpath
;
2960 pathnames
[other_source_index
] = oldpath
;
2961 pathnames
[target_index
] = newpath
;
2963 base
= strmap_get(&opt
->priv
->paths
, pathnames
[0]);
2964 side1
= strmap_get(&opt
->priv
->paths
, pathnames
[1]);
2965 side2
= strmap_get(&opt
->priv
->paths
, pathnames
[2]);
2971 clean
= handle_content_merge(opt
, pair
->one
->path
,
2976 1 + 2 * opt
->priv
->call_depth
,
2981 memcpy(&newinfo
->stages
[target_index
], &merged
,
2984 path_msg(opt
, CONFLICT_RENAME_COLLIDES
, 0,
2985 newpath
, oldpath
, NULL
, NULL
,
2986 _("CONFLICT (rename involved in "
2987 "collision): rename of %s -> %s has "
2988 "content conflicts AND collides "
2989 "with another path; this may result "
2990 "in nested conflict markers."),
2993 } else if (collision
&& source_deleted
) {
2995 * rename/add/delete or rename/rename(2to1)/delete:
2996 * since oldpath was deleted on the side that didn't
2997 * do the rename, there's not much of a content merge
2998 * we can do for the rename. oldinfo->merged.is_null
2999 * was already set, so we just leave things as-is so
3000 * they look like an add/add conflict.
3003 newinfo
->path_conflict
= 1;
3004 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
3005 newpath
, oldpath
, NULL
, NULL
,
3006 _("CONFLICT (rename/delete): %s renamed "
3007 "to %s in %s, but deleted in %s."),
3008 oldpath
, newpath
, rename_branch
, delete_branch
);
3011 * a few different cases...start by copying the
3012 * existing stage(s) from oldinfo over the newinfo
3013 * and update the pathname(s).
3015 memcpy(&newinfo
->stages
[0], &oldinfo
->stages
[0],
3016 sizeof(newinfo
->stages
[0]));
3017 newinfo
->filemask
|= (1 << MERGE_BASE
);
3018 newinfo
->pathnames
[0] = oldpath
;
3020 /* rename vs. typechange */
3021 /* Mark the original as resolved by removal */
3022 memcpy(&oldinfo
->stages
[0].oid
, null_oid(),
3023 sizeof(oldinfo
->stages
[0].oid
));
3024 oldinfo
->stages
[0].mode
= 0;
3025 oldinfo
->filemask
&= 0x06;
3026 } else if (source_deleted
) {
3028 newinfo
->path_conflict
= 1;
3029 path_msg(opt
, CONFLICT_RENAME_DELETE
, 0,
3030 newpath
, oldpath
, NULL
, NULL
,
3031 _("CONFLICT (rename/delete): %s renamed"
3032 " to %s in %s, but deleted in %s."),
3034 rename_branch
, delete_branch
);
3037 memcpy(&newinfo
->stages
[other_source_index
],
3038 &oldinfo
->stages
[other_source_index
],
3039 sizeof(newinfo
->stages
[0]));
3040 newinfo
->filemask
|= (1 << other_source_index
);
3041 newinfo
->pathnames
[other_source_index
] = oldpath
;
3045 if (!type_changed
) {
3046 /* Mark the original as resolved by removal */
3047 oldinfo
->merged
.is_null
= 1;
3048 oldinfo
->merged
.clean
= 1;
3056 static inline int possible_side_renames(struct rename_info
*renames
,
3057 unsigned side_index
)
3059 return renames
->pairs
[side_index
].nr
> 0 &&
3060 !strintmap_empty(&renames
->relevant_sources
[side_index
]);
3063 static inline int possible_renames(struct rename_info
*renames
)
3065 return possible_side_renames(renames
, 1) ||
3066 possible_side_renames(renames
, 2) ||
3067 !strmap_empty(&renames
->cached_pairs
[1]) ||
3068 !strmap_empty(&renames
->cached_pairs
[2]);
3071 static void resolve_diffpair_statuses(struct diff_queue_struct
*q
)
3074 * A simplified version of diff_resolve_rename_copy(); would probably
3075 * just use that function but it's static...
3078 struct diff_filepair
*p
;
3080 for (i
= 0; i
< q
->nr
; ++i
) {
3082 p
->status
= 0; /* undecided */
3083 if (!DIFF_FILE_VALID(p
->one
))
3084 p
->status
= DIFF_STATUS_ADDED
;
3085 else if (!DIFF_FILE_VALID(p
->two
))
3086 p
->status
= DIFF_STATUS_DELETED
;
3087 else if (DIFF_PAIR_RENAME(p
))
3088 p
->status
= DIFF_STATUS_RENAMED
;
3092 static void prune_cached_from_relevant(struct rename_info
*renames
,
3095 /* Reason for this function described in add_pair() */
3096 struct hashmap_iter iter
;
3097 struct strmap_entry
*entry
;
3099 /* Remove from relevant_sources all entries in cached_pairs[side] */
3100 strmap_for_each_entry(&renames
->cached_pairs
[side
], &iter
, entry
) {
3101 strintmap_remove(&renames
->relevant_sources
[side
],
3104 /* Remove from relevant_sources all entries in cached_irrelevant[side] */
3105 strset_for_each_entry(&renames
->cached_irrelevant
[side
], &iter
, entry
) {
3106 strintmap_remove(&renames
->relevant_sources
[side
],
3111 static void use_cached_pairs(struct merge_options
*opt
,
3112 struct strmap
*cached_pairs
,
3113 struct diff_queue_struct
*pairs
)
3115 struct hashmap_iter iter
;
3116 struct strmap_entry
*entry
;
3119 * Add to side_pairs all entries from renames->cached_pairs[side_index].
3120 * (Info in cached_irrelevant[side_index] is not relevant here.)
3122 strmap_for_each_entry(cached_pairs
, &iter
, entry
) {
3123 struct diff_filespec
*one
, *two
;
3124 const char *old_name
= entry
->key
;
3125 const char *new_name
= entry
->value
;
3127 new_name
= old_name
;
3130 * cached_pairs has *copies* of old_name and new_name,
3131 * because it has to persist across merges. Since
3132 * pool_alloc_filespec() will just re-use the existing
3133 * filenames, which will also get re-used by
3134 * opt->priv->paths if they become renames, and then
3135 * get freed at the end of the merge, that would leave
3136 * the copy in cached_pairs dangling. Avoid this by
3137 * making a copy here.
3139 old_name
= mem_pool_strdup(&opt
->priv
->pool
, old_name
);
3140 new_name
= mem_pool_strdup(&opt
->priv
->pool
, new_name
);
3142 /* We don't care about oid/mode, only filenames and status */
3143 one
= pool_alloc_filespec(&opt
->priv
->pool
, old_name
);
3144 two
= pool_alloc_filespec(&opt
->priv
->pool
, new_name
);
3145 pool_diff_queue(&opt
->priv
->pool
, pairs
, one
, two
);
3146 pairs
->queue
[pairs
->nr
-1]->status
= entry
->value
? 'R' : 'D';
3150 static void cache_new_pair(struct rename_info
*renames
,
3157 new_path
= xstrdup(new_path
);
3158 old_value
= strmap_put(&renames
->cached_pairs
[side
],
3159 old_path
, new_path
);
3160 strset_add(&renames
->cached_target_names
[side
], new_path
);
3167 static void possibly_cache_new_pair(struct rename_info
*renames
,
3168 struct diff_filepair
*p
,
3172 int dir_renamed_side
= 0;
3176 * Directory renames happen on the other side of history from
3177 * the side that adds new files to the old directory.
3179 dir_renamed_side
= 3 - side
;
3181 int val
= strintmap_get(&renames
->relevant_sources
[side
],
3183 if (val
== RELEVANT_NO_MORE
) {
3184 assert(p
->status
== 'D');
3185 strset_add(&renames
->cached_irrelevant
[side
],
3192 if (p
->status
== 'D') {
3194 * If we already had this delete, we'll just set it's value
3195 * to NULL again, so no harm.
3197 strmap_put(&renames
->cached_pairs
[side
], p
->one
->path
, NULL
);
3198 } else if (p
->status
== 'R') {
3200 new_path
= p
->two
->path
;
3202 cache_new_pair(renames
, dir_renamed_side
,
3203 p
->two
->path
, new_path
, 0);
3204 cache_new_pair(renames
, side
, p
->one
->path
, new_path
, 1);
3205 } else if (p
->status
== 'A' && new_path
) {
3206 cache_new_pair(renames
, dir_renamed_side
,
3207 p
->two
->path
, new_path
, 0);
3211 static int compare_pairs(const void *a_
, const void *b_
)
3213 const struct diff_filepair
*a
= *((const struct diff_filepair
**)a_
);
3214 const struct diff_filepair
*b
= *((const struct diff_filepair
**)b_
);
3216 return strcmp(a
->one
->path
, b
->one
->path
);
3219 /* Call diffcore_rename() to update deleted/added pairs into rename pairs */
3220 static int detect_regular_renames(struct merge_options
*opt
,
3221 unsigned side_index
)
3223 struct diff_options diff_opts
;
3224 struct rename_info
*renames
= &opt
->priv
->renames
;
3226 prune_cached_from_relevant(renames
, side_index
);
3227 if (!possible_side_renames(renames
, side_index
)) {
3229 * No rename detection needed for this side, but we still need
3230 * to make sure 'adds' are marked correctly in case the other
3231 * side had directory renames.
3233 resolve_diffpair_statuses(&renames
->pairs
[side_index
]);
3237 partial_clear_dir_rename_count(&renames
->dir_rename_count
[side_index
]);
3238 repo_diff_setup(opt
->repo
, &diff_opts
);
3239 diff_opts
.flags
.recursive
= 1;
3240 diff_opts
.flags
.rename_empty
= 0;
3241 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
3242 diff_opts
.rename_limit
= opt
->rename_limit
;
3243 if (opt
->rename_limit
<= 0)
3244 diff_opts
.rename_limit
= 7000;
3245 diff_opts
.rename_score
= opt
->rename_score
;
3246 diff_opts
.show_rename_progress
= opt
->show_rename_progress
;
3247 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3248 diff_setup_done(&diff_opts
);
3250 diff_queued_diff
= renames
->pairs
[side_index
];
3251 trace2_region_enter("diff", "diffcore_rename", opt
->repo
);
3252 diffcore_rename_extended(&diff_opts
,
3254 &renames
->relevant_sources
[side_index
],
3255 &renames
->dirs_removed
[side_index
],
3256 &renames
->dir_rename_count
[side_index
],
3257 &renames
->cached_pairs
[side_index
]);
3258 trace2_region_leave("diff", "diffcore_rename", opt
->repo
);
3259 resolve_diffpair_statuses(&diff_queued_diff
);
3261 if (diff_opts
.needed_rename_limit
> 0)
3262 renames
->redo_after_renames
= 0;
3263 if (diff_opts
.needed_rename_limit
> renames
->needed_limit
)
3264 renames
->needed_limit
= diff_opts
.needed_rename_limit
;
3266 renames
->pairs
[side_index
] = diff_queued_diff
;
3268 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
3269 diff_queued_diff
.nr
= 0;
3270 diff_queued_diff
.queue
= NULL
;
3271 diff_flush(&diff_opts
);
3277 * Get information of all renames which occurred in 'side_pairs', making use
3278 * of any implicit directory renames in side_dir_renames (also making use of
3279 * implicit directory renames rename_exclusions as needed by
3280 * check_for_directory_rename()). Add all (updated) renames into result.
3282 static int collect_renames(struct merge_options
*opt
,
3283 struct diff_queue_struct
*result
,
3284 unsigned side_index
,
3285 struct strmap
*collisions
,
3286 struct strmap
*dir_renames_for_side
,
3287 struct strmap
*rename_exclusions
)
3290 struct diff_queue_struct
*side_pairs
;
3291 struct rename_info
*renames
= &opt
->priv
->renames
;
3293 side_pairs
= &renames
->pairs
[side_index
];
3295 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3296 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3297 char *new_path
; /* non-NULL only with directory renames */
3299 if (p
->status
!= 'A' && p
->status
!= 'R') {
3300 possibly_cache_new_pair(renames
, p
, side_index
, NULL
);
3301 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3305 new_path
= check_for_directory_rename(opt
, p
->two
->path
,
3307 dir_renames_for_side
,
3312 possibly_cache_new_pair(renames
, p
, side_index
, new_path
);
3313 if (p
->status
!= 'R' && !new_path
) {
3314 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3319 apply_directory_rename_modifications(opt
, p
, new_path
);
3322 * p->score comes back from diffcore_rename_extended() with
3323 * the similarity of the renamed file. The similarity is
3324 * was used to determine that the two files were related
3325 * and are a rename, which we have already used, but beyond
3326 * that we have no use for the similarity. So p->score is
3327 * now irrelevant. However, process_renames() will need to
3328 * know which side of the merge this rename was associated
3329 * with, so overwrite p->score with that value.
3331 p
->score
= side_index
;
3332 result
->queue
[result
->nr
++] = p
;
3338 static int detect_and_process_renames(struct merge_options
*opt
,
3339 struct tree
*merge_base
,
3343 struct diff_queue_struct combined
= { 0 };
3344 struct rename_info
*renames
= &opt
->priv
->renames
;
3345 struct strmap collisions
[3];
3346 int need_dir_renames
, s
, i
, clean
= 1;
3347 unsigned detection_run
= 0;
3349 if (!possible_renames(renames
))
3352 trace2_region_enter("merge", "regular renames", opt
->repo
);
3353 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE1
);
3354 detection_run
|= detect_regular_renames(opt
, MERGE_SIDE2
);
3355 if (renames
->needed_limit
) {
3356 renames
->cached_pairs_valid_side
= 0;
3357 renames
->redo_after_renames
= 0;
3359 if (renames
->redo_after_renames
&& detection_run
) {
3361 struct diff_filepair
*p
;
3363 /* Cache the renames, we found */
3364 for (side
= MERGE_SIDE1
; side
<= MERGE_SIDE2
; side
++) {
3365 for (i
= 0; i
< renames
->pairs
[side
].nr
; ++i
) {
3366 p
= renames
->pairs
[side
].queue
[i
];
3367 possibly_cache_new_pair(renames
, p
, side
, NULL
);
3371 /* Restart the merge with the cached renames */
3372 renames
->redo_after_renames
= 2;
3373 trace2_region_leave("merge", "regular renames", opt
->repo
);
3376 use_cached_pairs(opt
, &renames
->cached_pairs
[1], &renames
->pairs
[1]);
3377 use_cached_pairs(opt
, &renames
->cached_pairs
[2], &renames
->pairs
[2]);
3378 trace2_region_leave("merge", "regular renames", opt
->repo
);
3380 trace2_region_enter("merge", "directory renames", opt
->repo
);
3382 !opt
->priv
->call_depth
&&
3383 (opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_TRUE
||
3384 opt
->detect_directory_renames
== MERGE_DIRECTORY_RENAMES_CONFLICT
);
3386 if (need_dir_renames
) {
3387 get_provisional_directory_renames(opt
, MERGE_SIDE1
, &clean
);
3388 get_provisional_directory_renames(opt
, MERGE_SIDE2
, &clean
);
3389 handle_directory_level_conflicts(opt
);
3392 ALLOC_GROW(combined
.queue
,
3393 renames
->pairs
[1].nr
+ renames
->pairs
[2].nr
,
3395 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
3396 int other_side
= 3 - i
;
3397 compute_collisions(&collisions
[i
],
3398 &renames
->dir_renames
[other_side
],
3399 &renames
->pairs
[i
]);
3401 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE1
,
3403 &renames
->dir_renames
[2],
3404 &renames
->dir_renames
[1]);
3405 clean
&= collect_renames(opt
, &combined
, MERGE_SIDE2
,
3407 &renames
->dir_renames
[1],
3408 &renames
->dir_renames
[2]);
3409 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++)
3410 free_collisions(&collisions
[i
]);
3411 STABLE_QSORT(combined
.queue
, combined
.nr
, compare_pairs
);
3412 trace2_region_leave("merge", "directory renames", opt
->repo
);
3414 trace2_region_enter("merge", "process renames", opt
->repo
);
3415 clean
&= process_renames(opt
, &combined
);
3416 trace2_region_leave("merge", "process renames", opt
->repo
);
3418 goto simple_cleanup
; /* collect_renames() handles some of cleanup */
3422 * Free now unneeded filepairs, which would have been handled
3423 * in collect_renames() normally but we skipped that code.
3425 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3426 struct diff_queue_struct
*side_pairs
;
3429 side_pairs
= &renames
->pairs
[s
];
3430 for (i
= 0; i
< side_pairs
->nr
; ++i
) {
3431 struct diff_filepair
*p
= side_pairs
->queue
[i
];
3432 pool_diff_free_filepair(&opt
->priv
->pool
, p
);
3437 /* Free memory for renames->pairs[] and combined */
3438 for (s
= MERGE_SIDE1
; s
<= MERGE_SIDE2
; s
++) {
3439 free(renames
->pairs
[s
].queue
);
3440 DIFF_QUEUE_CLEAR(&renames
->pairs
[s
]);
3442 for (i
= 0; i
< combined
.nr
; i
++)
3443 pool_diff_free_filepair(&opt
->priv
->pool
, combined
.queue
[i
]);
3444 free(combined
.queue
);
3449 /*** Function Grouping: functions related to process_entries() ***/
3451 static int sort_dirs_next_to_their_children(const char *one
, const char *two
)
3453 unsigned char c1
, c2
;
3456 * Here we only care that entries for directories appear adjacent
3457 * to and before files underneath the directory. We can achieve
3458 * that by pretending to add a trailing slash to every file and
3459 * then sorting. In other words, we do not want the natural
3464 * Instead, we want "foo" to sort as though it were "foo/", so that
3469 * To achieve this, we basically implement our own strcmp, except that
3470 * if we get to the end of either string instead of comparing NUL to
3471 * another character, we compare '/' to it.
3473 * If this unusual "sort as though '/' were appended" perplexes
3474 * you, perhaps it will help to note that this is not the final
3475 * sort. write_tree() will sort again without the trailing slash
3476 * magic, but just on paths immediately under a given tree.
3478 * The reason to not use df_name_compare directly was that it was
3479 * just too expensive (we don't have the string lengths handy), so
3480 * it was reimplemented.
3484 * NOTE: This function will never be called with two equal strings,
3485 * because it is used to sort the keys of a strmap, and strmaps have
3486 * unique keys by construction. That simplifies our c1==c2 handling
3490 while (*one
&& (*one
== *two
)) {
3495 c1
= *one
? *one
: '/';
3496 c2
= *two
? *two
: '/';
3499 /* Getting here means one is a leading directory of the other */
3500 return (*one
) ? 1 : -1;
3505 static int read_oid_strbuf(struct merge_options
*opt
,
3506 const struct object_id
*oid
,
3510 enum object_type type
;
3512 buf
= repo_read_object_file(the_repository
, oid
, &type
, &size
);
3514 return err(opt
, _("cannot read object %s"), oid_to_hex(oid
));
3515 if (type
!= OBJ_BLOB
) {
3517 return err(opt
, _("object %s is not a blob"), oid_to_hex(oid
));
3519 strbuf_attach(dst
, buf
, size
, size
+ 1);
3523 static int blob_unchanged(struct merge_options
*opt
,
3524 const struct version_info
*base
,
3525 const struct version_info
*side
,
3528 struct strbuf basebuf
= STRBUF_INIT
;
3529 struct strbuf sidebuf
= STRBUF_INIT
;
3530 int ret
= 0; /* assume changed for safety */
3531 struct index_state
*idx
= &opt
->priv
->attr_index
;
3533 if (!idx
->initialized
)
3534 initialize_attr_index(opt
);
3536 if (base
->mode
!= side
->mode
)
3538 if (oideq(&base
->oid
, &side
->oid
))
3541 if (read_oid_strbuf(opt
, &base
->oid
, &basebuf
) ||
3542 read_oid_strbuf(opt
, &side
->oid
, &sidebuf
))
3545 * Note: binary | is used so that both renormalizations are
3546 * performed. Comparison can be skipped if both files are
3547 * unchanged since their sha1s have already been compared.
3549 if (renormalize_buffer(idx
, path
, basebuf
.buf
, basebuf
.len
, &basebuf
) |
3550 renormalize_buffer(idx
, path
, sidebuf
.buf
, sidebuf
.len
, &sidebuf
))
3551 ret
= (basebuf
.len
== sidebuf
.len
&&
3552 !memcmp(basebuf
.buf
, sidebuf
.buf
, basebuf
.len
));
3555 strbuf_release(&basebuf
);
3556 strbuf_release(&sidebuf
);
3560 struct directory_versions
{
3562 * versions: list of (basename -> version_info)
3564 * The basenames are in reverse lexicographic order of full pathnames,
3565 * as processed in process_entries(). This puts all entries within
3566 * a directory together, and covers the directory itself after
3567 * everything within it, allowing us to write subtrees before needing
3568 * to record information for the tree itself.
3570 struct string_list versions
;
3573 * offsets: list of (full relative path directories -> integer offsets)
3575 * Since versions contains basenames from files in multiple different
3576 * directories, we need to know which entries in versions correspond
3577 * to which directories. Values of e.g.
3581 * Would mean that entries 0-1 of versions are files in the toplevel
3582 * directory, entries 2-4 are files under src/, and the remaining
3583 * entries starting at index 5 are files under src/moduleA/.
3585 struct string_list offsets
;
3588 * last_directory: directory that previously processed file found in
3590 * last_directory starts NULL, but records the directory in which the
3591 * previous file was found within. As soon as
3592 * directory(current_file) != last_directory
3593 * then we need to start updating accounting in versions & offsets.
3594 * Note that last_directory is always the last path in "offsets" (or
3595 * NULL if "offsets" is empty) so this exists just for quick access.
3597 const char *last_directory
;
3599 /* last_directory_len: cached computation of strlen(last_directory) */
3600 unsigned last_directory_len
;
3603 static int tree_entry_order(const void *a_
, const void *b_
)
3605 const struct string_list_item
*a
= a_
;
3606 const struct string_list_item
*b
= b_
;
3608 const struct merged_info
*ami
= a
->util
;
3609 const struct merged_info
*bmi
= b
->util
;
3610 return base_name_compare(a
->string
, strlen(a
->string
), ami
->result
.mode
,
3611 b
->string
, strlen(b
->string
), bmi
->result
.mode
);
3614 static int write_tree(struct object_id
*result_oid
,
3615 struct string_list
*versions
,
3616 unsigned int offset
,
3619 size_t maxlen
= 0, extra
;
3621 struct strbuf buf
= STRBUF_INIT
;
3624 assert(offset
<= versions
->nr
);
3625 nr
= versions
->nr
- offset
;
3627 /* No need for STABLE_QSORT -- filenames must be unique */
3628 QSORT(versions
->items
+ offset
, nr
, tree_entry_order
);
3630 /* Pre-allocate some space in buf */
3631 extra
= hash_size
+ 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
3632 for (i
= 0; i
< nr
; i
++) {
3633 maxlen
+= strlen(versions
->items
[offset
+i
].string
) + extra
;
3635 strbuf_grow(&buf
, maxlen
);
3637 /* Write each entry out to buf */
3638 for (i
= 0; i
< nr
; i
++) {
3639 struct merged_info
*mi
= versions
->items
[offset
+i
].util
;
3640 struct version_info
*ri
= &mi
->result
;
3641 strbuf_addf(&buf
, "%o %s%c",
3643 versions
->items
[offset
+i
].string
, '\0');
3644 strbuf_add(&buf
, ri
->oid
.hash
, hash_size
);
3647 /* Write this object file out, and record in result_oid */
3648 if (write_object_file(buf
.buf
, buf
.len
, OBJ_TREE
, result_oid
))
3650 strbuf_release(&buf
);
3654 static void record_entry_for_tree(struct directory_versions
*dir_metadata
,
3656 struct merged_info
*mi
)
3658 const char *basename
;
3661 /* nothing to record */
3664 basename
= path
+ mi
->basename_offset
;
3665 assert(strchr(basename
, '/') == NULL
);
3666 string_list_append(&dir_metadata
->versions
,
3667 basename
)->util
= &mi
->result
;
3670 static int write_completed_directory(struct merge_options
*opt
,
3671 const char *new_directory_name
,
3672 struct directory_versions
*info
)
3674 const char *prev_dir
;
3675 struct merged_info
*dir_info
= NULL
;
3676 unsigned int offset
, ret
= 0;
3679 * Some explanation of info->versions and info->offsets...
3681 * process_entries() iterates over all relevant files AND
3682 * directories in reverse lexicographic order, and calls this
3683 * function. Thus, an example of the paths that process_entries()
3684 * could operate on (along with the directories for those paths
3689 * src/moduleB/umm.c src/moduleB
3690 * src/moduleB/stuff.h src/moduleB
3691 * src/moduleB/baz.c src/moduleB
3693 * src/moduleA/foo.c src/moduleA
3694 * src/moduleA/bar.c src/moduleA
3701 * always contains the unprocessed entries and their
3702 * version_info information. For example, after the first five
3703 * entries above, info->versions would be:
3705 * xtract.c <xtract.c's version_info>
3706 * token.txt <token.txt's version_info>
3707 * umm.c <src/moduleB/umm.c's version_info>
3708 * stuff.h <src/moduleB/stuff.h's version_info>
3709 * baz.c <src/moduleB/baz.c's version_info>
3711 * Once a subdirectory is completed we remove the entries in
3712 * that subdirectory from info->versions, writing it as a tree
3713 * (write_tree()). Thus, as soon as we get to src/moduleB,
3714 * info->versions would be updated to
3716 * xtract.c <xtract.c's version_info>
3717 * token.txt <token.txt's version_info>
3718 * moduleB <src/moduleB's version_info>
3722 * helps us track which entries in info->versions correspond to
3723 * which directories. When we are N directories deep (e.g. 4
3724 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
3725 * directories (+1 because of toplevel dir). Corresponding to
3726 * the info->versions example above, after processing five entries
3727 * info->offsets will be:
3732 * which is used to know that xtract.c & token.txt are from the
3733 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
3734 * src/moduleB directory. Again, following the example above,
3735 * once we need to process src/moduleB, then info->offsets is
3741 * which says that moduleB (and only moduleB so far) is in the
3744 * One unique thing to note about info->offsets here is that
3745 * "src" was not added to info->offsets until there was a path
3746 * (a file OR directory) immediately below src/ that got
3749 * Since process_entry() just appends new entries to info->versions,
3750 * write_completed_directory() only needs to do work if the next path
3751 * is in a directory that is different than the last directory found
3756 * If we are working with the same directory as the last entry, there
3757 * is no work to do. (See comments above the directory_name member of
3758 * struct merged_info for why we can use pointer comparison instead of
3761 if (new_directory_name
== info
->last_directory
)
3765 * If we are just starting (last_directory is NULL), or last_directory
3766 * is a prefix of the current directory, then we can just update
3767 * info->offsets to record the offset where we started this directory
3768 * and update last_directory to have quick access to it.
3770 if (info
->last_directory
== NULL
||
3771 !strncmp(new_directory_name
, info
->last_directory
,
3772 info
->last_directory_len
)) {
3773 uintptr_t offset
= info
->versions
.nr
;
3775 info
->last_directory
= new_directory_name
;
3776 info
->last_directory_len
= strlen(info
->last_directory
);
3778 * Record the offset into info->versions where we will
3779 * start recording basenames of paths found within
3780 * new_directory_name.
3782 string_list_append(&info
->offsets
,
3783 info
->last_directory
)->util
= (void*)offset
;
3788 * The next entry that will be processed will be within
3789 * new_directory_name. Since at this point we know that
3790 * new_directory_name is within a different directory than
3791 * info->last_directory, we have all entries for info->last_directory
3792 * in info->versions and we need to create a tree object for them.
3794 dir_info
= strmap_get(&opt
->priv
->paths
, info
->last_directory
);
3796 offset
= (uintptr_t)info
->offsets
.items
[info
->offsets
.nr
-1].util
;
3797 if (offset
== info
->versions
.nr
) {
3799 * Actually, we don't need to create a tree object in this
3800 * case. Whenever all files within a directory disappear
3801 * during the merge (e.g. unmodified on one side and
3802 * deleted on the other, or files were renamed elsewhere),
3803 * then we get here and the directory itself needs to be
3804 * omitted from its parent tree as well.
3806 dir_info
->is_null
= 1;
3809 * Write out the tree to the git object directory, and also
3810 * record the mode and oid in dir_info->result.
3812 dir_info
->is_null
= 0;
3813 dir_info
->result
.mode
= S_IFDIR
;
3814 if (write_tree(&dir_info
->result
.oid
, &info
->versions
, offset
,
3815 opt
->repo
->hash_algo
->rawsz
) < 0)
3820 * We've now used several entries from info->versions and one entry
3821 * from info->offsets, so we get rid of those values.
3824 info
->versions
.nr
= offset
;
3827 * Now we've taken care of the completed directory, but we need to
3828 * prepare things since future entries will be in
3829 * new_directory_name. (In particular, process_entry() will be
3830 * appending new entries to info->versions.) So, we need to make
3831 * sure new_directory_name is the last entry in info->offsets.
3833 prev_dir
= info
->offsets
.nr
== 0 ? NULL
:
3834 info
->offsets
.items
[info
->offsets
.nr
-1].string
;
3835 if (new_directory_name
!= prev_dir
) {
3836 uintptr_t c
= info
->versions
.nr
;
3837 string_list_append(&info
->offsets
,
3838 new_directory_name
)->util
= (void*)c
;
3841 /* And, of course, we need to update last_directory to match. */
3842 info
->last_directory
= new_directory_name
;
3843 info
->last_directory_len
= strlen(info
->last_directory
);
3848 /* Per entry merge function */
3849 static int process_entry(struct merge_options
*opt
,
3851 struct conflict_info
*ci
,
3852 struct directory_versions
*dir_metadata
)
3854 int df_file_index
= 0;
3857 assert(ci
->filemask
>= 0 && ci
->filemask
<= 7);
3858 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
3859 assert(ci
->match_mask
== 0 || ci
->match_mask
== 3 ||
3860 ci
->match_mask
== 5 || ci
->match_mask
== 6);
3863 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
3864 if (ci
->filemask
== 0)
3865 /* nothing else to handle */
3867 assert(ci
->df_conflict
);
3870 if (ci
->df_conflict
&& ci
->merged
.result
.mode
== 0) {
3874 * directory no longer in the way, but we do have a file we
3875 * need to place here so we need to clean away the "directory
3876 * merges to nothing" result.
3878 ci
->df_conflict
= 0;
3879 assert(ci
->filemask
!= 0);
3880 ci
->merged
.clean
= 0;
3881 ci
->merged
.is_null
= 0;
3882 /* and we want to zero out any directory-related entries */
3883 ci
->match_mask
= (ci
->match_mask
& ~ci
->dirmask
);
3885 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3886 if (ci
->filemask
& (1 << i
))
3888 ci
->stages
[i
].mode
= 0;
3889 oidcpy(&ci
->stages
[i
].oid
, null_oid());
3891 } else if (ci
->df_conflict
&& ci
->merged
.result
.mode
!= 0) {
3893 * This started out as a D/F conflict, and the entries in
3894 * the competing directory were not removed by the merge as
3895 * evidenced by write_completed_directory() writing a value
3896 * to ci->merged.result.mode.
3898 struct conflict_info
*new_ci
;
3900 const char *old_path
= path
;
3903 assert(ci
->merged
.result
.mode
== S_IFDIR
);
3906 * If filemask is 1, we can just ignore the file as having
3907 * been deleted on both sides. We do not want to overwrite
3908 * ci->merged.result, since it stores the tree for all the
3911 if (ci
->filemask
== 1) {
3917 * This file still exists on at least one side, and we want
3918 * the directory to remain here, so we need to move this
3919 * path to some new location.
3921 new_ci
= mem_pool_calloc(&opt
->priv
->pool
, 1, sizeof(*new_ci
));
3923 /* We don't really want new_ci->merged.result copied, but it'll
3924 * be overwritten below so it doesn't matter. We also don't
3925 * want any directory mode/oid values copied, but we'll zero
3926 * those out immediately. We do want the rest of ci copied.
3928 memcpy(new_ci
, ci
, sizeof(*ci
));
3929 new_ci
->match_mask
= (new_ci
->match_mask
& ~new_ci
->dirmask
);
3930 new_ci
->dirmask
= 0;
3931 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
3932 if (new_ci
->filemask
& (1 << i
))
3934 /* zero out any entries related to directories */
3935 new_ci
->stages
[i
].mode
= 0;
3936 oidcpy(&new_ci
->stages
[i
].oid
, null_oid());
3940 * Find out which side this file came from; note that we
3941 * cannot just use ci->filemask, because renames could cause
3942 * the filemask to go back to 7. So we use dirmask, then
3943 * pick the opposite side's index.
3945 df_file_index
= (ci
->dirmask
& (1 << 1)) ? 2 : 1;
3946 branch
= (df_file_index
== 1) ? opt
->branch1
: opt
->branch2
;
3947 path
= unique_path(opt
, path
, branch
);
3948 strmap_put(&opt
->priv
->paths
, path
, new_ci
);
3950 path_msg(opt
, CONFLICT_FILE_DIRECTORY
, 0,
3951 path
, old_path
, NULL
, NULL
,
3952 _("CONFLICT (file/directory): directory in the way "
3953 "of %s from %s; moving it to %s instead."),
3954 old_path
, branch
, path
);
3957 * Zero out the filemask for the old ci. At this point, ci
3958 * was just an entry for a directory, so we don't need to
3959 * do anything more with it.
3964 * Now note that we're working on the new entry (path was
3971 * NOTE: Below there is a long switch-like if-elseif-elseif... block
3972 * which the code goes through even for the df_conflict cases
3975 if (ci
->match_mask
) {
3976 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
3977 if (ci
->match_mask
== 6) {
3978 /* stages[1] == stages[2] */
3979 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
3980 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
3982 /* determine the mask of the side that didn't match */
3983 unsigned int othermask
= 7 & ~ci
->match_mask
;
3984 int side
= (othermask
== 4) ? 2 : 1;
3986 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
3987 ci
->merged
.is_null
= !ci
->merged
.result
.mode
;
3988 if (ci
->merged
.is_null
)
3989 ci
->merged
.clean
= 1;
3990 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
3992 assert(othermask
== 2 || othermask
== 4);
3993 assert(ci
->merged
.is_null
==
3994 (ci
->filemask
== ci
->match_mask
));
3996 } else if (ci
->filemask
>= 6 &&
3997 (S_IFMT
& ci
->stages
[1].mode
) !=
3998 (S_IFMT
& ci
->stages
[2].mode
)) {
3999 /* Two different items from (file/submodule/symlink) */
4000 if (opt
->priv
->call_depth
) {
4001 /* Just use the version from the merge base */
4002 ci
->merged
.clean
= 0;
4003 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[0].oid
);
4004 ci
->merged
.result
.mode
= ci
->stages
[0].mode
;
4005 ci
->merged
.is_null
= (ci
->merged
.result
.mode
== 0);
4007 /* Handle by renaming one or both to separate paths. */
4008 unsigned o_mode
= ci
->stages
[0].mode
;
4009 unsigned a_mode
= ci
->stages
[1].mode
;
4010 unsigned b_mode
= ci
->stages
[2].mode
;
4011 struct conflict_info
*new_ci
;
4012 const char *a_path
= NULL
, *b_path
= NULL
;
4013 int rename_a
= 0, rename_b
= 0;
4015 new_ci
= mem_pool_alloc(&opt
->priv
->pool
,
4018 if (S_ISREG(a_mode
))
4020 else if (S_ISREG(b_mode
))
4028 a_path
= unique_path(opt
, path
, opt
->branch1
);
4030 b_path
= unique_path(opt
, path
, opt
->branch2
);
4032 if (rename_a
&& rename_b
) {
4033 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4034 path
, a_path
, b_path
, NULL
,
4035 _("CONFLICT (distinct types): %s had "
4036 "different types on each side; "
4037 "renamed both of them so each can "
4038 "be recorded somewhere."),
4041 path_msg(opt
, CONFLICT_DISTINCT_MODES
, 0,
4042 path
, rename_a
? a_path
: b_path
,
4044 _("CONFLICT (distinct types): %s had "
4045 "different types on each side; "
4046 "renamed one of them so each can be "
4047 "recorded somewhere."),
4051 ci
->merged
.clean
= 0;
4052 memcpy(new_ci
, ci
, sizeof(*new_ci
));
4054 /* Put b into new_ci, removing a from stages */
4055 new_ci
->merged
.result
.mode
= ci
->stages
[2].mode
;
4056 oidcpy(&new_ci
->merged
.result
.oid
, &ci
->stages
[2].oid
);
4057 new_ci
->stages
[1].mode
= 0;
4058 oidcpy(&new_ci
->stages
[1].oid
, null_oid());
4059 new_ci
->filemask
= 5;
4060 if ((S_IFMT
& b_mode
) != (S_IFMT
& o_mode
)) {
4061 new_ci
->stages
[0].mode
= 0;
4062 oidcpy(&new_ci
->stages
[0].oid
, null_oid());
4063 new_ci
->filemask
= 4;
4066 /* Leave only a in ci, fixing stages. */
4067 ci
->merged
.result
.mode
= ci
->stages
[1].mode
;
4068 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[1].oid
);
4069 ci
->stages
[2].mode
= 0;
4070 oidcpy(&ci
->stages
[2].oid
, null_oid());
4072 if ((S_IFMT
& a_mode
) != (S_IFMT
& o_mode
)) {
4073 ci
->stages
[0].mode
= 0;
4074 oidcpy(&ci
->stages
[0].oid
, null_oid());
4078 /* Insert entries into opt->priv_paths */
4079 assert(rename_a
|| rename_b
);
4081 strmap_put(&opt
->priv
->paths
, a_path
, ci
);
4085 strmap_put(&opt
->priv
->paths
, b_path
, new_ci
);
4087 if (rename_a
&& rename_b
)
4088 strmap_remove(&opt
->priv
->paths
, path
, 0);
4091 * Do special handling for b_path since process_entry()
4092 * won't be called on it specially.
4094 strmap_put(&opt
->priv
->conflicted
, b_path
, new_ci
);
4095 record_entry_for_tree(dir_metadata
, b_path
,
4099 * Remaining code for processing this entry should
4100 * think in terms of processing a_path.
4105 } else if (ci
->filemask
>= 6) {
4106 /* Need a two-way or three-way content merge */
4107 struct version_info merged_file
;
4109 struct version_info
*o
= &ci
->stages
[0];
4110 struct version_info
*a
= &ci
->stages
[1];
4111 struct version_info
*b
= &ci
->stages
[2];
4113 clean_merge
= handle_content_merge(opt
, path
, o
, a
, b
,
4115 opt
->priv
->call_depth
* 2,
4117 if (clean_merge
< 0)
4119 ci
->merged
.clean
= clean_merge
&&
4120 !ci
->df_conflict
&& !ci
->path_conflict
;
4121 ci
->merged
.result
.mode
= merged_file
.mode
;
4122 ci
->merged
.is_null
= (merged_file
.mode
== 0);
4123 oidcpy(&ci
->merged
.result
.oid
, &merged_file
.oid
);
4124 if (clean_merge
&& ci
->df_conflict
) {
4125 assert(df_file_index
== 1 || df_file_index
== 2);
4126 ci
->filemask
= 1 << df_file_index
;
4127 ci
->stages
[df_file_index
].mode
= merged_file
.mode
;
4128 oidcpy(&ci
->stages
[df_file_index
].oid
, &merged_file
.oid
);
4131 const char *reason
= _("content");
4132 if (ci
->filemask
== 6)
4133 reason
= _("add/add");
4134 if (S_ISGITLINK(merged_file
.mode
))
4135 reason
= _("submodule");
4136 path_msg(opt
, CONFLICT_CONTENTS
, 0,
4137 path
, NULL
, NULL
, NULL
,
4138 _("CONFLICT (%s): Merge conflict in %s"),
4141 } else if (ci
->filemask
== 3 || ci
->filemask
== 5) {
4143 const char *modify_branch
, *delete_branch
;
4144 int side
= (ci
->filemask
== 5) ? 2 : 1;
4145 int index
= opt
->priv
->call_depth
? 0 : side
;
4147 ci
->merged
.result
.mode
= ci
->stages
[index
].mode
;
4148 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[index
].oid
);
4149 ci
->merged
.clean
= 0;
4151 modify_branch
= (side
== 1) ? opt
->branch1
: opt
->branch2
;
4152 delete_branch
= (side
== 1) ? opt
->branch2
: opt
->branch1
;
4154 if (opt
->renormalize
&&
4155 blob_unchanged(opt
, &ci
->stages
[0], &ci
->stages
[side
],
4157 if (!ci
->path_conflict
) {
4159 * Blob unchanged after renormalization, so
4160 * there's no modify/delete conflict after all;
4161 * we can just remove the file.
4163 ci
->merged
.is_null
= 1;
4164 ci
->merged
.clean
= 1;
4166 * file goes away => even if there was a
4167 * directory/file conflict there isn't one now.
4169 ci
->df_conflict
= 0;
4171 /* rename/delete, so conflict remains */
4173 } else if (ci
->path_conflict
&&
4174 oideq(&ci
->stages
[0].oid
, &ci
->stages
[side
].oid
)) {
4176 * This came from a rename/delete; no action to take,
4177 * but avoid printing "modify/delete" conflict notice
4178 * since the contents were not modified.
4181 path_msg(opt
, CONFLICT_MODIFY_DELETE
, 0,
4182 path
, NULL
, NULL
, NULL
,
4183 _("CONFLICT (modify/delete): %s deleted in %s "
4184 "and modified in %s. Version %s of %s left "
4186 path
, delete_branch
, modify_branch
,
4187 modify_branch
, path
);
4189 } else if (ci
->filemask
== 2 || ci
->filemask
== 4) {
4190 /* Added on one side */
4191 int side
= (ci
->filemask
== 4) ? 2 : 1;
4192 ci
->merged
.result
.mode
= ci
->stages
[side
].mode
;
4193 oidcpy(&ci
->merged
.result
.oid
, &ci
->stages
[side
].oid
);
4194 ci
->merged
.clean
= !ci
->df_conflict
&& !ci
->path_conflict
;
4195 } else if (ci
->filemask
== 1) {
4196 /* Deleted on both sides */
4197 ci
->merged
.is_null
= 1;
4198 ci
->merged
.result
.mode
= 0;
4199 oidcpy(&ci
->merged
.result
.oid
, null_oid());
4200 assert(!ci
->df_conflict
);
4201 ci
->merged
.clean
= !ci
->path_conflict
;
4205 * If still conflicted, record it separately. This allows us to later
4206 * iterate over just conflicted entries when updating the index instead
4207 * of iterating over all entries.
4209 if (!ci
->merged
.clean
)
4210 strmap_put(&opt
->priv
->conflicted
, path
, ci
);
4212 /* Record metadata for ci->merged in dir_metadata */
4213 record_entry_for_tree(dir_metadata
, path
, &ci
->merged
);
4217 static void prefetch_for_content_merges(struct merge_options
*opt
,
4218 struct string_list
*plist
)
4220 struct string_list_item
*e
;
4221 struct oid_array to_fetch
= OID_ARRAY_INIT
;
4223 if (opt
->repo
!= the_repository
|| !repo_has_promisor_remote(the_repository
))
4226 for (e
= &plist
->items
[plist
->nr
-1]; e
>= plist
->items
; --e
) {
4227 /* char *path = e->string; */
4228 struct conflict_info
*ci
= e
->util
;
4231 /* Ignore clean entries */
4232 if (ci
->merged
.clean
)
4235 /* Ignore entries that don't need a content merge */
4236 if (ci
->match_mask
|| ci
->filemask
< 6 ||
4237 !S_ISREG(ci
->stages
[1].mode
) ||
4238 !S_ISREG(ci
->stages
[2].mode
) ||
4239 oideq(&ci
->stages
[1].oid
, &ci
->stages
[2].oid
))
4242 /* Also don't need content merge if base matches either side */
4243 if (ci
->filemask
== 7 &&
4244 S_ISREG(ci
->stages
[0].mode
) &&
4245 (oideq(&ci
->stages
[0].oid
, &ci
->stages
[1].oid
) ||
4246 oideq(&ci
->stages
[0].oid
, &ci
->stages
[2].oid
)))
4249 for (i
= 0; i
< 3; i
++) {
4250 unsigned side_mask
= (1 << i
);
4251 struct version_info
*vi
= &ci
->stages
[i
];
4253 if ((ci
->filemask
& side_mask
) &&
4254 S_ISREG(vi
->mode
) &&
4255 oid_object_info_extended(opt
->repo
, &vi
->oid
, NULL
,
4256 OBJECT_INFO_FOR_PREFETCH
))
4257 oid_array_append(&to_fetch
, &vi
->oid
);
4261 promisor_remote_get_direct(opt
->repo
, to_fetch
.oid
, to_fetch
.nr
);
4262 oid_array_clear(&to_fetch
);
4265 static int process_entries(struct merge_options
*opt
,
4266 struct object_id
*result_oid
)
4268 struct hashmap_iter iter
;
4269 struct strmap_entry
*e
;
4270 struct string_list plist
= STRING_LIST_INIT_NODUP
;
4271 struct string_list_item
*entry
;
4272 struct directory_versions dir_metadata
= { STRING_LIST_INIT_NODUP
,
4273 STRING_LIST_INIT_NODUP
,
4277 trace2_region_enter("merge", "process_entries setup", opt
->repo
);
4278 if (strmap_empty(&opt
->priv
->paths
)) {
4279 oidcpy(result_oid
, opt
->repo
->hash_algo
->empty_tree
);
4283 /* Hack to pre-allocate plist to the desired size */
4284 trace2_region_enter("merge", "plist grow", opt
->repo
);
4285 ALLOC_GROW(plist
.items
, strmap_get_size(&opt
->priv
->paths
), plist
.alloc
);
4286 trace2_region_leave("merge", "plist grow", opt
->repo
);
4288 /* Put every entry from paths into plist, then sort */
4289 trace2_region_enter("merge", "plist copy", opt
->repo
);
4290 strmap_for_each_entry(&opt
->priv
->paths
, &iter
, e
) {
4291 string_list_append(&plist
, e
->key
)->util
= e
->value
;
4293 trace2_region_leave("merge", "plist copy", opt
->repo
);
4295 trace2_region_enter("merge", "plist special sort", opt
->repo
);
4296 plist
.cmp
= sort_dirs_next_to_their_children
;
4297 string_list_sort(&plist
);
4298 trace2_region_leave("merge", "plist special sort", opt
->repo
);
4300 trace2_region_leave("merge", "process_entries setup", opt
->repo
);
4303 * Iterate over the items in reverse order, so we can handle paths
4304 * below a directory before needing to handle the directory itself.
4306 * This allows us to write subtrees before we need to write trees,
4307 * and it also enables sane handling of directory/file conflicts
4308 * (because it allows us to know whether the directory is still in
4309 * the way when it is time to process the file at the same path).
4311 trace2_region_enter("merge", "processing", opt
->repo
);
4312 prefetch_for_content_merges(opt
, &plist
);
4313 for (entry
= &plist
.items
[plist
.nr
-1]; entry
>= plist
.items
; --entry
) {
4314 char *path
= entry
->string
;
4316 * NOTE: mi may actually be a pointer to a conflict_info, but
4317 * we have to check mi->clean first to see if it's safe to
4318 * reassign to such a pointer type.
4320 struct merged_info
*mi
= entry
->util
;
4322 if (write_completed_directory(opt
, mi
->directory_name
,
4323 &dir_metadata
) < 0) {
4328 record_entry_for_tree(&dir_metadata
, path
, mi
);
4330 struct conflict_info
*ci
= (struct conflict_info
*)mi
;
4331 if (process_entry(opt
, path
, ci
, &dir_metadata
) < 0) {
4337 trace2_region_leave("merge", "processing", opt
->repo
);
4339 trace2_region_enter("merge", "process_entries cleanup", opt
->repo
);
4340 if (dir_metadata
.offsets
.nr
!= 1 ||
4341 (uintptr_t)dir_metadata
.offsets
.items
[0].util
!= 0) {
4342 printf("dir_metadata.offsets.nr = %"PRIuMAX
" (should be 1)\n",
4343 (uintmax_t)dir_metadata
.offsets
.nr
);
4344 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
4345 (unsigned)(uintptr_t)dir_metadata
.offsets
.items
[0].util
);
4347 BUG("dir_metadata accounting completely off; shouldn't happen");
4349 if (write_tree(result_oid
, &dir_metadata
.versions
, 0,
4350 opt
->repo
->hash_algo
->rawsz
) < 0)
4353 string_list_clear(&plist
, 0);
4354 string_list_clear(&dir_metadata
.versions
, 0);
4355 string_list_clear(&dir_metadata
.offsets
, 0);
4356 trace2_region_leave("merge", "process_entries cleanup", opt
->repo
);
4361 /*** Function Grouping: functions related to merge_switch_to_result() ***/
4363 static int checkout(struct merge_options
*opt
,
4367 /* Switch the index/working copy from old to new */
4369 struct tree_desc trees
[2];
4370 struct unpack_trees_options unpack_opts
;
4372 memset(&unpack_opts
, 0, sizeof(unpack_opts
));
4373 unpack_opts
.head_idx
= -1;
4374 unpack_opts
.src_index
= opt
->repo
->index
;
4375 unpack_opts
.dst_index
= opt
->repo
->index
;
4377 setup_unpack_trees_porcelain(&unpack_opts
, "merge");
4380 * NOTE: if this were just "git checkout" code, we would probably
4381 * read or refresh the cache and check for a conflicted index, but
4382 * builtin/merge.c or sequencer.c really needs to read the index
4383 * and check for conflicted entries before starting merging for a
4384 * good user experience (no sense waiting for merges/rebases before
4385 * erroring out), so there's no reason to duplicate that work here.
4388 /* 2-way merge to the new branch */
4389 unpack_opts
.update
= 1;
4390 unpack_opts
.merge
= 1;
4391 unpack_opts
.quiet
= 0; /* FIXME: sequencer might want quiet? */
4392 unpack_opts
.verbose_update
= (opt
->verbosity
> 2);
4393 unpack_opts
.fn
= twoway_merge
;
4394 unpack_opts
.preserve_ignored
= 0; /* FIXME: !opts->overwrite_ignore */
4396 init_tree_desc(&trees
[0], prev
->buffer
, prev
->size
);
4398 init_tree_desc(&trees
[1], next
->buffer
, next
->size
);
4400 ret
= unpack_trees(2, trees
, &unpack_opts
);
4401 clear_unpack_trees_porcelain(&unpack_opts
);
4405 static int record_conflicted_index_entries(struct merge_options
*opt
)
4407 struct hashmap_iter iter
;
4408 struct strmap_entry
*e
;
4409 struct index_state
*index
= opt
->repo
->index
;
4410 struct checkout state
= CHECKOUT_INIT
;
4412 int original_cache_nr
;
4414 if (strmap_empty(&opt
->priv
->conflicted
))
4418 * We are in a conflicted state. These conflicts might be inside
4419 * sparse-directory entries, so check if any entries are outside
4420 * of the sparse-checkout cone preemptively.
4422 * We set original_cache_nr below, but that might change if
4423 * index_name_pos() calls ask for paths within sparse directories.
4425 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4426 if (!path_in_sparse_checkout(e
->key
, index
)) {
4427 ensure_full_index(index
);
4432 /* If any entries have skip_worktree set, we'll have to check 'em out */
4435 state
.refresh_cache
= 1;
4436 state
.istate
= index
;
4437 original_cache_nr
= index
->cache_nr
;
4439 /* Append every entry from conflicted into index, then sort */
4440 strmap_for_each_entry(&opt
->priv
->conflicted
, &iter
, e
) {
4441 const char *path
= e
->key
;
4442 struct conflict_info
*ci
= e
->value
;
4444 struct cache_entry
*ce
;
4450 * The index will already have a stage=0 entry for this path,
4451 * because we created an as-merged-as-possible version of the
4452 * file and checkout() moved the working copy and index over
4455 * However, previous iterations through this loop will have
4456 * added unstaged entries to the end of the cache which
4457 * ignore the standard alphabetical ordering of cache
4458 * entries and break invariants needed for index_name_pos()
4459 * to work. However, we know the entry we want is before
4460 * those appended cache entries, so do a temporary swap on
4461 * cache_nr to only look through entries of interest.
4463 SWAP(index
->cache_nr
, original_cache_nr
);
4464 pos
= index_name_pos(index
, path
, strlen(path
));
4465 SWAP(index
->cache_nr
, original_cache_nr
);
4467 if (ci
->filemask
!= 1)
4468 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path
);
4469 cache_tree_invalidate_path(index
, path
);
4471 ce
= index
->cache
[pos
];
4474 * Clean paths with CE_SKIP_WORKTREE set will not be
4475 * written to the working tree by the unpack_trees()
4476 * call in checkout(). Our conflicted entries would
4477 * have appeared clean to that code since we ignored
4478 * the higher order stages. Thus, we need override
4479 * the CE_SKIP_WORKTREE bit and manually write those
4480 * files to the working disk here.
4482 if (ce_skip_worktree(ce
))
4483 errs
|= checkout_entry(ce
, &state
, NULL
, NULL
);
4486 * Mark this cache entry for removal and instead add
4487 * new stage>0 entries corresponding to the
4488 * conflicts. If there are many conflicted entries, we
4489 * want to avoid memmove'ing O(NM) entries by
4490 * inserting the new entries one at a time. So,
4491 * instead, we just add the new cache entries to the
4492 * end (ignoring normal index requirements on sort
4493 * order) and sort the index once we're all done.
4495 ce
->ce_flags
|= CE_REMOVE
;
4498 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4499 struct version_info
*vi
;
4500 if (!(ci
->filemask
& (1ul << i
)))
4502 vi
= &ci
->stages
[i
];
4503 ce
= make_cache_entry(index
, vi
->mode
, &vi
->oid
,
4505 add_index_entry(index
, ce
, ADD_CACHE_JUST_APPEND
);
4510 * Remove the unused cache entries (and invalidate the relevant
4511 * cache-trees), then sort the index entries to get the conflicted
4512 * entries we added to the end into their right locations.
4514 remove_marked_cache_entries(index
, 1);
4516 * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily
4517 * on filename and secondarily on stage, and (name, stage #) are a
4520 QSORT(index
->cache
, index
->cache_nr
, cmp_cache_name_compare
);
4525 static void print_submodule_conflict_suggestion(struct string_list
*csub
) {
4526 struct string_list_item
*item
;
4527 struct strbuf msg
= STRBUF_INIT
;
4528 struct strbuf tmp
= STRBUF_INIT
;
4529 struct strbuf subs
= STRBUF_INIT
;
4534 strbuf_add_separated_string_list(&subs
, " ", csub
);
4535 for_each_string_list_item(item
, csub
) {
4536 struct conflicted_submodule_item
*util
= item
->util
;
4539 * NEEDSWORK: The steps to resolve these errors deserve a more
4540 * detailed explanation than what is currently printed below.
4542 if (util
->flag
== CONFLICT_SUBMODULE_NOT_INITIALIZED
||
4543 util
->flag
== CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE
)
4547 * TRANSLATORS: This is a line of advice to resolve a merge
4548 * conflict in a submodule. The first argument is the submodule
4549 * name, and the second argument is the abbreviated id of the
4550 * commit that needs to be merged. For example:
4551 * - go to submodule (mysubmodule), and either merge commit abc1234"
4553 strbuf_addf(&tmp
, _(" - go to submodule (%s), and either merge commit %s\n"
4554 " or update to an existing commit which has merged those changes\n"),
4555 item
->string
, util
->abbrev
);
4559 * TRANSLATORS: This is a detailed message for resolving submodule
4560 * conflicts. The first argument is string containing one step per
4561 * submodule. The second is a space-separated list of submodule names.
4564 _("Recursive merging with submodules currently only supports trivial cases.\n"
4565 "Please manually handle the merging of each conflicted submodule.\n"
4566 "This can be accomplished with the following steps:\n"
4568 " - come back to superproject and run:\n\n"
4570 " to record the above merge or update\n"
4571 " - resolve any other conflicts in the superproject\n"
4572 " - commit the resulting index in the superproject\n"),
4575 printf("%s", msg
.buf
);
4577 strbuf_release(&subs
);
4578 strbuf_release(&tmp
);
4579 strbuf_release(&msg
);
4582 void merge_display_update_messages(struct merge_options
*opt
,
4584 struct merge_result
*result
)
4586 struct merge_options_internal
*opti
= result
->priv
;
4587 struct hashmap_iter iter
;
4588 struct strmap_entry
*e
;
4589 struct string_list olist
= STRING_LIST_INIT_NODUP
;
4591 if (opt
->record_conflict_msgs_as_headers
)
4592 BUG("Either display conflict messages or record them as headers, not both");
4594 trace2_region_enter("merge", "display messages", opt
->repo
);
4596 /* Hack to pre-allocate olist to the desired size */
4597 ALLOC_GROW(olist
.items
, strmap_get_size(&opti
->conflicts
),
4600 /* Put every entry from output into olist, then sort */
4601 strmap_for_each_entry(&opti
->conflicts
, &iter
, e
) {
4602 string_list_append(&olist
, e
->key
)->util
= e
->value
;
4604 string_list_sort(&olist
);
4606 /* Iterate over the items, printing them */
4607 for (int path_nr
= 0; path_nr
< olist
.nr
; ++path_nr
) {
4608 struct string_list
*conflicts
= olist
.items
[path_nr
].util
;
4609 for (int i
= 0; i
< conflicts
->nr
; i
++) {
4610 struct logical_conflict_info
*info
=
4611 conflicts
->items
[i
].util
;
4614 printf("%lu", (unsigned long)info
->paths
.nr
);
4616 for (int n
= 0; n
< info
->paths
.nr
; n
++) {
4617 fputs(info
->paths
.v
[n
], stdout
);
4620 fputs(type_short_descriptions
[info
->type
],
4624 puts(conflicts
->items
[i
].string
);
4629 string_list_clear(&olist
, 0);
4631 print_submodule_conflict_suggestion(&opti
->conflicted_submodules
);
4633 /* Also include needed rename limit adjustment now */
4634 diff_warn_rename_limit("merge.renamelimit",
4635 opti
->renames
.needed_limit
, 0);
4637 trace2_region_leave("merge", "display messages", opt
->repo
);
4640 void merge_get_conflicted_files(struct merge_result
*result
,
4641 struct string_list
*conflicted_files
)
4643 struct hashmap_iter iter
;
4644 struct strmap_entry
*e
;
4645 struct merge_options_internal
*opti
= result
->priv
;
4647 strmap_for_each_entry(&opti
->conflicted
, &iter
, e
) {
4648 const char *path
= e
->key
;
4649 struct conflict_info
*ci
= e
->value
;
4654 for (i
= MERGE_BASE
; i
<= MERGE_SIDE2
; i
++) {
4655 struct stage_info
*si
;
4657 if (!(ci
->filemask
& (1ul << i
)))
4660 si
= xmalloc(sizeof(*si
));
4662 si
->mode
= ci
->stages
[i
].mode
;
4663 oidcpy(&si
->oid
, &ci
->stages
[i
].oid
);
4664 string_list_append(conflicted_files
, path
)->util
= si
;
4667 /* string_list_sort() uses a stable sort, so we're good */
4668 string_list_sort(conflicted_files
);
4671 void merge_switch_to_result(struct merge_options
*opt
,
4673 struct merge_result
*result
,
4674 int update_worktree_and_index
,
4675 int display_update_msgs
)
4677 assert(opt
->priv
== NULL
);
4678 if (result
->clean
>= 0 && update_worktree_and_index
) {
4679 const char *filename
;
4682 trace2_region_enter("merge", "checkout", opt
->repo
);
4683 if (checkout(opt
, head
, result
->tree
)) {
4684 /* failure to function */
4686 merge_finalize(opt
, result
);
4687 trace2_region_leave("merge", "checkout", opt
->repo
);
4690 trace2_region_leave("merge", "checkout", opt
->repo
);
4692 trace2_region_enter("merge", "record_conflicted", opt
->repo
);
4693 opt
->priv
= result
->priv
;
4694 if (record_conflicted_index_entries(opt
)) {
4695 /* failure to function */
4698 merge_finalize(opt
, result
);
4699 trace2_region_leave("merge", "record_conflicted",
4704 trace2_region_leave("merge", "record_conflicted", opt
->repo
);
4706 trace2_region_enter("merge", "write_auto_merge", opt
->repo
);
4707 filename
= git_path_auto_merge(opt
->repo
);
4708 fp
= xfopen(filename
, "w");
4709 fprintf(fp
, "%s\n", oid_to_hex(&result
->tree
->object
.oid
));
4711 trace2_region_leave("merge", "write_auto_merge", opt
->repo
);
4713 if (display_update_msgs
)
4714 merge_display_update_messages(opt
, /* detailed */ 0, result
);
4716 merge_finalize(opt
, result
);
4719 void merge_finalize(struct merge_options
*opt
,
4720 struct merge_result
*result
)
4722 struct merge_options_internal
*opti
= result
->priv
;
4724 if (opt
->renormalize
)
4725 git_attr_set_direction(GIT_ATTR_CHECKIN
);
4726 assert(opt
->priv
== NULL
);
4728 clear_or_reinit_internal_opts(opti
, 0);
4729 FREE_AND_NULL(opti
);
4732 /*** Function Grouping: helper functions for merge_incore_*() ***/
4734 static struct tree
*shift_tree_object(struct repository
*repo
,
4735 struct tree
*one
, struct tree
*two
,
4736 const char *subtree_shift
)
4738 struct object_id shifted
;
4740 if (!*subtree_shift
) {
4741 shift_tree(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
4743 shift_tree_by(repo
, &one
->object
.oid
, &two
->object
.oid
, &shifted
,
4746 if (oideq(&two
->object
.oid
, &shifted
))
4748 return lookup_tree(repo
, &shifted
);
4751 static inline void set_commit_tree(struct commit
*c
, struct tree
*t
)
4756 static struct commit
*make_virtual_commit(struct repository
*repo
,
4758 const char *comment
)
4760 struct commit
*commit
= alloc_commit_node(repo
);
4762 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
4763 set_commit_tree(commit
, tree
);
4764 commit
->object
.parsed
= 1;
4768 static void merge_start(struct merge_options
*opt
, struct merge_result
*result
)
4770 struct rename_info
*renames
;
4772 struct mem_pool
*pool
= NULL
;
4774 /* Sanity checks on opt */
4775 trace2_region_enter("merge", "sanity checks", opt
->repo
);
4778 assert(opt
->branch1
&& opt
->branch2
);
4780 assert(opt
->detect_directory_renames
>= MERGE_DIRECTORY_RENAMES_NONE
&&
4781 opt
->detect_directory_renames
<= MERGE_DIRECTORY_RENAMES_TRUE
);
4782 assert(opt
->rename_limit
>= -1);
4783 assert(opt
->rename_score
>= 0 && opt
->rename_score
<= MAX_SCORE
);
4784 assert(opt
->show_rename_progress
>= 0 && opt
->show_rename_progress
<= 1);
4786 assert(opt
->xdl_opts
>= 0);
4787 assert(opt
->recursive_variant
>= MERGE_VARIANT_NORMAL
&&
4788 opt
->recursive_variant
<= MERGE_VARIANT_THEIRS
);
4790 if (opt
->msg_header_prefix
)
4791 assert(opt
->record_conflict_msgs_as_headers
);
4794 * detect_renames, verbosity, buffer_output, and obuf are ignored
4795 * fields that were used by "recursive" rather than "ort" -- but
4796 * sanity check them anyway.
4798 assert(opt
->detect_renames
>= -1 &&
4799 opt
->detect_renames
<= DIFF_DETECT_COPY
);
4800 assert(opt
->verbosity
>= 0 && opt
->verbosity
<= 5);
4801 assert(opt
->buffer_output
<= 2);
4802 assert(opt
->obuf
.len
== 0);
4804 assert(opt
->priv
== NULL
);
4805 if (result
->_properly_initialized
!= 0 &&
4806 result
->_properly_initialized
!= RESULT_INITIALIZED
)
4807 BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run");
4808 assert(!!result
->priv
== !!result
->_properly_initialized
);
4810 opt
->priv
= result
->priv
;
4811 result
->priv
= NULL
;
4813 * opt->priv non-NULL means we had results from a previous
4814 * run; do a few sanity checks that user didn't mess with
4815 * it in an obvious fashion.
4817 assert(opt
->priv
->call_depth
== 0);
4818 assert(!opt
->priv
->toplevel_dir
||
4819 0 == strlen(opt
->priv
->toplevel_dir
));
4821 trace2_region_leave("merge", "sanity checks", opt
->repo
);
4823 /* Default to histogram diff. Actually, just hardcode it...for now. */
4824 opt
->xdl_opts
= DIFF_WITH_ALG(opt
, HISTOGRAM_DIFF
);
4826 /* Handle attr direction stuff for renormalization */
4827 if (opt
->renormalize
)
4828 git_attr_set_direction(GIT_ATTR_CHECKOUT
);
4830 /* Initialization of opt->priv, our internal merge data */
4831 trace2_region_enter("merge", "allocate/init", opt
->repo
);
4833 clear_or_reinit_internal_opts(opt
->priv
, 1);
4834 string_list_init_nodup(&opt
->priv
->conflicted_submodules
);
4835 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4838 opt
->priv
= xcalloc(1, sizeof(*opt
->priv
));
4840 /* Initialization of various renames fields */
4841 renames
= &opt
->priv
->renames
;
4842 mem_pool_init(&opt
->priv
->pool
, 0);
4843 pool
= &opt
->priv
->pool
;
4844 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4845 strintmap_init_with_options(&renames
->dirs_removed
[i
],
4846 NOT_RELEVANT
, pool
, 0);
4847 strmap_init_with_options(&renames
->dir_rename_count
[i
],
4849 strmap_init_with_options(&renames
->dir_renames
[i
],
4852 * relevant_sources uses -1 for the default, because we need
4853 * to be able to distinguish not-in-strintmap from valid
4854 * relevant_source values from enum file_rename_relevance.
4855 * In particular, possibly_cache_new_pair() expects a negative
4856 * value for not-found entries.
4858 strintmap_init_with_options(&renames
->relevant_sources
[i
],
4859 -1 /* explicitly invalid */,
4861 strmap_init_with_options(&renames
->cached_pairs
[i
],
4863 strset_init_with_options(&renames
->cached_irrelevant
[i
],
4865 strset_init_with_options(&renames
->cached_target_names
[i
],
4868 for (i
= MERGE_SIDE1
; i
<= MERGE_SIDE2
; i
++) {
4869 strintmap_init_with_options(&renames
->deferred
[i
].possible_trivial_merges
,
4871 strset_init_with_options(&renames
->deferred
[i
].target_dirs
,
4873 renames
->deferred
[i
].trivial_merges_okay
= 1; /* 1 == maybe */
4877 * Although we initialize opt->priv->paths with strdup_strings=0,
4878 * that's just to avoid making yet another copy of an allocated
4879 * string. Putting the entry into paths means we are taking
4880 * ownership, so we will later free it.
4882 * In contrast, conflicted just has a subset of keys from paths, so
4883 * we don't want to free those (it'd be a duplicate free).
4885 strmap_init_with_options(&opt
->priv
->paths
, pool
, 0);
4886 strmap_init_with_options(&opt
->priv
->conflicted
, pool
, 0);
4889 * keys & string_lists in conflicts will sometimes need to outlive
4890 * "paths", so it will have a copy of relevant keys. It's probably
4891 * a small subset of the overall paths that have special output.
4893 strmap_init(&opt
->priv
->conflicts
);
4895 trace2_region_leave("merge", "allocate/init", opt
->repo
);
4898 static void merge_check_renames_reusable(struct merge_options
*opt
,
4899 struct merge_result
*result
,
4900 struct tree
*merge_base
,
4904 struct rename_info
*renames
;
4905 struct tree
**merge_trees
;
4906 struct merge_options_internal
*opti
= result
->priv
;
4911 renames
= &opti
->renames
;
4912 merge_trees
= renames
->merge_trees
;
4915 * Handle case where previous merge operation did not want cache to
4916 * take effect, e.g. because rename/rename(1to1) makes it invalid.
4918 if (!merge_trees
[0]) {
4919 assert(!merge_trees
[0] && !merge_trees
[1] && !merge_trees
[2]);
4920 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4925 * Handle other cases; note that merge_trees[0..2] will only
4926 * be NULL if opti is, or if all three were manually set to
4927 * NULL by e.g. rename/rename(1to1) handling.
4929 assert(merge_trees
[0] && merge_trees
[1] && merge_trees
[2]);
4931 /* Check if we meet a condition for re-using cached_pairs */
4932 if (oideq(&merge_base
->object
.oid
, &merge_trees
[2]->object
.oid
) &&
4933 oideq(&side1
->object
.oid
, &result
->tree
->object
.oid
))
4934 renames
->cached_pairs_valid_side
= MERGE_SIDE1
;
4935 else if (oideq(&merge_base
->object
.oid
, &merge_trees
[1]->object
.oid
) &&
4936 oideq(&side2
->object
.oid
, &result
->tree
->object
.oid
))
4937 renames
->cached_pairs_valid_side
= MERGE_SIDE2
;
4939 renames
->cached_pairs_valid_side
= 0; /* neither side valid */
4942 /*** Function Grouping: merge_incore_*() and their internal variants ***/
4945 * Originally from merge_trees_internal(); heavily adapted, though.
4947 static void merge_ort_nonrecursive_internal(struct merge_options
*opt
,
4948 struct tree
*merge_base
,
4951 struct merge_result
*result
)
4953 struct object_id working_tree_oid
;
4955 if (opt
->subtree_shift
) {
4956 side2
= shift_tree_object(opt
->repo
, side1
, side2
,
4957 opt
->subtree_shift
);
4958 merge_base
= shift_tree_object(opt
->repo
, side1
, merge_base
,
4959 opt
->subtree_shift
);
4963 trace2_region_enter("merge", "collect_merge_info", opt
->repo
);
4964 if (collect_merge_info(opt
, merge_base
, side1
, side2
) != 0) {
4966 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
4967 * base, and 2-3) the trees for the two trees we're merging.
4969 err(opt
, _("collecting merge info failed for trees %s, %s, %s"),
4970 oid_to_hex(&merge_base
->object
.oid
),
4971 oid_to_hex(&side1
->object
.oid
),
4972 oid_to_hex(&side2
->object
.oid
));
4976 trace2_region_leave("merge", "collect_merge_info", opt
->repo
);
4978 trace2_region_enter("merge", "renames", opt
->repo
);
4979 result
->clean
= detect_and_process_renames(opt
, merge_base
,
4981 trace2_region_leave("merge", "renames", opt
->repo
);
4982 if (opt
->priv
->renames
.redo_after_renames
== 2) {
4983 trace2_region_enter("merge", "reset_maps", opt
->repo
);
4984 clear_or_reinit_internal_opts(opt
->priv
, 1);
4985 trace2_region_leave("merge", "reset_maps", opt
->repo
);
4989 trace2_region_enter("merge", "process_entries", opt
->repo
);
4990 if (process_entries(opt
, &working_tree_oid
) < 0)
4992 trace2_region_leave("merge", "process_entries", opt
->repo
);
4994 /* Set return values */
4995 result
->path_messages
= &opt
->priv
->conflicts
;
4997 if (result
->clean
>= 0) {
4998 result
->tree
= parse_tree_indirect(&working_tree_oid
);
4999 /* existence of conflicted entries implies unclean */
5000 result
->clean
&= strmap_empty(&opt
->priv
->conflicted
);
5002 if (!opt
->priv
->call_depth
) {
5003 result
->priv
= opt
->priv
;
5004 result
->_properly_initialized
= RESULT_INITIALIZED
;
5010 * Originally from merge_recursive_internal(); somewhat adapted, though.
5012 static void merge_ort_internal(struct merge_options
*opt
,
5013 struct commit_list
*merge_bases
,
5016 struct merge_result
*result
)
5018 struct commit
*next
;
5019 struct commit
*merged_merge_bases
;
5020 const char *ancestor_name
;
5021 struct strbuf merge_base_abbrev
= STRBUF_INIT
;
5024 merge_bases
= repo_get_merge_bases(the_repository
, h1
, h2
);
5025 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
5026 merge_bases
= reverse_commit_list(merge_bases
);
5029 merged_merge_bases
= pop_commit(&merge_bases
);
5030 if (!merged_merge_bases
) {
5031 /* if there is no common ancestor, use an empty tree */
5034 tree
= lookup_tree(opt
->repo
, opt
->repo
->hash_algo
->empty_tree
);
5035 merged_merge_bases
= make_virtual_commit(opt
->repo
, tree
,
5037 ancestor_name
= "empty tree";
5038 } else if (merge_bases
) {
5039 ancestor_name
= "merged common ancestors";
5041 strbuf_add_unique_abbrev(&merge_base_abbrev
,
5042 &merged_merge_bases
->object
.oid
,
5044 ancestor_name
= merge_base_abbrev
.buf
;
5047 for (next
= pop_commit(&merge_bases
); next
;
5048 next
= pop_commit(&merge_bases
)) {
5049 const char *saved_b1
, *saved_b2
;
5050 struct commit
*prev
= merged_merge_bases
;
5052 opt
->priv
->call_depth
++;
5054 * When the merge fails, the result contains files
5055 * with conflict markers. The cleanness flag is
5056 * ignored (unless indicating an error), it was never
5057 * actually used, as result of merge_trees has always
5058 * overwritten it: the committed "conflicts" were
5061 saved_b1
= opt
->branch1
;
5062 saved_b2
= opt
->branch2
;
5063 opt
->branch1
= "Temporary merge branch 1";
5064 opt
->branch2
= "Temporary merge branch 2";
5065 merge_ort_internal(opt
, NULL
, prev
, next
, result
);
5066 if (result
->clean
< 0)
5068 opt
->branch1
= saved_b1
;
5069 opt
->branch2
= saved_b2
;
5070 opt
->priv
->call_depth
--;
5072 merged_merge_bases
= make_virtual_commit(opt
->repo
,
5075 commit_list_insert(prev
, &merged_merge_bases
->parents
);
5076 commit_list_insert(next
, &merged_merge_bases
->parents
->next
);
5078 clear_or_reinit_internal_opts(opt
->priv
, 1);
5081 opt
->ancestor
= ancestor_name
;
5082 merge_ort_nonrecursive_internal(opt
,
5083 repo_get_commit_tree(opt
->repo
,
5084 merged_merge_bases
),
5085 repo_get_commit_tree(opt
->repo
, h1
),
5086 repo_get_commit_tree(opt
->repo
, h2
),
5088 strbuf_release(&merge_base_abbrev
);
5089 opt
->ancestor
= NULL
; /* avoid accidental re-use of opt->ancestor */
5092 void merge_incore_nonrecursive(struct merge_options
*opt
,
5093 struct tree
*merge_base
,
5096 struct merge_result
*result
)
5098 trace2_region_enter("merge", "incore_nonrecursive", opt
->repo
);
5100 trace2_region_enter("merge", "merge_start", opt
->repo
);
5101 assert(opt
->ancestor
!= NULL
);
5102 merge_check_renames_reusable(opt
, result
, merge_base
, side1
, side2
);
5103 merge_start(opt
, result
);
5105 * Record the trees used in this merge, so if there's a next merge in
5106 * a cherry-pick or rebase sequence it might be able to take advantage
5107 * of the cached_pairs in that next merge.
5109 opt
->priv
->renames
.merge_trees
[0] = merge_base
;
5110 opt
->priv
->renames
.merge_trees
[1] = side1
;
5111 opt
->priv
->renames
.merge_trees
[2] = side2
;
5112 trace2_region_leave("merge", "merge_start", opt
->repo
);
5114 merge_ort_nonrecursive_internal(opt
, merge_base
, side1
, side2
, result
);
5115 trace2_region_leave("merge", "incore_nonrecursive", opt
->repo
);
5118 void merge_incore_recursive(struct merge_options
*opt
,
5119 struct commit_list
*merge_bases
,
5120 struct commit
*side1
,
5121 struct commit
*side2
,
5122 struct merge_result
*result
)
5124 trace2_region_enter("merge", "incore_recursive", opt
->repo
);
5126 /* We set the ancestor label based on the merge_bases */
5127 assert(opt
->ancestor
== NULL
);
5129 trace2_region_enter("merge", "merge_start", opt
->repo
);
5130 merge_start(opt
, result
);
5131 trace2_region_leave("merge", "merge_start", opt
->repo
);
5133 merge_ort_internal(opt
, merge_bases
, side1
, side2
, result
);
5134 trace2_region_leave("merge", "incore_recursive", opt
->repo
);